src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_VF_avx_256_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_VF_avx_256_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  93     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  96     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  97     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 104     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 105     __m256d          dummy_mask,cutoff_mask;
 106     __m128           tmpmask0,tmpmask1;
 107     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 108     __m256d          one     = _mm256_set1_pd(1.0);
 109     __m256d          two     = _mm256_set1_pd(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm256_set1_pd(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     krf              = _mm256_set1_pd(fr->ic->k_rf);
 124     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 125     crf              = _mm256_set1_pd(fr->ic->c_rf);
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     /* Setup water-specific parameters */
 131     inr              = nlist->iinr[0];
 132     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 133     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 134     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 135     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 136
 137     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 138     rcutoff_scalar   = fr->rcoulomb;
 139     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 140     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 141
 142     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 143     rvdw             = _mm256_set1_pd(fr->rvdw);
 144
 145     /* Avoid stupid compiler warnings */
 146     jnrA = jnrB = jnrC = jnrD = 0;
 147     j_coord_offsetA = 0;
 148     j_coord_offsetB = 0;
 149     j_coord_offsetC = 0;
 150     j_coord_offsetD = 0;
 151
 152     outeriter        = 0;
 153     inneriter        = 0;
 154
 155     for(iidx=0;iidx<4*DIM;iidx++)
 156     {
 157         scratch[iidx] = 0.0;
 158     }
 159
 160     /* Start outer loop over neighborlists */
 161     for(iidx=0; iidx<nri; iidx++)
 162     {
 163         /* Load shift vector for this list */
 164         i_shift_offset   = DIM*shiftidx[iidx];
 165
 166         /* Load limits for loop over neighbors */
 167         j_index_start    = jindex[iidx];
 168         j_index_end      = jindex[iidx+1];
 169
 170         /* Get outer coordinate index */
 171         inr              = iinr[iidx];
 172         i_coord_offset   = DIM*inr;
 173
 174         /* Load i particle coords and add shift vector */
 175         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 176                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 177
 178         fix0             = _mm256_setzero_pd();
 179         fiy0             = _mm256_setzero_pd();
 180         fiz0             = _mm256_setzero_pd();
 181         fix1             = _mm256_setzero_pd();
 182         fiy1             = _mm256_setzero_pd();
 183         fiz1             = _mm256_setzero_pd();
 184         fix2             = _mm256_setzero_pd();
 185         fiy2             = _mm256_setzero_pd();
 186         fiz2             = _mm256_setzero_pd();
 187
 188         /* Reset potential sums */
 189         velecsum         = _mm256_setzero_pd();
 190         vvdwsum          = _mm256_setzero_pd();
 191
 192         /* Start inner kernel loop */
 193         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 194         {
 195
 196             /* Get j neighbor index, and coordinate index */
 197             jnrA             = jjnr[jidx];
 198             jnrB             = jjnr[jidx+1];
 199             jnrC             = jjnr[jidx+2];
 200             jnrD             = jjnr[jidx+3];
 201             j_coord_offsetA  = DIM*jnrA;
 202             j_coord_offsetB  = DIM*jnrB;
 203             j_coord_offsetC  = DIM*jnrC;
 204             j_coord_offsetD  = DIM*jnrD;
 205
 206             /* load j atom coordinates */
 207             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 208                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 209                                                  &jx0,&jy0,&jz0);
 210
 211             /* Calculate displacement vector */
 212             dx00             = _mm256_sub_pd(ix0,jx0);
 213             dy00             = _mm256_sub_pd(iy0,jy0);
 214             dz00             = _mm256_sub_pd(iz0,jz0);
 215             dx10             = _mm256_sub_pd(ix1,jx0);
 216             dy10             = _mm256_sub_pd(iy1,jy0);
 217             dz10             = _mm256_sub_pd(iz1,jz0);
 218             dx20             = _mm256_sub_pd(ix2,jx0);
 219             dy20             = _mm256_sub_pd(iy2,jy0);
 220             dz20             = _mm256_sub_pd(iz2,jz0);
 221
 222             /* Calculate squared distance and things based on it */
 223             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 224             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 225             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 226
 227             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 228             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 229             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 230
 231             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 232             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 233             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 234
 235             /* Load parameters for j particles */
 236             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 237                                                                  charge+jnrC+0,charge+jnrD+0);
 238             vdwjidx0A        = 2*vdwtype[jnrA+0];
 239             vdwjidx0B        = 2*vdwtype[jnrB+0];
 240             vdwjidx0C        = 2*vdwtype[jnrC+0];
 241             vdwjidx0D        = 2*vdwtype[jnrD+0];
 242
 243             fjx0             = _mm256_setzero_pd();
 244             fjy0             = _mm256_setzero_pd();
 245             fjz0             = _mm256_setzero_pd();
 246
 247             /**************************
 248              * CALCULATE INTERACTIONS *
 249              **************************/
 250
 251             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 252             {
 253
 254             /* Compute parameters for interactions between i and j atoms */
 255             qq00             = _mm256_mul_pd(iq0,jq0);
 256             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 257                                             vdwioffsetptr0+vdwjidx0B,
 258                                             vdwioffsetptr0+vdwjidx0C,
 259                                             vdwioffsetptr0+vdwjidx0D,
 260                                             &c6_00,&c12_00);
 261
 262             /* REACTION-FIELD ELECTROSTATICS */
 263             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 264             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 265
 266             /* LENNARD-JONES DISPERSION/REPULSION */
 267
 268             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 269             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 270             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 271             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) ,
 272                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 273             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 274
 275             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 276
 277             /* Update potential sum for this i atom from the interaction with this j atom. */
 278             velec            = _mm256_and_pd(velec,cutoff_mask);
 279             velecsum         = _mm256_add_pd(velecsum,velec);
 280             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 281             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 282
 283             fscal            = _mm256_add_pd(felec,fvdw);
 284
 285             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 286
 287             /* Calculate temporary vectorial force */
 288             tx               = _mm256_mul_pd(fscal,dx00);
 289             ty               = _mm256_mul_pd(fscal,dy00);
 290             tz               = _mm256_mul_pd(fscal,dz00);
 291
 292             /* Update vectorial force */
 293             fix0             = _mm256_add_pd(fix0,tx);
 294             fiy0             = _mm256_add_pd(fiy0,ty);
 295             fiz0             = _mm256_add_pd(fiz0,tz);
 296
 297             fjx0             = _mm256_add_pd(fjx0,tx);
 298             fjy0             = _mm256_add_pd(fjy0,ty);
 299             fjz0             = _mm256_add_pd(fjz0,tz);
 300
 301             }
 302
 303             /**************************
 304              * CALCULATE INTERACTIONS *
 305              **************************/
 306
 307             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 308             {
 309
 310             /* Compute parameters for interactions between i and j atoms */
 311             qq10             = _mm256_mul_pd(iq1,jq0);
 312
 313             /* REACTION-FIELD ELECTROSTATICS */
 314             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 315             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 316
 317             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 318
 319             /* Update potential sum for this i atom from the interaction with this j atom. */
 320             velec            = _mm256_and_pd(velec,cutoff_mask);
 321             velecsum         = _mm256_add_pd(velecsum,velec);
 322
 323             fscal            = felec;
 324
 325             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 326
 327             /* Calculate temporary vectorial force */
 328             tx               = _mm256_mul_pd(fscal,dx10);
 329             ty               = _mm256_mul_pd(fscal,dy10);
 330             tz               = _mm256_mul_pd(fscal,dz10);
 331
 332             /* Update vectorial force */
 333             fix1             = _mm256_add_pd(fix1,tx);
 334             fiy1             = _mm256_add_pd(fiy1,ty);
 335             fiz1             = _mm256_add_pd(fiz1,tz);
 336
 337             fjx0             = _mm256_add_pd(fjx0,tx);
 338             fjy0             = _mm256_add_pd(fjy0,ty);
 339             fjz0             = _mm256_add_pd(fjz0,tz);
 340
 341             }
 342
 343             /**************************
 344              * CALCULATE INTERACTIONS *
 345              **************************/
 346
 347             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 348             {
 349
 350             /* Compute parameters for interactions between i and j atoms */
 351             qq20             = _mm256_mul_pd(iq2,jq0);
 352
 353             /* REACTION-FIELD ELECTROSTATICS */
 354             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 355             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 356
 357             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 358
 359             /* Update potential sum for this i atom from the interaction with this j atom. */
 360             velec            = _mm256_and_pd(velec,cutoff_mask);
 361             velecsum         = _mm256_add_pd(velecsum,velec);
 362
 363             fscal            = felec;
 364
 365             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 366
 367             /* Calculate temporary vectorial force */
 368             tx               = _mm256_mul_pd(fscal,dx20);
 369             ty               = _mm256_mul_pd(fscal,dy20);
 370             tz               = _mm256_mul_pd(fscal,dz20);
 371
 372             /* Update vectorial force */
 373             fix2             = _mm256_add_pd(fix2,tx);
 374             fiy2             = _mm256_add_pd(fiy2,ty);
 375             fiz2             = _mm256_add_pd(fiz2,tz);
 376
 377             fjx0             = _mm256_add_pd(fjx0,tx);
 378             fjy0             = _mm256_add_pd(fjy0,ty);
 379             fjz0             = _mm256_add_pd(fjz0,tz);
 380
 381             }
 382
 383             fjptrA             = f+j_coord_offsetA;
 384             fjptrB             = f+j_coord_offsetB;
 385             fjptrC             = f+j_coord_offsetC;
 386             fjptrD             = f+j_coord_offsetD;
 387
 388             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 389
 390             /* Inner loop uses 129 flops */
 391         }
 392
 393         if(jidx<j_index_end)
 394         {
 395
 396             /* Get j neighbor index, and coordinate index */
 397             jnrlistA         = jjnr[jidx];
 398             jnrlistB         = jjnr[jidx+1];
 399             jnrlistC         = jjnr[jidx+2];
 400             jnrlistD         = jjnr[jidx+3];
 401             /* Sign of each element will be negative for non-real atoms.
 402              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 403              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 404              */
 405             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 406
 407             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 408             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 409             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 410
 411             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 412             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 413             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 414             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 415             j_coord_offsetA  = DIM*jnrA;
 416             j_coord_offsetB  = DIM*jnrB;
 417             j_coord_offsetC  = DIM*jnrC;
 418             j_coord_offsetD  = DIM*jnrD;
 419
 420             /* load j atom coordinates */
 421             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 422                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 423                                                  &jx0,&jy0,&jz0);
 424
 425             /* Calculate displacement vector */
 426             dx00             = _mm256_sub_pd(ix0,jx0);
 427             dy00             = _mm256_sub_pd(iy0,jy0);
 428             dz00             = _mm256_sub_pd(iz0,jz0);
 429             dx10             = _mm256_sub_pd(ix1,jx0);
 430             dy10             = _mm256_sub_pd(iy1,jy0);
 431             dz10             = _mm256_sub_pd(iz1,jz0);
 432             dx20             = _mm256_sub_pd(ix2,jx0);
 433             dy20             = _mm256_sub_pd(iy2,jy0);
 434             dz20             = _mm256_sub_pd(iz2,jz0);
 435
 436             /* Calculate squared distance and things based on it */
 437             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 438             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 439             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 440
 441             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 442             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 443             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 444
 445             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 446             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 447             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 448
 449             /* Load parameters for j particles */
 450             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 451                                                                  charge+jnrC+0,charge+jnrD+0);
 452             vdwjidx0A        = 2*vdwtype[jnrA+0];
 453             vdwjidx0B        = 2*vdwtype[jnrB+0];
 454             vdwjidx0C        = 2*vdwtype[jnrC+0];
 455             vdwjidx0D        = 2*vdwtype[jnrD+0];
 456
 457             fjx0             = _mm256_setzero_pd();
 458             fjy0             = _mm256_setzero_pd();
 459             fjz0             = _mm256_setzero_pd();
 460
 461             /**************************
 462              * CALCULATE INTERACTIONS *
 463              **************************/
 464
 465             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 466             {
 467
 468             /* Compute parameters for interactions between i and j atoms */
 469             qq00             = _mm256_mul_pd(iq0,jq0);
 470             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 471                                             vdwioffsetptr0+vdwjidx0B,
 472                                             vdwioffsetptr0+vdwjidx0C,
 473                                             vdwioffsetptr0+vdwjidx0D,
 474                                             &c6_00,&c12_00);
 475
 476             /* REACTION-FIELD ELECTROSTATICS */
 477             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 478             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 479
 480             /* LENNARD-JONES DISPERSION/REPULSION */
 481
 482             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 483             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 484             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 485             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) ,
 486                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 487             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 488
 489             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 490
 491             /* Update potential sum for this i atom from the interaction with this j atom. */
 492             velec            = _mm256_and_pd(velec,cutoff_mask);
 493             velec            = _mm256_andnot_pd(dummy_mask,velec);
 494             velecsum         = _mm256_add_pd(velecsum,velec);
 495             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 496             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 497             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 498
 499             fscal            = _mm256_add_pd(felec,fvdw);
 500
 501             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 502
 503             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 504
 505             /* Calculate temporary vectorial force */
 506             tx               = _mm256_mul_pd(fscal,dx00);
 507             ty               = _mm256_mul_pd(fscal,dy00);
 508             tz               = _mm256_mul_pd(fscal,dz00);
 509
 510             /* Update vectorial force */
 511             fix0             = _mm256_add_pd(fix0,tx);
 512             fiy0             = _mm256_add_pd(fiy0,ty);
 513             fiz0             = _mm256_add_pd(fiz0,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
 521             /**************************
 522              * CALCULATE INTERACTIONS *
 523              **************************/
 524
 525             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 526             {
 527
 528             /* Compute parameters for interactions between i and j atoms */
 529             qq10             = _mm256_mul_pd(iq1,jq0);
 530
 531             /* REACTION-FIELD ELECTROSTATICS */
 532             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 533             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 534
 535             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 536
 537             /* Update potential sum for this i atom from the interaction with this j atom. */
 538             velec            = _mm256_and_pd(velec,cutoff_mask);
 539             velec            = _mm256_andnot_pd(dummy_mask,velec);
 540             velecsum         = _mm256_add_pd(velecsum,velec);
 541
 542             fscal            = felec;
 543
 544             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 545
 546             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 547
 548             /* Calculate temporary vectorial force */
 549             tx               = _mm256_mul_pd(fscal,dx10);
 550             ty               = _mm256_mul_pd(fscal,dy10);
 551             tz               = _mm256_mul_pd(fscal,dz10);
 552
 553             /* Update vectorial force */
 554             fix1             = _mm256_add_pd(fix1,tx);
 555             fiy1             = _mm256_add_pd(fiy1,ty);
 556             fiz1             = _mm256_add_pd(fiz1,tz);
 557
 558             fjx0             = _mm256_add_pd(fjx0,tx);
 559             fjy0             = _mm256_add_pd(fjy0,ty);
 560             fjz0             = _mm256_add_pd(fjz0,tz);
 561
 562             }
 563
 564             /**************************
 565              * CALCULATE INTERACTIONS *
 566              **************************/
 567
 568             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 569             {
 570
 571             /* Compute parameters for interactions between i and j atoms */
 572             qq20             = _mm256_mul_pd(iq2,jq0);
 573
 574             /* REACTION-FIELD ELECTROSTATICS */
 575             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 576             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 577
 578             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 579
 580             /* Update potential sum for this i atom from the interaction with this j atom. */
 581             velec            = _mm256_and_pd(velec,cutoff_mask);
 582             velec            = _mm256_andnot_pd(dummy_mask,velec);
 583             velecsum         = _mm256_add_pd(velecsum,velec);
 584
 585             fscal            = felec;
 586
 587             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 588
 589             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 590
 591             /* Calculate temporary vectorial force */
 592             tx               = _mm256_mul_pd(fscal,dx20);
 593             ty               = _mm256_mul_pd(fscal,dy20);
 594             tz               = _mm256_mul_pd(fscal,dz20);
 595
 596             /* Update vectorial force */
 597             fix2             = _mm256_add_pd(fix2,tx);
 598             fiy2             = _mm256_add_pd(fiy2,ty);
 599             fiz2             = _mm256_add_pd(fiz2,tz);
 600
 601             fjx0             = _mm256_add_pd(fjx0,tx);
 602             fjy0             = _mm256_add_pd(fjy0,ty);
 603             fjz0             = _mm256_add_pd(fjz0,tz);
 604
 605             }
 606
 607             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 608             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 609             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 610             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 611
 612             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 613
 614             /* Inner loop uses 129 flops */
 615         }
 616
 617         /* End of innermost loop */
 618
 619         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 620                                                  f+i_coord_offset,fshift+i_shift_offset);
 621
 622         ggid                        = gid[iidx];
 623         /* Update potential energies */
 624         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 625         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 626
 627         /* Increment number of inner iterations */
 628         inneriter                  += j_index_end - j_index_start;
 629
 630         /* Outer loop uses 20 flops */
 631     }
 632
 633     /* Increment number of outer iterations */
 634     outeriter        += nri;
 635
 636     /* Update outer/inner flops */
 637
 638     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*129);
 639 }
 640 /*
 641  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_avx_256_double
 642  * Electrostatics interaction: ReactionField
 643  * VdW interaction:            LennardJones
 644  * Geometry:                   Water3-Particle
 645  * Calculate force/pot:        Force
 646  */
 647 void
 648 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_avx_256_double
 649                     (t_nblist                    * gmx_restrict       nlist,
 650                      rvec                        * gmx_restrict          xx,
 651                      rvec                        * gmx_restrict          ff,
 652                      t_forcerec                  * gmx_restrict          fr,
 653                      t_mdatoms                   * gmx_restrict     mdatoms,
 654                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 655                      t_nrnb                      * gmx_restrict        nrnb)
 656 {
 657     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 658      * just 0 for non-waters.
 659      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 660      * jnr indices corresponding to data put in the four positions in the SIMD register.
 661      */
 662     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 663     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 664     int              jnrA,jnrB,jnrC,jnrD;
 665     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 666     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 667     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 668     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 669     real             rcutoff_scalar;
 670     real             *shiftvec,*fshift,*x,*f;
 671     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 672     real             scratch[4*DIM];
 673     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 674     real *           vdwioffsetptr0;
 675     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 676     real *           vdwioffsetptr1;
 677     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 678     real *           vdwioffsetptr2;
 679     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 680     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 681     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 682     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 683     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 684     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 685     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 686     real             *charge;
 687     int              nvdwtype;
 688     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 689     int              *vdwtype;
 690     real             *vdwparam;
 691     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 692     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 693     __m256d          dummy_mask,cutoff_mask;
 694     __m128           tmpmask0,tmpmask1;
 695     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 696     __m256d          one     = _mm256_set1_pd(1.0);
 697     __m256d          two     = _mm256_set1_pd(2.0);
 698     x                = xx[0];
 699     f                = ff[0];
 700
 701     nri              = nlist->nri;
 702     iinr             = nlist->iinr;
 703     jindex           = nlist->jindex;
 704     jjnr             = nlist->jjnr;
 705     shiftidx         = nlist->shift;
 706     gid              = nlist->gid;
 707     shiftvec         = fr->shift_vec[0];
 708     fshift           = fr->fshift[0];
 709     facel            = _mm256_set1_pd(fr->epsfac);
 710     charge           = mdatoms->chargeA;
 711     krf              = _mm256_set1_pd(fr->ic->k_rf);
 712     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 713     crf              = _mm256_set1_pd(fr->ic->c_rf);
 714     nvdwtype         = fr->ntype;
 715     vdwparam         = fr->nbfp;
 716     vdwtype          = mdatoms->typeA;
 717
 718     /* Setup water-specific parameters */
 719     inr              = nlist->iinr[0];
 720     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 721     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 722     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 723     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 724
 725     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 726     rcutoff_scalar   = fr->rcoulomb;
 727     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 728     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 729
 730     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 731     rvdw             = _mm256_set1_pd(fr->rvdw);
 732
 733     /* Avoid stupid compiler warnings */
 734     jnrA = jnrB = jnrC = jnrD = 0;
 735     j_coord_offsetA = 0;
 736     j_coord_offsetB = 0;
 737     j_coord_offsetC = 0;
 738     j_coord_offsetD = 0;
 739
 740     outeriter        = 0;
 741     inneriter        = 0;
 742
 743     for(iidx=0;iidx<4*DIM;iidx++)
 744     {
 745         scratch[iidx] = 0.0;
 746     }
 747
 748     /* Start outer loop over neighborlists */
 749     for(iidx=0; iidx<nri; iidx++)
 750     {
 751         /* Load shift vector for this list */
 752         i_shift_offset   = DIM*shiftidx[iidx];
 753
 754         /* Load limits for loop over neighbors */
 755         j_index_start    = jindex[iidx];
 756         j_index_end      = jindex[iidx+1];
 757
 758         /* Get outer coordinate index */
 759         inr              = iinr[iidx];
 760         i_coord_offset   = DIM*inr;
 761
 762         /* Load i particle coords and add shift vector */
 763         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 764                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 765
 766         fix0             = _mm256_setzero_pd();
 767         fiy0             = _mm256_setzero_pd();
 768         fiz0             = _mm256_setzero_pd();
 769         fix1             = _mm256_setzero_pd();
 770         fiy1             = _mm256_setzero_pd();
 771         fiz1             = _mm256_setzero_pd();
 772         fix2             = _mm256_setzero_pd();
 773         fiy2             = _mm256_setzero_pd();
 774         fiz2             = _mm256_setzero_pd();
 775
 776         /* Start inner kernel loop */
 777         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 778         {
 779
 780             /* Get j neighbor index, and coordinate index */
 781             jnrA             = jjnr[jidx];
 782             jnrB             = jjnr[jidx+1];
 783             jnrC             = jjnr[jidx+2];
 784             jnrD             = jjnr[jidx+3];
 785             j_coord_offsetA  = DIM*jnrA;
 786             j_coord_offsetB  = DIM*jnrB;
 787             j_coord_offsetC  = DIM*jnrC;
 788             j_coord_offsetD  = DIM*jnrD;
 789
 790             /* load j atom coordinates */
 791             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 792                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 793                                                  &jx0,&jy0,&jz0);
 794
 795             /* Calculate displacement vector */
 796             dx00             = _mm256_sub_pd(ix0,jx0);
 797             dy00             = _mm256_sub_pd(iy0,jy0);
 798             dz00             = _mm256_sub_pd(iz0,jz0);
 799             dx10             = _mm256_sub_pd(ix1,jx0);
 800             dy10             = _mm256_sub_pd(iy1,jy0);
 801             dz10             = _mm256_sub_pd(iz1,jz0);
 802             dx20             = _mm256_sub_pd(ix2,jx0);
 803             dy20             = _mm256_sub_pd(iy2,jy0);
 804             dz20             = _mm256_sub_pd(iz2,jz0);
 805
 806             /* Calculate squared distance and things based on it */
 807             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 808             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 809             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 810
 811             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 812             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 813             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 814
 815             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 816             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 817             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 818
 819             /* Load parameters for j particles */
 820             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 821                                                                  charge+jnrC+0,charge+jnrD+0);
 822             vdwjidx0A        = 2*vdwtype[jnrA+0];
 823             vdwjidx0B        = 2*vdwtype[jnrB+0];
 824             vdwjidx0C        = 2*vdwtype[jnrC+0];
 825             vdwjidx0D        = 2*vdwtype[jnrD+0];
 826
 827             fjx0             = _mm256_setzero_pd();
 828             fjy0             = _mm256_setzero_pd();
 829             fjz0             = _mm256_setzero_pd();
 830
 831             /**************************
 832              * CALCULATE INTERACTIONS *
 833              **************************/
 834
 835             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 836             {
 837
 838             /* Compute parameters for interactions between i and j atoms */
 839             qq00             = _mm256_mul_pd(iq0,jq0);
 840             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 841                                             vdwioffsetptr0+vdwjidx0B,
 842                                             vdwioffsetptr0+vdwjidx0C,
 843                                             vdwioffsetptr0+vdwjidx0D,
 844                                             &c6_00,&c12_00);
 845
 846             /* REACTION-FIELD ELECTROSTATICS */
 847             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 848
 849             /* LENNARD-JONES DISPERSION/REPULSION */
 850
 851             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 852             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 853
 854             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 855
 856             fscal            = _mm256_add_pd(felec,fvdw);
 857
 858             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 859
 860             /* Calculate temporary vectorial force */
 861             tx               = _mm256_mul_pd(fscal,dx00);
 862             ty               = _mm256_mul_pd(fscal,dy00);
 863             tz               = _mm256_mul_pd(fscal,dz00);
 864
 865             /* Update vectorial force */
 866             fix0             = _mm256_add_pd(fix0,tx);
 867             fiy0             = _mm256_add_pd(fiy0,ty);
 868             fiz0             = _mm256_add_pd(fiz0,tz);
 869
 870             fjx0             = _mm256_add_pd(fjx0,tx);
 871             fjy0             = _mm256_add_pd(fjy0,ty);
 872             fjz0             = _mm256_add_pd(fjz0,tz);
 873
 874             }
 875
 876             /**************************
 877              * CALCULATE INTERACTIONS *
 878              **************************/
 879
 880             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 881             {
 882
 883             /* Compute parameters for interactions between i and j atoms */
 884             qq10             = _mm256_mul_pd(iq1,jq0);
 885
 886             /* REACTION-FIELD ELECTROSTATICS */
 887             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 888
 889             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 890
 891             fscal            = felec;
 892
 893             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 894
 895             /* Calculate temporary vectorial force */
 896             tx               = _mm256_mul_pd(fscal,dx10);
 897             ty               = _mm256_mul_pd(fscal,dy10);
 898             tz               = _mm256_mul_pd(fscal,dz10);
 899
 900             /* Update vectorial force */
 901             fix1             = _mm256_add_pd(fix1,tx);
 902             fiy1             = _mm256_add_pd(fiy1,ty);
 903             fiz1             = _mm256_add_pd(fiz1,tz);
 904
 905             fjx0             = _mm256_add_pd(fjx0,tx);
 906             fjy0             = _mm256_add_pd(fjy0,ty);
 907             fjz0             = _mm256_add_pd(fjz0,tz);
 908
 909             }
 910
 911             /**************************
 912              * CALCULATE INTERACTIONS *
 913              **************************/
 914
 915             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 916             {
 917
 918             /* Compute parameters for interactions between i and j atoms */
 919             qq20             = _mm256_mul_pd(iq2,jq0);
 920
 921             /* REACTION-FIELD ELECTROSTATICS */
 922             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 923
 924             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 925
 926             fscal            = felec;
 927
 928             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 929
 930             /* Calculate temporary vectorial force */
 931             tx               = _mm256_mul_pd(fscal,dx20);
 932             ty               = _mm256_mul_pd(fscal,dy20);
 933             tz               = _mm256_mul_pd(fscal,dz20);
 934
 935             /* Update vectorial force */
 936             fix2             = _mm256_add_pd(fix2,tx);
 937             fiy2             = _mm256_add_pd(fiy2,ty);
 938             fiz2             = _mm256_add_pd(fiz2,tz);
 939
 940             fjx0             = _mm256_add_pd(fjx0,tx);
 941             fjy0             = _mm256_add_pd(fjy0,ty);
 942             fjz0             = _mm256_add_pd(fjz0,tz);
 943
 944             }
 945
 946             fjptrA             = f+j_coord_offsetA;
 947             fjptrB             = f+j_coord_offsetB;
 948             fjptrC             = f+j_coord_offsetC;
 949             fjptrD             = f+j_coord_offsetD;
 950
 951             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 952
 953             /* Inner loop uses 100 flops */
 954         }
 955
 956         if(jidx<j_index_end)
 957         {
 958
 959             /* Get j neighbor index, and coordinate index */
 960             jnrlistA         = jjnr[jidx];
 961             jnrlistB         = jjnr[jidx+1];
 962             jnrlistC         = jjnr[jidx+2];
 963             jnrlistD         = jjnr[jidx+3];
 964             /* Sign of each element will be negative for non-real atoms.
 965              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 966              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 967              */
 968             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 969
 970             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 971             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 972             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 973
 974             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 975             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 976             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 977             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 978             j_coord_offsetA  = DIM*jnrA;
 979             j_coord_offsetB  = DIM*jnrB;
 980             j_coord_offsetC  = DIM*jnrC;
 981             j_coord_offsetD  = DIM*jnrD;
 982
 983             /* load j atom coordinates */
 984             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 985                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 986                                                  &jx0,&jy0,&jz0);
 987
 988             /* Calculate displacement vector */
 989             dx00             = _mm256_sub_pd(ix0,jx0);
 990             dy00             = _mm256_sub_pd(iy0,jy0);
 991             dz00             = _mm256_sub_pd(iz0,jz0);
 992             dx10             = _mm256_sub_pd(ix1,jx0);
 993             dy10             = _mm256_sub_pd(iy1,jy0);
 994             dz10             = _mm256_sub_pd(iz1,jz0);
 995             dx20             = _mm256_sub_pd(ix2,jx0);
 996             dy20             = _mm256_sub_pd(iy2,jy0);
 997             dz20             = _mm256_sub_pd(iz2,jz0);
 998
 999             /* Calculate squared distance and things based on it */
1000             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1001             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1002             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1003
1004             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1005             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1006             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1007
1008             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1009             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1010             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1011
1012             /* Load parameters for j particles */
1013             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1014                                                                  charge+jnrC+0,charge+jnrD+0);
1015             vdwjidx0A        = 2*vdwtype[jnrA+0];
1016             vdwjidx0B        = 2*vdwtype[jnrB+0];
1017             vdwjidx0C        = 2*vdwtype[jnrC+0];
1018             vdwjidx0D        = 2*vdwtype[jnrD+0];
1019
1020             fjx0             = _mm256_setzero_pd();
1021             fjy0             = _mm256_setzero_pd();
1022             fjz0             = _mm256_setzero_pd();
1023
1024             /**************************
1025              * CALCULATE INTERACTIONS *
1026              **************************/
1027
1028             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1029             {
1030
1031             /* Compute parameters for interactions between i and j atoms */
1032             qq00             = _mm256_mul_pd(iq0,jq0);
1033             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1034                                             vdwioffsetptr0+vdwjidx0B,
1035                                             vdwioffsetptr0+vdwjidx0C,
1036                                             vdwioffsetptr0+vdwjidx0D,
1037                                             &c6_00,&c12_00);
1038
1039             /* REACTION-FIELD ELECTROSTATICS */
1040             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
1041
1042             /* LENNARD-JONES DISPERSION/REPULSION */
1043
1044             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1045             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1046
1047             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1048
1049             fscal            = _mm256_add_pd(felec,fvdw);
1050
1051             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1052
1053             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1054
1055             /* Calculate temporary vectorial force */
1056             tx               = _mm256_mul_pd(fscal,dx00);
1057             ty               = _mm256_mul_pd(fscal,dy00);
1058             tz               = _mm256_mul_pd(fscal,dz00);
1059
1060             /* Update vectorial force */
1061             fix0             = _mm256_add_pd(fix0,tx);
1062             fiy0             = _mm256_add_pd(fiy0,ty);
1063             fiz0             = _mm256_add_pd(fiz0,tz);
1064
1065             fjx0             = _mm256_add_pd(fjx0,tx);
1066             fjy0             = _mm256_add_pd(fjy0,ty);
1067             fjz0             = _mm256_add_pd(fjz0,tz);
1068
1069             }
1070
1071             /**************************
1072              * CALCULATE INTERACTIONS *
1073              **************************/
1074
1075             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1076             {
1077
1078             /* Compute parameters for interactions between i and j atoms */
1079             qq10             = _mm256_mul_pd(iq1,jq0);
1080
1081             /* REACTION-FIELD ELECTROSTATICS */
1082             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1083
1084             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1085
1086             fscal            = felec;
1087
1088             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1089
1090             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1091
1092             /* Calculate temporary vectorial force */
1093             tx               = _mm256_mul_pd(fscal,dx10);
1094             ty               = _mm256_mul_pd(fscal,dy10);
1095             tz               = _mm256_mul_pd(fscal,dz10);
1096
1097             /* Update vectorial force */
1098             fix1             = _mm256_add_pd(fix1,tx);
1099             fiy1             = _mm256_add_pd(fiy1,ty);
1100             fiz1             = _mm256_add_pd(fiz1,tz);
1101
1102             fjx0             = _mm256_add_pd(fjx0,tx);
1103             fjy0             = _mm256_add_pd(fjy0,ty);
1104             fjz0             = _mm256_add_pd(fjz0,tz);
1105
1106             }
1107
1108             /**************************
1109              * CALCULATE INTERACTIONS *
1110              **************************/
1111
1112             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1113             {
1114
1115             /* Compute parameters for interactions between i and j atoms */
1116             qq20             = _mm256_mul_pd(iq2,jq0);
1117
1118             /* REACTION-FIELD ELECTROSTATICS */
1119             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1120
1121             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1122
1123             fscal            = felec;
1124
1125             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1126
1127             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1128
1129             /* Calculate temporary vectorial force */
1130             tx               = _mm256_mul_pd(fscal,dx20);
1131             ty               = _mm256_mul_pd(fscal,dy20);
1132             tz               = _mm256_mul_pd(fscal,dz20);
1133
1134             /* Update vectorial force */
1135             fix2             = _mm256_add_pd(fix2,tx);
1136             fiy2             = _mm256_add_pd(fiy2,ty);
1137             fiz2             = _mm256_add_pd(fiz2,tz);
1138
1139             fjx0             = _mm256_add_pd(fjx0,tx);
1140             fjy0             = _mm256_add_pd(fjy0,ty);
1141             fjz0             = _mm256_add_pd(fjz0,tz);
1142
1143             }
1144
1145             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1146             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1147             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1148             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1149
1150             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1151
1152             /* Inner loop uses 100 flops */
1153         }
1154
1155         /* End of innermost loop */
1156
1157         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1158                                                  f+i_coord_offset,fshift+i_shift_offset);
1159
1160         /* Increment number of inner iterations */
1161         inneriter                  += j_index_end - j_index_start;
1162
1163         /* Outer loop uses 18 flops */
1164     }
1165
1166     /* Increment number of outer iterations */
1167     outeriter        += nri;
1168
1169     /* Update outer/inner flops */
1170
1171     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*100);
1172 }