src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRFCut_VdwLJSh_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_ElecRFCut_VdwLJSh_GeomW3P1_VF_avx_256_double
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwLJSh_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     krf              = _mm256_set1_pd(fr->ic->k_rf);
 108     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 109     crf              = _mm256_set1_pd(fr->ic->c_rf);
 110     nvdwtype         = fr->ntype;
 111     vdwparam         = fr->nbfp;
 112     vdwtype          = mdatoms->typeA;
 113
 114     /* Setup water-specific parameters */
 115     inr              = nlist->iinr[0];
 116     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 117     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 118     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 119     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 120
 121     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 122     rcutoff_scalar   = fr->rcoulomb;
 123     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 124     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 125
 126     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 127     rvdw             = _mm256_set1_pd(fr->rvdw);
 128
 129     /* Avoid stupid compiler warnings */
 130     jnrA = jnrB = jnrC = jnrD = 0;
 131     j_coord_offsetA = 0;
 132     j_coord_offsetB = 0;
 133     j_coord_offsetC = 0;
 134     j_coord_offsetD = 0;
 135
 136     outeriter        = 0;
 137     inneriter        = 0;
 138
 139     for(iidx=0;iidx<4*DIM;iidx++)
 140     {
 141         scratch[iidx] = 0.0;
 142     }
 143
 144     /* Start outer loop over neighborlists */
 145     for(iidx=0; iidx<nri; iidx++)
 146     {
 147         /* Load shift vector for this list */
 148         i_shift_offset   = DIM*shiftidx[iidx];
 149
 150         /* Load limits for loop over neighbors */
 151         j_index_start    = jindex[iidx];
 152         j_index_end      = jindex[iidx+1];
 153
 154         /* Get outer coordinate index */
 155         inr              = iinr[iidx];
 156         i_coord_offset   = DIM*inr;
 157
 158         /* Load i particle coords and add shift vector */
 159         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 160                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 161
 162         fix0             = _mm256_setzero_pd();
 163         fiy0             = _mm256_setzero_pd();
 164         fiz0             = _mm256_setzero_pd();
 165         fix1             = _mm256_setzero_pd();
 166         fiy1             = _mm256_setzero_pd();
 167         fiz1             = _mm256_setzero_pd();
 168         fix2             = _mm256_setzero_pd();
 169         fiy2             = _mm256_setzero_pd();
 170         fiz2             = _mm256_setzero_pd();
 171
 172         /* Reset potential sums */
 173         velecsum         = _mm256_setzero_pd();
 174         vvdwsum          = _mm256_setzero_pd();
 175
 176         /* Start inner kernel loop */
 177         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 178         {
 179
 180             /* Get j neighbor index, and coordinate index */
 181             jnrA             = jjnr[jidx];
 182             jnrB             = jjnr[jidx+1];
 183             jnrC             = jjnr[jidx+2];
 184             jnrD             = jjnr[jidx+3];
 185             j_coord_offsetA  = DIM*jnrA;
 186             j_coord_offsetB  = DIM*jnrB;
 187             j_coord_offsetC  = DIM*jnrC;
 188             j_coord_offsetD  = DIM*jnrD;
 189
 190             /* load j atom coordinates */
 191             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 192                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 193                                                  &jx0,&jy0,&jz0);
 194
 195             /* Calculate displacement vector */
 196             dx00             = _mm256_sub_pd(ix0,jx0);
 197             dy00             = _mm256_sub_pd(iy0,jy0);
 198             dz00             = _mm256_sub_pd(iz0,jz0);
 199             dx10             = _mm256_sub_pd(ix1,jx0);
 200             dy10             = _mm256_sub_pd(iy1,jy0);
 201             dz10             = _mm256_sub_pd(iz1,jz0);
 202             dx20             = _mm256_sub_pd(ix2,jx0);
 203             dy20             = _mm256_sub_pd(iy2,jy0);
 204             dz20             = _mm256_sub_pd(iz2,jz0);
 205
 206             /* Calculate squared distance and things based on it */
 207             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 208             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 209             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 210
 211             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 212             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 213             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 214
 215             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 216             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 217             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 218
 219             /* Load parameters for j particles */
 220             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 221                                                                  charge+jnrC+0,charge+jnrD+0);
 222             vdwjidx0A        = 2*vdwtype[jnrA+0];
 223             vdwjidx0B        = 2*vdwtype[jnrB+0];
 224             vdwjidx0C        = 2*vdwtype[jnrC+0];
 225             vdwjidx0D        = 2*vdwtype[jnrD+0];
 226
 227             fjx0             = _mm256_setzero_pd();
 228             fjy0             = _mm256_setzero_pd();
 229             fjz0             = _mm256_setzero_pd();
 230
 231             /**************************
 232              * CALCULATE INTERACTIONS *
 233              **************************/
 234
 235             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 236             {
 237
 238             /* Compute parameters for interactions between i and j atoms */
 239             qq00             = _mm256_mul_pd(iq0,jq0);
 240             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 241                                             vdwioffsetptr0+vdwjidx0B,
 242                                             vdwioffsetptr0+vdwjidx0C,
 243                                             vdwioffsetptr0+vdwjidx0D,
 244                                             &c6_00,&c12_00);
 245
 246             /* REACTION-FIELD ELECTROSTATICS */
 247             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 248             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 249
 250             /* LENNARD-JONES DISPERSION/REPULSION */
 251
 252             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 253             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 254             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 255             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 256                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 257             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 258
 259             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 260
 261             /* Update potential sum for this i atom from the interaction with this j atom. */
 262             velec            = _mm256_and_pd(velec,cutoff_mask);
 263             velecsum         = _mm256_add_pd(velecsum,velec);
 264             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 265             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 266
 267             fscal            = _mm256_add_pd(felec,fvdw);
 268
 269             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 270
 271             /* Calculate temporary vectorial force */
 272             tx               = _mm256_mul_pd(fscal,dx00);
 273             ty               = _mm256_mul_pd(fscal,dy00);
 274             tz               = _mm256_mul_pd(fscal,dz00);
 275
 276             /* Update vectorial force */
 277             fix0             = _mm256_add_pd(fix0,tx);
 278             fiy0             = _mm256_add_pd(fiy0,ty);
 279             fiz0             = _mm256_add_pd(fiz0,tz);
 280
 281             fjx0             = _mm256_add_pd(fjx0,tx);
 282             fjy0             = _mm256_add_pd(fjy0,ty);
 283             fjz0             = _mm256_add_pd(fjz0,tz);
 284
 285             }
 286
 287             /**************************
 288              * CALCULATE INTERACTIONS *
 289              **************************/
 290
 291             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 292             {
 293
 294             /* Compute parameters for interactions between i and j atoms */
 295             qq10             = _mm256_mul_pd(iq1,jq0);
 296
 297             /* REACTION-FIELD ELECTROSTATICS */
 298             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 299             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 300
 301             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 302
 303             /* Update potential sum for this i atom from the interaction with this j atom. */
 304             velec            = _mm256_and_pd(velec,cutoff_mask);
 305             velecsum         = _mm256_add_pd(velecsum,velec);
 306
 307             fscal            = felec;
 308
 309             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 310
 311             /* Calculate temporary vectorial force */
 312             tx               = _mm256_mul_pd(fscal,dx10);
 313             ty               = _mm256_mul_pd(fscal,dy10);
 314             tz               = _mm256_mul_pd(fscal,dz10);
 315
 316             /* Update vectorial force */
 317             fix1             = _mm256_add_pd(fix1,tx);
 318             fiy1             = _mm256_add_pd(fiy1,ty);
 319             fiz1             = _mm256_add_pd(fiz1,tz);
 320
 321             fjx0             = _mm256_add_pd(fjx0,tx);
 322             fjy0             = _mm256_add_pd(fjy0,ty);
 323             fjz0             = _mm256_add_pd(fjz0,tz);
 324
 325             }
 326
 327             /**************************
 328              * CALCULATE INTERACTIONS *
 329              **************************/
 330
 331             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 332             {
 333
 334             /* Compute parameters for interactions between i and j atoms */
 335             qq20             = _mm256_mul_pd(iq2,jq0);
 336
 337             /* REACTION-FIELD ELECTROSTATICS */
 338             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 339             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 340
 341             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 342
 343             /* Update potential sum for this i atom from the interaction with this j atom. */
 344             velec            = _mm256_and_pd(velec,cutoff_mask);
 345             velecsum         = _mm256_add_pd(velecsum,velec);
 346
 347             fscal            = felec;
 348
 349             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 350
 351             /* Calculate temporary vectorial force */
 352             tx               = _mm256_mul_pd(fscal,dx20);
 353             ty               = _mm256_mul_pd(fscal,dy20);
 354             tz               = _mm256_mul_pd(fscal,dz20);
 355
 356             /* Update vectorial force */
 357             fix2             = _mm256_add_pd(fix2,tx);
 358             fiy2             = _mm256_add_pd(fiy2,ty);
 359             fiz2             = _mm256_add_pd(fiz2,tz);
 360
 361             fjx0             = _mm256_add_pd(fjx0,tx);
 362             fjy0             = _mm256_add_pd(fjy0,ty);
 363             fjz0             = _mm256_add_pd(fjz0,tz);
 364
 365             }
 366
 367             fjptrA             = f+j_coord_offsetA;
 368             fjptrB             = f+j_coord_offsetB;
 369             fjptrC             = f+j_coord_offsetC;
 370             fjptrD             = f+j_coord_offsetD;
 371
 372             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 373
 374             /* Inner loop uses 129 flops */
 375         }
 376
 377         if(jidx<j_index_end)
 378         {
 379
 380             /* Get j neighbor index, and coordinate index */
 381             jnrlistA         = jjnr[jidx];
 382             jnrlistB         = jjnr[jidx+1];
 383             jnrlistC         = jjnr[jidx+2];
 384             jnrlistD         = jjnr[jidx+3];
 385             /* Sign of each element will be negative for non-real atoms.
 386              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 387              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 388              */
 389             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 390
 391             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 392             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 393             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 394
 395             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 396             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 397             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 398             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 399             j_coord_offsetA  = DIM*jnrA;
 400             j_coord_offsetB  = DIM*jnrB;
 401             j_coord_offsetC  = DIM*jnrC;
 402             j_coord_offsetD  = DIM*jnrD;
 403
 404             /* load j atom coordinates */
 405             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 406                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 407                                                  &jx0,&jy0,&jz0);
 408
 409             /* Calculate displacement vector */
 410             dx00             = _mm256_sub_pd(ix0,jx0);
 411             dy00             = _mm256_sub_pd(iy0,jy0);
 412             dz00             = _mm256_sub_pd(iz0,jz0);
 413             dx10             = _mm256_sub_pd(ix1,jx0);
 414             dy10             = _mm256_sub_pd(iy1,jy0);
 415             dz10             = _mm256_sub_pd(iz1,jz0);
 416             dx20             = _mm256_sub_pd(ix2,jx0);
 417             dy20             = _mm256_sub_pd(iy2,jy0);
 418             dz20             = _mm256_sub_pd(iz2,jz0);
 419
 420             /* Calculate squared distance and things based on it */
 421             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 422             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 423             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 424
 425             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 426             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 427             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 428
 429             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 430             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 431             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 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             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 450             {
 451
 452             /* Compute parameters for interactions between i and j atoms */
 453             qq00             = _mm256_mul_pd(iq0,jq0);
 454             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 455                                             vdwioffsetptr0+vdwjidx0B,
 456                                             vdwioffsetptr0+vdwjidx0C,
 457                                             vdwioffsetptr0+vdwjidx0D,
 458                                             &c6_00,&c12_00);
 459
 460             /* REACTION-FIELD ELECTROSTATICS */
 461             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 462             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 463
 464             /* LENNARD-JONES DISPERSION/REPULSION */
 465
 466             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 467             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 468             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 469             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 470                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 471             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 472
 473             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 474
 475             /* Update potential sum for this i atom from the interaction with this j atom. */
 476             velec            = _mm256_and_pd(velec,cutoff_mask);
 477             velec            = _mm256_andnot_pd(dummy_mask,velec);
 478             velecsum         = _mm256_add_pd(velecsum,velec);
 479             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 480             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 481             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 482
 483             fscal            = _mm256_add_pd(felec,fvdw);
 484
 485             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 486
 487             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 488
 489             /* Calculate temporary vectorial force */
 490             tx               = _mm256_mul_pd(fscal,dx00);
 491             ty               = _mm256_mul_pd(fscal,dy00);
 492             tz               = _mm256_mul_pd(fscal,dz00);
 493
 494             /* Update vectorial force */
 495             fix0             = _mm256_add_pd(fix0,tx);
 496             fiy0             = _mm256_add_pd(fiy0,ty);
 497             fiz0             = _mm256_add_pd(fiz0,tz);
 498
 499             fjx0             = _mm256_add_pd(fjx0,tx);
 500             fjy0             = _mm256_add_pd(fjy0,ty);
 501             fjz0             = _mm256_add_pd(fjz0,tz);
 502
 503             }
 504
 505             /**************************
 506              * CALCULATE INTERACTIONS *
 507              **************************/
 508
 509             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 510             {
 511
 512             /* Compute parameters for interactions between i and j atoms */
 513             qq10             = _mm256_mul_pd(iq1,jq0);
 514
 515             /* REACTION-FIELD ELECTROSTATICS */
 516             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 517             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 518
 519             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 520
 521             /* Update potential sum for this i atom from the interaction with this j atom. */
 522             velec            = _mm256_and_pd(velec,cutoff_mask);
 523             velec            = _mm256_andnot_pd(dummy_mask,velec);
 524             velecsum         = _mm256_add_pd(velecsum,velec);
 525
 526             fscal            = felec;
 527
 528             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 529
 530             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 531
 532             /* Calculate temporary vectorial force */
 533             tx               = _mm256_mul_pd(fscal,dx10);
 534             ty               = _mm256_mul_pd(fscal,dy10);
 535             tz               = _mm256_mul_pd(fscal,dz10);
 536
 537             /* Update vectorial force */
 538             fix1             = _mm256_add_pd(fix1,tx);
 539             fiy1             = _mm256_add_pd(fiy1,ty);
 540             fiz1             = _mm256_add_pd(fiz1,tz);
 541
 542             fjx0             = _mm256_add_pd(fjx0,tx);
 543             fjy0             = _mm256_add_pd(fjy0,ty);
 544             fjz0             = _mm256_add_pd(fjz0,tz);
 545
 546             }
 547
 548             /**************************
 549              * CALCULATE INTERACTIONS *
 550              **************************/
 551
 552             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 553             {
 554
 555             /* Compute parameters for interactions between i and j atoms */
 556             qq20             = _mm256_mul_pd(iq2,jq0);
 557
 558             /* REACTION-FIELD ELECTROSTATICS */
 559             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 560             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 561
 562             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 563
 564             /* Update potential sum for this i atom from the interaction with this j atom. */
 565             velec            = _mm256_and_pd(velec,cutoff_mask);
 566             velec            = _mm256_andnot_pd(dummy_mask,velec);
 567             velecsum         = _mm256_add_pd(velecsum,velec);
 568
 569             fscal            = felec;
 570
 571             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 572
 573             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 574
 575             /* Calculate temporary vectorial force */
 576             tx               = _mm256_mul_pd(fscal,dx20);
 577             ty               = _mm256_mul_pd(fscal,dy20);
 578             tz               = _mm256_mul_pd(fscal,dz20);
 579
 580             /* Update vectorial force */
 581             fix2             = _mm256_add_pd(fix2,tx);
 582             fiy2             = _mm256_add_pd(fiy2,ty);
 583             fiz2             = _mm256_add_pd(fiz2,tz);
 584
 585             fjx0             = _mm256_add_pd(fjx0,tx);
 586             fjy0             = _mm256_add_pd(fjy0,ty);
 587             fjz0             = _mm256_add_pd(fjz0,tz);
 588
 589             }
 590
 591             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 592             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 593             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 594             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 595
 596             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 597
 598             /* Inner loop uses 129 flops */
 599         }
 600
 601         /* End of innermost loop */
 602
 603         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 604                                                  f+i_coord_offset,fshift+i_shift_offset);
 605
 606         ggid                        = gid[iidx];
 607         /* Update potential energies */
 608         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 609         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 610
 611         /* Increment number of inner iterations */
 612         inneriter                  += j_index_end - j_index_start;
 613
 614         /* Outer loop uses 20 flops */
 615     }
 616
 617     /* Increment number of outer iterations */
 618     outeriter        += nri;
 619
 620     /* Update outer/inner flops */
 621
 622     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*129);
 623 }
 624 /*
 625  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_avx_256_double
 626  * Electrostatics interaction: ReactionField
 627  * VdW interaction:            LennardJones
 628  * Geometry:                   Water3-Particle
 629  * Calculate force/pot:        Force
 630  */
 631 void
 632 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_avx_256_double
 633                     (t_nblist * gmx_restrict                nlist,
 634                      rvec * gmx_restrict                    xx,
 635                      rvec * gmx_restrict                    ff,
 636                      t_forcerec * gmx_restrict              fr,
 637                      t_mdatoms * gmx_restrict               mdatoms,
 638                      nb_kernel_data_t * gmx_restrict        kernel_data,
 639                      t_nrnb * gmx_restrict                  nrnb)
 640 {
 641     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 642      * just 0 for non-waters.
 643      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 644      * jnr indices corresponding to data put in the four positions in the SIMD register.
 645      */
 646     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 647     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 648     int              jnrA,jnrB,jnrC,jnrD;
 649     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 650     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 651     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 652     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 653     real             rcutoff_scalar;
 654     real             *shiftvec,*fshift,*x,*f;
 655     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 656     real             scratch[4*DIM];
 657     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 658     real *           vdwioffsetptr0;
 659     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 660     real *           vdwioffsetptr1;
 661     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 662     real *           vdwioffsetptr2;
 663     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 664     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 665     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 666     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 667     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 668     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 669     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 670     real             *charge;
 671     int              nvdwtype;
 672     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 673     int              *vdwtype;
 674     real             *vdwparam;
 675     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 676     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 677     __m256d          dummy_mask,cutoff_mask;
 678     __m128           tmpmask0,tmpmask1;
 679     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 680     __m256d          one     = _mm256_set1_pd(1.0);
 681     __m256d          two     = _mm256_set1_pd(2.0);
 682     x                = xx[0];
 683     f                = ff[0];
 684
 685     nri              = nlist->nri;
 686     iinr             = nlist->iinr;
 687     jindex           = nlist->jindex;
 688     jjnr             = nlist->jjnr;
 689     shiftidx         = nlist->shift;
 690     gid              = nlist->gid;
 691     shiftvec         = fr->shift_vec[0];
 692     fshift           = fr->fshift[0];
 693     facel            = _mm256_set1_pd(fr->epsfac);
 694     charge           = mdatoms->chargeA;
 695     krf              = _mm256_set1_pd(fr->ic->k_rf);
 696     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 697     crf              = _mm256_set1_pd(fr->ic->c_rf);
 698     nvdwtype         = fr->ntype;
 699     vdwparam         = fr->nbfp;
 700     vdwtype          = mdatoms->typeA;
 701
 702     /* Setup water-specific parameters */
 703     inr              = nlist->iinr[0];
 704     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 705     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 706     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 707     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 708
 709     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 710     rcutoff_scalar   = fr->rcoulomb;
 711     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 712     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 713
 714     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 715     rvdw             = _mm256_set1_pd(fr->rvdw);
 716
 717     /* Avoid stupid compiler warnings */
 718     jnrA = jnrB = jnrC = jnrD = 0;
 719     j_coord_offsetA = 0;
 720     j_coord_offsetB = 0;
 721     j_coord_offsetC = 0;
 722     j_coord_offsetD = 0;
 723
 724     outeriter        = 0;
 725     inneriter        = 0;
 726
 727     for(iidx=0;iidx<4*DIM;iidx++)
 728     {
 729         scratch[iidx] = 0.0;
 730     }
 731
 732     /* Start outer loop over neighborlists */
 733     for(iidx=0; iidx<nri; iidx++)
 734     {
 735         /* Load shift vector for this list */
 736         i_shift_offset   = DIM*shiftidx[iidx];
 737
 738         /* Load limits for loop over neighbors */
 739         j_index_start    = jindex[iidx];
 740         j_index_end      = jindex[iidx+1];
 741
 742         /* Get outer coordinate index */
 743         inr              = iinr[iidx];
 744         i_coord_offset   = DIM*inr;
 745
 746         /* Load i particle coords and add shift vector */
 747         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 748                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 749
 750         fix0             = _mm256_setzero_pd();
 751         fiy0             = _mm256_setzero_pd();
 752         fiz0             = _mm256_setzero_pd();
 753         fix1             = _mm256_setzero_pd();
 754         fiy1             = _mm256_setzero_pd();
 755         fiz1             = _mm256_setzero_pd();
 756         fix2             = _mm256_setzero_pd();
 757         fiy2             = _mm256_setzero_pd();
 758         fiz2             = _mm256_setzero_pd();
 759
 760         /* Start inner kernel loop */
 761         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 762         {
 763
 764             /* Get j neighbor index, and coordinate index */
 765             jnrA             = jjnr[jidx];
 766             jnrB             = jjnr[jidx+1];
 767             jnrC             = jjnr[jidx+2];
 768             jnrD             = jjnr[jidx+3];
 769             j_coord_offsetA  = DIM*jnrA;
 770             j_coord_offsetB  = DIM*jnrB;
 771             j_coord_offsetC  = DIM*jnrC;
 772             j_coord_offsetD  = DIM*jnrD;
 773
 774             /* load j atom coordinates */
 775             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 776                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 777                                                  &jx0,&jy0,&jz0);
 778
 779             /* Calculate displacement vector */
 780             dx00             = _mm256_sub_pd(ix0,jx0);
 781             dy00             = _mm256_sub_pd(iy0,jy0);
 782             dz00             = _mm256_sub_pd(iz0,jz0);
 783             dx10             = _mm256_sub_pd(ix1,jx0);
 784             dy10             = _mm256_sub_pd(iy1,jy0);
 785             dz10             = _mm256_sub_pd(iz1,jz0);
 786             dx20             = _mm256_sub_pd(ix2,jx0);
 787             dy20             = _mm256_sub_pd(iy2,jy0);
 788             dz20             = _mm256_sub_pd(iz2,jz0);
 789
 790             /* Calculate squared distance and things based on it */
 791             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 792             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 793             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 794
 795             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 796             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 797             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 798
 799             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 800             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 801             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 802
 803             /* Load parameters for j particles */
 804             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 805                                                                  charge+jnrC+0,charge+jnrD+0);
 806             vdwjidx0A        = 2*vdwtype[jnrA+0];
 807             vdwjidx0B        = 2*vdwtype[jnrB+0];
 808             vdwjidx0C        = 2*vdwtype[jnrC+0];
 809             vdwjidx0D        = 2*vdwtype[jnrD+0];
 810
 811             fjx0             = _mm256_setzero_pd();
 812             fjy0             = _mm256_setzero_pd();
 813             fjz0             = _mm256_setzero_pd();
 814
 815             /**************************
 816              * CALCULATE INTERACTIONS *
 817              **************************/
 818
 819             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 820             {
 821
 822             /* Compute parameters for interactions between i and j atoms */
 823             qq00             = _mm256_mul_pd(iq0,jq0);
 824             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 825                                             vdwioffsetptr0+vdwjidx0B,
 826                                             vdwioffsetptr0+vdwjidx0C,
 827                                             vdwioffsetptr0+vdwjidx0D,
 828                                             &c6_00,&c12_00);
 829
 830             /* REACTION-FIELD ELECTROSTATICS */
 831             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 832
 833             /* LENNARD-JONES DISPERSION/REPULSION */
 834
 835             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 836             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 837
 838             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 839
 840             fscal            = _mm256_add_pd(felec,fvdw);
 841
 842             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 843
 844             /* Calculate temporary vectorial force */
 845             tx               = _mm256_mul_pd(fscal,dx00);
 846             ty               = _mm256_mul_pd(fscal,dy00);
 847             tz               = _mm256_mul_pd(fscal,dz00);
 848
 849             /* Update vectorial force */
 850             fix0             = _mm256_add_pd(fix0,tx);
 851             fiy0             = _mm256_add_pd(fiy0,ty);
 852             fiz0             = _mm256_add_pd(fiz0,tz);
 853
 854             fjx0             = _mm256_add_pd(fjx0,tx);
 855             fjy0             = _mm256_add_pd(fjy0,ty);
 856             fjz0             = _mm256_add_pd(fjz0,tz);
 857
 858             }
 859
 860             /**************************
 861              * CALCULATE INTERACTIONS *
 862              **************************/
 863
 864             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 865             {
 866
 867             /* Compute parameters for interactions between i and j atoms */
 868             qq10             = _mm256_mul_pd(iq1,jq0);
 869
 870             /* REACTION-FIELD ELECTROSTATICS */
 871             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 872
 873             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 874
 875             fscal            = felec;
 876
 877             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 878
 879             /* Calculate temporary vectorial force */
 880             tx               = _mm256_mul_pd(fscal,dx10);
 881             ty               = _mm256_mul_pd(fscal,dy10);
 882             tz               = _mm256_mul_pd(fscal,dz10);
 883
 884             /* Update vectorial force */
 885             fix1             = _mm256_add_pd(fix1,tx);
 886             fiy1             = _mm256_add_pd(fiy1,ty);
 887             fiz1             = _mm256_add_pd(fiz1,tz);
 888
 889             fjx0             = _mm256_add_pd(fjx0,tx);
 890             fjy0             = _mm256_add_pd(fjy0,ty);
 891             fjz0             = _mm256_add_pd(fjz0,tz);
 892
 893             }
 894
 895             /**************************
 896              * CALCULATE INTERACTIONS *
 897              **************************/
 898
 899             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 900             {
 901
 902             /* Compute parameters for interactions between i and j atoms */
 903             qq20             = _mm256_mul_pd(iq2,jq0);
 904
 905             /* REACTION-FIELD ELECTROSTATICS */
 906             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 907
 908             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 909
 910             fscal            = felec;
 911
 912             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 913
 914             /* Calculate temporary vectorial force */
 915             tx               = _mm256_mul_pd(fscal,dx20);
 916             ty               = _mm256_mul_pd(fscal,dy20);
 917             tz               = _mm256_mul_pd(fscal,dz20);
 918
 919             /* Update vectorial force */
 920             fix2             = _mm256_add_pd(fix2,tx);
 921             fiy2             = _mm256_add_pd(fiy2,ty);
 922             fiz2             = _mm256_add_pd(fiz2,tz);
 923
 924             fjx0             = _mm256_add_pd(fjx0,tx);
 925             fjy0             = _mm256_add_pd(fjy0,ty);
 926             fjz0             = _mm256_add_pd(fjz0,tz);
 927
 928             }
 929
 930             fjptrA             = f+j_coord_offsetA;
 931             fjptrB             = f+j_coord_offsetB;
 932             fjptrC             = f+j_coord_offsetC;
 933             fjptrD             = f+j_coord_offsetD;
 934
 935             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 936
 937             /* Inner loop uses 100 flops */
 938         }
 939
 940         if(jidx<j_index_end)
 941         {
 942
 943             /* Get j neighbor index, and coordinate index */
 944             jnrlistA         = jjnr[jidx];
 945             jnrlistB         = jjnr[jidx+1];
 946             jnrlistC         = jjnr[jidx+2];
 947             jnrlistD         = jjnr[jidx+3];
 948             /* Sign of each element will be negative for non-real atoms.
 949              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 950              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 951              */
 952             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 953
 954             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 955             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 956             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 957
 958             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 959             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 960             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 961             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 962             j_coord_offsetA  = DIM*jnrA;
 963             j_coord_offsetB  = DIM*jnrB;
 964             j_coord_offsetC  = DIM*jnrC;
 965             j_coord_offsetD  = DIM*jnrD;
 966
 967             /* load j atom coordinates */
 968             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 969                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 970                                                  &jx0,&jy0,&jz0);
 971
 972             /* Calculate displacement vector */
 973             dx00             = _mm256_sub_pd(ix0,jx0);
 974             dy00             = _mm256_sub_pd(iy0,jy0);
 975             dz00             = _mm256_sub_pd(iz0,jz0);
 976             dx10             = _mm256_sub_pd(ix1,jx0);
 977             dy10             = _mm256_sub_pd(iy1,jy0);
 978             dz10             = _mm256_sub_pd(iz1,jz0);
 979             dx20             = _mm256_sub_pd(ix2,jx0);
 980             dy20             = _mm256_sub_pd(iy2,jy0);
 981             dz20             = _mm256_sub_pd(iz2,jz0);
 982
 983             /* Calculate squared distance and things based on it */
 984             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 985             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 986             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 987
 988             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 989             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 990             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 991
 992             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 993             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 994             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 995
 996             /* Load parameters for j particles */
 997             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 998                                                                  charge+jnrC+0,charge+jnrD+0);
 999             vdwjidx0A        = 2*vdwtype[jnrA+0];
1000             vdwjidx0B        = 2*vdwtype[jnrB+0];
1001             vdwjidx0C        = 2*vdwtype[jnrC+0];
1002             vdwjidx0D        = 2*vdwtype[jnrD+0];
1003
1004             fjx0             = _mm256_setzero_pd();
1005             fjy0             = _mm256_setzero_pd();
1006             fjz0             = _mm256_setzero_pd();
1007
1008             /**************************
1009              * CALCULATE INTERACTIONS *
1010              **************************/
1011
1012             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1013             {
1014
1015             /* Compute parameters for interactions between i and j atoms */
1016             qq00             = _mm256_mul_pd(iq0,jq0);
1017             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1018                                             vdwioffsetptr0+vdwjidx0B,
1019                                             vdwioffsetptr0+vdwjidx0C,
1020                                             vdwioffsetptr0+vdwjidx0D,
1021                                             &c6_00,&c12_00);
1022
1023             /* REACTION-FIELD ELECTROSTATICS */
1024             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
1025
1026             /* LENNARD-JONES DISPERSION/REPULSION */
1027
1028             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1029             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1030
1031             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1032
1033             fscal            = _mm256_add_pd(felec,fvdw);
1034
1035             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1036
1037             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1038
1039             /* Calculate temporary vectorial force */
1040             tx               = _mm256_mul_pd(fscal,dx00);
1041             ty               = _mm256_mul_pd(fscal,dy00);
1042             tz               = _mm256_mul_pd(fscal,dz00);
1043
1044             /* Update vectorial force */
1045             fix0             = _mm256_add_pd(fix0,tx);
1046             fiy0             = _mm256_add_pd(fiy0,ty);
1047             fiz0             = _mm256_add_pd(fiz0,tz);
1048
1049             fjx0             = _mm256_add_pd(fjx0,tx);
1050             fjy0             = _mm256_add_pd(fjy0,ty);
1051             fjz0             = _mm256_add_pd(fjz0,tz);
1052
1053             }
1054
1055             /**************************
1056              * CALCULATE INTERACTIONS *
1057              **************************/
1058
1059             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1060             {
1061
1062             /* Compute parameters for interactions between i and j atoms */
1063             qq10             = _mm256_mul_pd(iq1,jq0);
1064
1065             /* REACTION-FIELD ELECTROSTATICS */
1066             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1067
1068             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1069
1070             fscal            = felec;
1071
1072             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1073
1074             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1075
1076             /* Calculate temporary vectorial force */
1077             tx               = _mm256_mul_pd(fscal,dx10);
1078             ty               = _mm256_mul_pd(fscal,dy10);
1079             tz               = _mm256_mul_pd(fscal,dz10);
1080
1081             /* Update vectorial force */
1082             fix1             = _mm256_add_pd(fix1,tx);
1083             fiy1             = _mm256_add_pd(fiy1,ty);
1084             fiz1             = _mm256_add_pd(fiz1,tz);
1085
1086             fjx0             = _mm256_add_pd(fjx0,tx);
1087             fjy0             = _mm256_add_pd(fjy0,ty);
1088             fjz0             = _mm256_add_pd(fjz0,tz);
1089
1090             }
1091
1092             /**************************
1093              * CALCULATE INTERACTIONS *
1094              **************************/
1095
1096             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1097             {
1098
1099             /* Compute parameters for interactions between i and j atoms */
1100             qq20             = _mm256_mul_pd(iq2,jq0);
1101
1102             /* REACTION-FIELD ELECTROSTATICS */
1103             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1104
1105             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1106
1107             fscal            = felec;
1108
1109             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1110
1111             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1112
1113             /* Calculate temporary vectorial force */
1114             tx               = _mm256_mul_pd(fscal,dx20);
1115             ty               = _mm256_mul_pd(fscal,dy20);
1116             tz               = _mm256_mul_pd(fscal,dz20);
1117
1118             /* Update vectorial force */
1119             fix2             = _mm256_add_pd(fix2,tx);
1120             fiy2             = _mm256_add_pd(fiy2,ty);
1121             fiz2             = _mm256_add_pd(fiz2,tz);
1122
1123             fjx0             = _mm256_add_pd(fjx0,tx);
1124             fjy0             = _mm256_add_pd(fjy0,ty);
1125             fjz0             = _mm256_add_pd(fjz0,tz);
1126
1127             }
1128
1129             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1130             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1131             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1132             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1133
1134             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1135
1136             /* Inner loop uses 100 flops */
1137         }
1138
1139         /* End of innermost loop */
1140
1141         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1142                                                  f+i_coord_offset,fshift+i_shift_offset);
1143
1144         /* Increment number of inner iterations */
1145         inneriter                  += j_index_end - j_index_start;
1146
1147         /* Outer loop uses 18 flops */
1148     }
1149
1150     /* Increment number of outer iterations */
1151     outeriter        += nri;
1152
1153     /* Update outer/inner flops */
1154
1155     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*100);
1156 }