src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRF_VdwLJ_GeomW3P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, 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_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW3P1_VF_avx_256_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRF_VdwLJ_GeomW3P1_VF_avx_256_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  93     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  96     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  97     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 104     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 105     __m256           dummy_mask,cutoff_mask;
 106     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 107     __m256           one     = _mm256_set1_ps(1.0);
 108     __m256           two     = _mm256_set1_ps(2.0);
 109     x                = xx[0];
 110     f                = ff[0];
 111
 112     nri              = nlist->nri;
 113     iinr             = nlist->iinr;
 114     jindex           = nlist->jindex;
 115     jjnr             = nlist->jjnr;
 116     shiftidx         = nlist->shift;
 117     gid              = nlist->gid;
 118     shiftvec         = fr->shift_vec[0];
 119     fshift           = fr->fshift[0];
 120     facel            = _mm256_set1_ps(fr->epsfac);
 121     charge           = mdatoms->chargeA;
 122     krf              = _mm256_set1_ps(fr->ic->k_rf);
 123     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 124     crf              = _mm256_set1_ps(fr->ic->c_rf);
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     /* Setup water-specific parameters */
 130     inr              = nlist->iinr[0];
 131     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 132     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 133     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 134     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 135
 136     /* Avoid stupid compiler warnings */
 137     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 138     j_coord_offsetA = 0;
 139     j_coord_offsetB = 0;
 140     j_coord_offsetC = 0;
 141     j_coord_offsetD = 0;
 142     j_coord_offsetE = 0;
 143     j_coord_offsetF = 0;
 144     j_coord_offsetG = 0;
 145     j_coord_offsetH = 0;
 146
 147     outeriter        = 0;
 148     inneriter        = 0;
 149
 150     for(iidx=0;iidx<4*DIM;iidx++)
 151     {
 152         scratch[iidx] = 0.0;
 153     }
 154
 155     /* Start outer loop over neighborlists */
 156     for(iidx=0; iidx<nri; iidx++)
 157     {
 158         /* Load shift vector for this list */
 159         i_shift_offset   = DIM*shiftidx[iidx];
 160
 161         /* Load limits for loop over neighbors */
 162         j_index_start    = jindex[iidx];
 163         j_index_end      = jindex[iidx+1];
 164
 165         /* Get outer coordinate index */
 166         inr              = iinr[iidx];
 167         i_coord_offset   = DIM*inr;
 168
 169         /* Load i particle coords and add shift vector */
 170         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 171                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 172
 173         fix0             = _mm256_setzero_ps();
 174         fiy0             = _mm256_setzero_ps();
 175         fiz0             = _mm256_setzero_ps();
 176         fix1             = _mm256_setzero_ps();
 177         fiy1             = _mm256_setzero_ps();
 178         fiz1             = _mm256_setzero_ps();
 179         fix2             = _mm256_setzero_ps();
 180         fiy2             = _mm256_setzero_ps();
 181         fiz2             = _mm256_setzero_ps();
 182
 183         /* Reset potential sums */
 184         velecsum         = _mm256_setzero_ps();
 185         vvdwsum          = _mm256_setzero_ps();
 186
 187         /* Start inner kernel loop */
 188         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 189         {
 190
 191             /* Get j neighbor index, and coordinate index */
 192             jnrA             = jjnr[jidx];
 193             jnrB             = jjnr[jidx+1];
 194             jnrC             = jjnr[jidx+2];
 195             jnrD             = jjnr[jidx+3];
 196             jnrE             = jjnr[jidx+4];
 197             jnrF             = jjnr[jidx+5];
 198             jnrG             = jjnr[jidx+6];
 199             jnrH             = jjnr[jidx+7];
 200             j_coord_offsetA  = DIM*jnrA;
 201             j_coord_offsetB  = DIM*jnrB;
 202             j_coord_offsetC  = DIM*jnrC;
 203             j_coord_offsetD  = DIM*jnrD;
 204             j_coord_offsetE  = DIM*jnrE;
 205             j_coord_offsetF  = DIM*jnrF;
 206             j_coord_offsetG  = DIM*jnrG;
 207             j_coord_offsetH  = DIM*jnrH;
 208
 209             /* load j atom coordinates */
 210             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 211                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 212                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 213                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 214                                                  &jx0,&jy0,&jz0);
 215
 216             /* Calculate displacement vector */
 217             dx00             = _mm256_sub_ps(ix0,jx0);
 218             dy00             = _mm256_sub_ps(iy0,jy0);
 219             dz00             = _mm256_sub_ps(iz0,jz0);
 220             dx10             = _mm256_sub_ps(ix1,jx0);
 221             dy10             = _mm256_sub_ps(iy1,jy0);
 222             dz10             = _mm256_sub_ps(iz1,jz0);
 223             dx20             = _mm256_sub_ps(ix2,jx0);
 224             dy20             = _mm256_sub_ps(iy2,jy0);
 225             dz20             = _mm256_sub_ps(iz2,jz0);
 226
 227             /* Calculate squared distance and things based on it */
 228             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 229             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 230             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 231
 232             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 233             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 234             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 235
 236             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 237             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 238             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 239
 240             /* Load parameters for j particles */
 241             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 242                                                                  charge+jnrC+0,charge+jnrD+0,
 243                                                                  charge+jnrE+0,charge+jnrF+0,
 244                                                                  charge+jnrG+0,charge+jnrH+0);
 245             vdwjidx0A        = 2*vdwtype[jnrA+0];
 246             vdwjidx0B        = 2*vdwtype[jnrB+0];
 247             vdwjidx0C        = 2*vdwtype[jnrC+0];
 248             vdwjidx0D        = 2*vdwtype[jnrD+0];
 249             vdwjidx0E        = 2*vdwtype[jnrE+0];
 250             vdwjidx0F        = 2*vdwtype[jnrF+0];
 251             vdwjidx0G        = 2*vdwtype[jnrG+0];
 252             vdwjidx0H        = 2*vdwtype[jnrH+0];
 253
 254             fjx0             = _mm256_setzero_ps();
 255             fjy0             = _mm256_setzero_ps();
 256             fjz0             = _mm256_setzero_ps();
 257
 258             /**************************
 259              * CALCULATE INTERACTIONS *
 260              **************************/
 261
 262             /* Compute parameters for interactions between i and j atoms */
 263             qq00             = _mm256_mul_ps(iq0,jq0);
 264             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 265                                             vdwioffsetptr0+vdwjidx0B,
 266                                             vdwioffsetptr0+vdwjidx0C,
 267                                             vdwioffsetptr0+vdwjidx0D,
 268                                             vdwioffsetptr0+vdwjidx0E,
 269                                             vdwioffsetptr0+vdwjidx0F,
 270                                             vdwioffsetptr0+vdwjidx0G,
 271                                             vdwioffsetptr0+vdwjidx0H,
 272                                             &c6_00,&c12_00);
 273
 274             /* REACTION-FIELD ELECTROSTATICS */
 275             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 276             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 277
 278             /* LENNARD-JONES DISPERSION/REPULSION */
 279
 280             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 281             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 282             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 283             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 284             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 285
 286             /* Update potential sum for this i atom from the interaction with this j atom. */
 287             velecsum         = _mm256_add_ps(velecsum,velec);
 288             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 289
 290             fscal            = _mm256_add_ps(felec,fvdw);
 291
 292             /* Calculate temporary vectorial force */
 293             tx               = _mm256_mul_ps(fscal,dx00);
 294             ty               = _mm256_mul_ps(fscal,dy00);
 295             tz               = _mm256_mul_ps(fscal,dz00);
 296
 297             /* Update vectorial force */
 298             fix0             = _mm256_add_ps(fix0,tx);
 299             fiy0             = _mm256_add_ps(fiy0,ty);
 300             fiz0             = _mm256_add_ps(fiz0,tz);
 301
 302             fjx0             = _mm256_add_ps(fjx0,tx);
 303             fjy0             = _mm256_add_ps(fjy0,ty);
 304             fjz0             = _mm256_add_ps(fjz0,tz);
 305
 306             /**************************
 307              * CALCULATE INTERACTIONS *
 308              **************************/
 309
 310             /* Compute parameters for interactions between i and j atoms */
 311             qq10             = _mm256_mul_ps(iq1,jq0);
 312
 313             /* REACTION-FIELD ELECTROSTATICS */
 314             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 315             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 316
 317             /* Update potential sum for this i atom from the interaction with this j atom. */
 318             velecsum         = _mm256_add_ps(velecsum,velec);
 319
 320             fscal            = felec;
 321
 322             /* Calculate temporary vectorial force */
 323             tx               = _mm256_mul_ps(fscal,dx10);
 324             ty               = _mm256_mul_ps(fscal,dy10);
 325             tz               = _mm256_mul_ps(fscal,dz10);
 326
 327             /* Update vectorial force */
 328             fix1             = _mm256_add_ps(fix1,tx);
 329             fiy1             = _mm256_add_ps(fiy1,ty);
 330             fiz1             = _mm256_add_ps(fiz1,tz);
 331
 332             fjx0             = _mm256_add_ps(fjx0,tx);
 333             fjy0             = _mm256_add_ps(fjy0,ty);
 334             fjz0             = _mm256_add_ps(fjz0,tz);
 335
 336             /**************************
 337              * CALCULATE INTERACTIONS *
 338              **************************/
 339
 340             /* Compute parameters for interactions between i and j atoms */
 341             qq20             = _mm256_mul_ps(iq2,jq0);
 342
 343             /* REACTION-FIELD ELECTROSTATICS */
 344             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 345             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 346
 347             /* Update potential sum for this i atom from the interaction with this j atom. */
 348             velecsum         = _mm256_add_ps(velecsum,velec);
 349
 350             fscal            = felec;
 351
 352             /* Calculate temporary vectorial force */
 353             tx               = _mm256_mul_ps(fscal,dx20);
 354             ty               = _mm256_mul_ps(fscal,dy20);
 355             tz               = _mm256_mul_ps(fscal,dz20);
 356
 357             /* Update vectorial force */
 358             fix2             = _mm256_add_ps(fix2,tx);
 359             fiy2             = _mm256_add_ps(fiy2,ty);
 360             fiz2             = _mm256_add_ps(fiz2,tz);
 361
 362             fjx0             = _mm256_add_ps(fjx0,tx);
 363             fjy0             = _mm256_add_ps(fjy0,ty);
 364             fjz0             = _mm256_add_ps(fjz0,tz);
 365
 366             fjptrA             = f+j_coord_offsetA;
 367             fjptrB             = f+j_coord_offsetB;
 368             fjptrC             = f+j_coord_offsetC;
 369             fjptrD             = f+j_coord_offsetD;
 370             fjptrE             = f+j_coord_offsetE;
 371             fjptrF             = f+j_coord_offsetF;
 372             fjptrG             = f+j_coord_offsetG;
 373             fjptrH             = f+j_coord_offsetH;
 374
 375             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 376
 377             /* Inner loop uses 111 flops */
 378         }
 379
 380         if(jidx<j_index_end)
 381         {
 382
 383             /* Get j neighbor index, and coordinate index */
 384             jnrlistA         = jjnr[jidx];
 385             jnrlistB         = jjnr[jidx+1];
 386             jnrlistC         = jjnr[jidx+2];
 387             jnrlistD         = jjnr[jidx+3];
 388             jnrlistE         = jjnr[jidx+4];
 389             jnrlistF         = jjnr[jidx+5];
 390             jnrlistG         = jjnr[jidx+6];
 391             jnrlistH         = jjnr[jidx+7];
 392             /* Sign of each element will be negative for non-real atoms.
 393              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 394              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 395              */
 396             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 397                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 398
 399             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 400             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 401             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 402             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 403             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 404             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 405             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 406             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 407             j_coord_offsetA  = DIM*jnrA;
 408             j_coord_offsetB  = DIM*jnrB;
 409             j_coord_offsetC  = DIM*jnrC;
 410             j_coord_offsetD  = DIM*jnrD;
 411             j_coord_offsetE  = DIM*jnrE;
 412             j_coord_offsetF  = DIM*jnrF;
 413             j_coord_offsetG  = DIM*jnrG;
 414             j_coord_offsetH  = DIM*jnrH;
 415
 416             /* load j atom coordinates */
 417             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 418                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 419                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 420                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 421                                                  &jx0,&jy0,&jz0);
 422
 423             /* Calculate displacement vector */
 424             dx00             = _mm256_sub_ps(ix0,jx0);
 425             dy00             = _mm256_sub_ps(iy0,jy0);
 426             dz00             = _mm256_sub_ps(iz0,jz0);
 427             dx10             = _mm256_sub_ps(ix1,jx0);
 428             dy10             = _mm256_sub_ps(iy1,jy0);
 429             dz10             = _mm256_sub_ps(iz1,jz0);
 430             dx20             = _mm256_sub_ps(ix2,jx0);
 431             dy20             = _mm256_sub_ps(iy2,jy0);
 432             dz20             = _mm256_sub_ps(iz2,jz0);
 433
 434             /* Calculate squared distance and things based on it */
 435             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 436             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 437             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 438
 439             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 440             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 441             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 442
 443             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 444             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 445             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 446
 447             /* Load parameters for j particles */
 448             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 449                                                                  charge+jnrC+0,charge+jnrD+0,
 450                                                                  charge+jnrE+0,charge+jnrF+0,
 451                                                                  charge+jnrG+0,charge+jnrH+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             vdwjidx0E        = 2*vdwtype[jnrE+0];
 457             vdwjidx0F        = 2*vdwtype[jnrF+0];
 458             vdwjidx0G        = 2*vdwtype[jnrG+0];
 459             vdwjidx0H        = 2*vdwtype[jnrH+0];
 460
 461             fjx0             = _mm256_setzero_ps();
 462             fjy0             = _mm256_setzero_ps();
 463             fjz0             = _mm256_setzero_ps();
 464
 465             /**************************
 466              * CALCULATE INTERACTIONS *
 467              **************************/
 468
 469             /* Compute parameters for interactions between i and j atoms */
 470             qq00             = _mm256_mul_ps(iq0,jq0);
 471             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 472                                             vdwioffsetptr0+vdwjidx0B,
 473                                             vdwioffsetptr0+vdwjidx0C,
 474                                             vdwioffsetptr0+vdwjidx0D,
 475                                             vdwioffsetptr0+vdwjidx0E,
 476                                             vdwioffsetptr0+vdwjidx0F,
 477                                             vdwioffsetptr0+vdwjidx0G,
 478                                             vdwioffsetptr0+vdwjidx0H,
 479                                             &c6_00,&c12_00);
 480
 481             /* REACTION-FIELD ELECTROSTATICS */
 482             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 483             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 484
 485             /* LENNARD-JONES DISPERSION/REPULSION */
 486
 487             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 488             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 489             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 490             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 491             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 492
 493             /* Update potential sum for this i atom from the interaction with this j atom. */
 494             velec            = _mm256_andnot_ps(dummy_mask,velec);
 495             velecsum         = _mm256_add_ps(velecsum,velec);
 496             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 497             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 498
 499             fscal            = _mm256_add_ps(felec,fvdw);
 500
 501             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 502
 503             /* Calculate temporary vectorial force */
 504             tx               = _mm256_mul_ps(fscal,dx00);
 505             ty               = _mm256_mul_ps(fscal,dy00);
 506             tz               = _mm256_mul_ps(fscal,dz00);
 507
 508             /* Update vectorial force */
 509             fix0             = _mm256_add_ps(fix0,tx);
 510             fiy0             = _mm256_add_ps(fiy0,ty);
 511             fiz0             = _mm256_add_ps(fiz0,tz);
 512
 513             fjx0             = _mm256_add_ps(fjx0,tx);
 514             fjy0             = _mm256_add_ps(fjy0,ty);
 515             fjz0             = _mm256_add_ps(fjz0,tz);
 516
 517             /**************************
 518              * CALCULATE INTERACTIONS *
 519              **************************/
 520
 521             /* Compute parameters for interactions between i and j atoms */
 522             qq10             = _mm256_mul_ps(iq1,jq0);
 523
 524             /* REACTION-FIELD ELECTROSTATICS */
 525             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 526             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 527
 528             /* Update potential sum for this i atom from the interaction with this j atom. */
 529             velec            = _mm256_andnot_ps(dummy_mask,velec);
 530             velecsum         = _mm256_add_ps(velecsum,velec);
 531
 532             fscal            = felec;
 533
 534             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 535
 536             /* Calculate temporary vectorial force */
 537             tx               = _mm256_mul_ps(fscal,dx10);
 538             ty               = _mm256_mul_ps(fscal,dy10);
 539             tz               = _mm256_mul_ps(fscal,dz10);
 540
 541             /* Update vectorial force */
 542             fix1             = _mm256_add_ps(fix1,tx);
 543             fiy1             = _mm256_add_ps(fiy1,ty);
 544             fiz1             = _mm256_add_ps(fiz1,tz);
 545
 546             fjx0             = _mm256_add_ps(fjx0,tx);
 547             fjy0             = _mm256_add_ps(fjy0,ty);
 548             fjz0             = _mm256_add_ps(fjz0,tz);
 549
 550             /**************************
 551              * CALCULATE INTERACTIONS *
 552              **************************/
 553
 554             /* Compute parameters for interactions between i and j atoms */
 555             qq20             = _mm256_mul_ps(iq2,jq0);
 556
 557             /* REACTION-FIELD ELECTROSTATICS */
 558             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 559             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 560
 561             /* Update potential sum for this i atom from the interaction with this j atom. */
 562             velec            = _mm256_andnot_ps(dummy_mask,velec);
 563             velecsum         = _mm256_add_ps(velecsum,velec);
 564
 565             fscal            = felec;
 566
 567             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 568
 569             /* Calculate temporary vectorial force */
 570             tx               = _mm256_mul_ps(fscal,dx20);
 571             ty               = _mm256_mul_ps(fscal,dy20);
 572             tz               = _mm256_mul_ps(fscal,dz20);
 573
 574             /* Update vectorial force */
 575             fix2             = _mm256_add_ps(fix2,tx);
 576             fiy2             = _mm256_add_ps(fiy2,ty);
 577             fiz2             = _mm256_add_ps(fiz2,tz);
 578
 579             fjx0             = _mm256_add_ps(fjx0,tx);
 580             fjy0             = _mm256_add_ps(fjy0,ty);
 581             fjz0             = _mm256_add_ps(fjz0,tz);
 582
 583             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 584             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 585             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 586             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 587             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 588             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 589             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 590             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 591
 592             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 593
 594             /* Inner loop uses 111 flops */
 595         }
 596
 597         /* End of innermost loop */
 598
 599         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 600                                                  f+i_coord_offset,fshift+i_shift_offset);
 601
 602         ggid                        = gid[iidx];
 603         /* Update potential energies */
 604         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 605         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 606
 607         /* Increment number of inner iterations */
 608         inneriter                  += j_index_end - j_index_start;
 609
 610         /* Outer loop uses 20 flops */
 611     }
 612
 613     /* Increment number of outer iterations */
 614     outeriter        += nri;
 615
 616     /* Update outer/inner flops */
 617
 618     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*111);
 619 }
 620 /*
 621  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW3P1_F_avx_256_single
 622  * Electrostatics interaction: ReactionField
 623  * VdW interaction:            LennardJones
 624  * Geometry:                   Water3-Particle
 625  * Calculate force/pot:        Force
 626  */
 627 void
 628 nb_kernel_ElecRF_VdwLJ_GeomW3P1_F_avx_256_single
 629                     (t_nblist                    * gmx_restrict       nlist,
 630                      rvec                        * gmx_restrict          xx,
 631                      rvec                        * gmx_restrict          ff,
 632                      t_forcerec                  * gmx_restrict          fr,
 633                      t_mdatoms                   * gmx_restrict     mdatoms,
 634                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 635                      t_nrnb                      * gmx_restrict        nrnb)
 636 {
 637     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 638      * just 0 for non-waters.
 639      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 640      * jnr indices corresponding to data put in the four positions in the SIMD register.
 641      */
 642     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 643     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 644     int              jnrA,jnrB,jnrC,jnrD;
 645     int              jnrE,jnrF,jnrG,jnrH;
 646     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 647     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 648     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 649     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 650     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 651     real             rcutoff_scalar;
 652     real             *shiftvec,*fshift,*x,*f;
 653     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 654     real             scratch[4*DIM];
 655     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 656     real *           vdwioffsetptr0;
 657     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 658     real *           vdwioffsetptr1;
 659     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 660     real *           vdwioffsetptr2;
 661     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 662     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 663     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 664     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 665     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 666     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 667     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 668     real             *charge;
 669     int              nvdwtype;
 670     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 671     int              *vdwtype;
 672     real             *vdwparam;
 673     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 674     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 675     __m256           dummy_mask,cutoff_mask;
 676     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 677     __m256           one     = _mm256_set1_ps(1.0);
 678     __m256           two     = _mm256_set1_ps(2.0);
 679     x                = xx[0];
 680     f                = ff[0];
 681
 682     nri              = nlist->nri;
 683     iinr             = nlist->iinr;
 684     jindex           = nlist->jindex;
 685     jjnr             = nlist->jjnr;
 686     shiftidx         = nlist->shift;
 687     gid              = nlist->gid;
 688     shiftvec         = fr->shift_vec[0];
 689     fshift           = fr->fshift[0];
 690     facel            = _mm256_set1_ps(fr->epsfac);
 691     charge           = mdatoms->chargeA;
 692     krf              = _mm256_set1_ps(fr->ic->k_rf);
 693     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 694     crf              = _mm256_set1_ps(fr->ic->c_rf);
 695     nvdwtype         = fr->ntype;
 696     vdwparam         = fr->nbfp;
 697     vdwtype          = mdatoms->typeA;
 698
 699     /* Setup water-specific parameters */
 700     inr              = nlist->iinr[0];
 701     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 702     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 703     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 704     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 705
 706     /* Avoid stupid compiler warnings */
 707     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 708     j_coord_offsetA = 0;
 709     j_coord_offsetB = 0;
 710     j_coord_offsetC = 0;
 711     j_coord_offsetD = 0;
 712     j_coord_offsetE = 0;
 713     j_coord_offsetF = 0;
 714     j_coord_offsetG = 0;
 715     j_coord_offsetH = 0;
 716
 717     outeriter        = 0;
 718     inneriter        = 0;
 719
 720     for(iidx=0;iidx<4*DIM;iidx++)
 721     {
 722         scratch[iidx] = 0.0;
 723     }
 724
 725     /* Start outer loop over neighborlists */
 726     for(iidx=0; iidx<nri; iidx++)
 727     {
 728         /* Load shift vector for this list */
 729         i_shift_offset   = DIM*shiftidx[iidx];
 730
 731         /* Load limits for loop over neighbors */
 732         j_index_start    = jindex[iidx];
 733         j_index_end      = jindex[iidx+1];
 734
 735         /* Get outer coordinate index */
 736         inr              = iinr[iidx];
 737         i_coord_offset   = DIM*inr;
 738
 739         /* Load i particle coords and add shift vector */
 740         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 741                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 742
 743         fix0             = _mm256_setzero_ps();
 744         fiy0             = _mm256_setzero_ps();
 745         fiz0             = _mm256_setzero_ps();
 746         fix1             = _mm256_setzero_ps();
 747         fiy1             = _mm256_setzero_ps();
 748         fiz1             = _mm256_setzero_ps();
 749         fix2             = _mm256_setzero_ps();
 750         fiy2             = _mm256_setzero_ps();
 751         fiz2             = _mm256_setzero_ps();
 752
 753         /* Start inner kernel loop */
 754         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 755         {
 756
 757             /* Get j neighbor index, and coordinate index */
 758             jnrA             = jjnr[jidx];
 759             jnrB             = jjnr[jidx+1];
 760             jnrC             = jjnr[jidx+2];
 761             jnrD             = jjnr[jidx+3];
 762             jnrE             = jjnr[jidx+4];
 763             jnrF             = jjnr[jidx+5];
 764             jnrG             = jjnr[jidx+6];
 765             jnrH             = jjnr[jidx+7];
 766             j_coord_offsetA  = DIM*jnrA;
 767             j_coord_offsetB  = DIM*jnrB;
 768             j_coord_offsetC  = DIM*jnrC;
 769             j_coord_offsetD  = DIM*jnrD;
 770             j_coord_offsetE  = DIM*jnrE;
 771             j_coord_offsetF  = DIM*jnrF;
 772             j_coord_offsetG  = DIM*jnrG;
 773             j_coord_offsetH  = DIM*jnrH;
 774
 775             /* load j atom coordinates */
 776             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 777                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 778                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 779                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 780                                                  &jx0,&jy0,&jz0);
 781
 782             /* Calculate displacement vector */
 783             dx00             = _mm256_sub_ps(ix0,jx0);
 784             dy00             = _mm256_sub_ps(iy0,jy0);
 785             dz00             = _mm256_sub_ps(iz0,jz0);
 786             dx10             = _mm256_sub_ps(ix1,jx0);
 787             dy10             = _mm256_sub_ps(iy1,jy0);
 788             dz10             = _mm256_sub_ps(iz1,jz0);
 789             dx20             = _mm256_sub_ps(ix2,jx0);
 790             dy20             = _mm256_sub_ps(iy2,jy0);
 791             dz20             = _mm256_sub_ps(iz2,jz0);
 792
 793             /* Calculate squared distance and things based on it */
 794             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 795             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 796             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 797
 798             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 799             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 800             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 801
 802             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 803             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 804             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 805
 806             /* Load parameters for j particles */
 807             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 808                                                                  charge+jnrC+0,charge+jnrD+0,
 809                                                                  charge+jnrE+0,charge+jnrF+0,
 810                                                                  charge+jnrG+0,charge+jnrH+0);
 811             vdwjidx0A        = 2*vdwtype[jnrA+0];
 812             vdwjidx0B        = 2*vdwtype[jnrB+0];
 813             vdwjidx0C        = 2*vdwtype[jnrC+0];
 814             vdwjidx0D        = 2*vdwtype[jnrD+0];
 815             vdwjidx0E        = 2*vdwtype[jnrE+0];
 816             vdwjidx0F        = 2*vdwtype[jnrF+0];
 817             vdwjidx0G        = 2*vdwtype[jnrG+0];
 818             vdwjidx0H        = 2*vdwtype[jnrH+0];
 819
 820             fjx0             = _mm256_setzero_ps();
 821             fjy0             = _mm256_setzero_ps();
 822             fjz0             = _mm256_setzero_ps();
 823
 824             /**************************
 825              * CALCULATE INTERACTIONS *
 826              **************************/
 827
 828             /* Compute parameters for interactions between i and j atoms */
 829             qq00             = _mm256_mul_ps(iq0,jq0);
 830             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 831                                             vdwioffsetptr0+vdwjidx0B,
 832                                             vdwioffsetptr0+vdwjidx0C,
 833                                             vdwioffsetptr0+vdwjidx0D,
 834                                             vdwioffsetptr0+vdwjidx0E,
 835                                             vdwioffsetptr0+vdwjidx0F,
 836                                             vdwioffsetptr0+vdwjidx0G,
 837                                             vdwioffsetptr0+vdwjidx0H,
 838                                             &c6_00,&c12_00);
 839
 840             /* REACTION-FIELD ELECTROSTATICS */
 841             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 842
 843             /* LENNARD-JONES DISPERSION/REPULSION */
 844
 845             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 846             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 847
 848             fscal            = _mm256_add_ps(felec,fvdw);
 849
 850             /* Calculate temporary vectorial force */
 851             tx               = _mm256_mul_ps(fscal,dx00);
 852             ty               = _mm256_mul_ps(fscal,dy00);
 853             tz               = _mm256_mul_ps(fscal,dz00);
 854
 855             /* Update vectorial force */
 856             fix0             = _mm256_add_ps(fix0,tx);
 857             fiy0             = _mm256_add_ps(fiy0,ty);
 858             fiz0             = _mm256_add_ps(fiz0,tz);
 859
 860             fjx0             = _mm256_add_ps(fjx0,tx);
 861             fjy0             = _mm256_add_ps(fjy0,ty);
 862             fjz0             = _mm256_add_ps(fjz0,tz);
 863
 864             /**************************
 865              * CALCULATE INTERACTIONS *
 866              **************************/
 867
 868             /* Compute parameters for interactions between i and j atoms */
 869             qq10             = _mm256_mul_ps(iq1,jq0);
 870
 871             /* REACTION-FIELD ELECTROSTATICS */
 872             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 873
 874             fscal            = felec;
 875
 876             /* Calculate temporary vectorial force */
 877             tx               = _mm256_mul_ps(fscal,dx10);
 878             ty               = _mm256_mul_ps(fscal,dy10);
 879             tz               = _mm256_mul_ps(fscal,dz10);
 880
 881             /* Update vectorial force */
 882             fix1             = _mm256_add_ps(fix1,tx);
 883             fiy1             = _mm256_add_ps(fiy1,ty);
 884             fiz1             = _mm256_add_ps(fiz1,tz);
 885
 886             fjx0             = _mm256_add_ps(fjx0,tx);
 887             fjy0             = _mm256_add_ps(fjy0,ty);
 888             fjz0             = _mm256_add_ps(fjz0,tz);
 889
 890             /**************************
 891              * CALCULATE INTERACTIONS *
 892              **************************/
 893
 894             /* Compute parameters for interactions between i and j atoms */
 895             qq20             = _mm256_mul_ps(iq2,jq0);
 896
 897             /* REACTION-FIELD ELECTROSTATICS */
 898             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 899
 900             fscal            = felec;
 901
 902             /* Calculate temporary vectorial force */
 903             tx               = _mm256_mul_ps(fscal,dx20);
 904             ty               = _mm256_mul_ps(fscal,dy20);
 905             tz               = _mm256_mul_ps(fscal,dz20);
 906
 907             /* Update vectorial force */
 908             fix2             = _mm256_add_ps(fix2,tx);
 909             fiy2             = _mm256_add_ps(fiy2,ty);
 910             fiz2             = _mm256_add_ps(fiz2,tz);
 911
 912             fjx0             = _mm256_add_ps(fjx0,tx);
 913             fjy0             = _mm256_add_ps(fjy0,ty);
 914             fjz0             = _mm256_add_ps(fjz0,tz);
 915
 916             fjptrA             = f+j_coord_offsetA;
 917             fjptrB             = f+j_coord_offsetB;
 918             fjptrC             = f+j_coord_offsetC;
 919             fjptrD             = f+j_coord_offsetD;
 920             fjptrE             = f+j_coord_offsetE;
 921             fjptrF             = f+j_coord_offsetF;
 922             fjptrG             = f+j_coord_offsetG;
 923             fjptrH             = f+j_coord_offsetH;
 924
 925             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 926
 927             /* Inner loop uses 91 flops */
 928         }
 929
 930         if(jidx<j_index_end)
 931         {
 932
 933             /* Get j neighbor index, and coordinate index */
 934             jnrlistA         = jjnr[jidx];
 935             jnrlistB         = jjnr[jidx+1];
 936             jnrlistC         = jjnr[jidx+2];
 937             jnrlistD         = jjnr[jidx+3];
 938             jnrlistE         = jjnr[jidx+4];
 939             jnrlistF         = jjnr[jidx+5];
 940             jnrlistG         = jjnr[jidx+6];
 941             jnrlistH         = jjnr[jidx+7];
 942             /* Sign of each element will be negative for non-real atoms.
 943              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 944              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 945              */
 946             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 947                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 948
 949             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 950             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 951             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 952             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 953             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 954             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 955             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 956             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 957             j_coord_offsetA  = DIM*jnrA;
 958             j_coord_offsetB  = DIM*jnrB;
 959             j_coord_offsetC  = DIM*jnrC;
 960             j_coord_offsetD  = DIM*jnrD;
 961             j_coord_offsetE  = DIM*jnrE;
 962             j_coord_offsetF  = DIM*jnrF;
 963             j_coord_offsetG  = DIM*jnrG;
 964             j_coord_offsetH  = DIM*jnrH;
 965
 966             /* load j atom coordinates */
 967             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 968                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 969                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 970                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 971                                                  &jx0,&jy0,&jz0);
 972
 973             /* Calculate displacement vector */
 974             dx00             = _mm256_sub_ps(ix0,jx0);
 975             dy00             = _mm256_sub_ps(iy0,jy0);
 976             dz00             = _mm256_sub_ps(iz0,jz0);
 977             dx10             = _mm256_sub_ps(ix1,jx0);
 978             dy10             = _mm256_sub_ps(iy1,jy0);
 979             dz10             = _mm256_sub_ps(iz1,jz0);
 980             dx20             = _mm256_sub_ps(ix2,jx0);
 981             dy20             = _mm256_sub_ps(iy2,jy0);
 982             dz20             = _mm256_sub_ps(iz2,jz0);
 983
 984             /* Calculate squared distance and things based on it */
 985             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 986             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 987             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 988
 989             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 990             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 991             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 992
 993             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 994             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 995             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 996
 997             /* Load parameters for j particles */
 998             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 999                                                                  charge+jnrC+0,charge+jnrD+0,
1000                                                                  charge+jnrE+0,charge+jnrF+0,
1001                                                                  charge+jnrG+0,charge+jnrH+0);
1002             vdwjidx0A        = 2*vdwtype[jnrA+0];
1003             vdwjidx0B        = 2*vdwtype[jnrB+0];
1004             vdwjidx0C        = 2*vdwtype[jnrC+0];
1005             vdwjidx0D        = 2*vdwtype[jnrD+0];
1006             vdwjidx0E        = 2*vdwtype[jnrE+0];
1007             vdwjidx0F        = 2*vdwtype[jnrF+0];
1008             vdwjidx0G        = 2*vdwtype[jnrG+0];
1009             vdwjidx0H        = 2*vdwtype[jnrH+0];
1010
1011             fjx0             = _mm256_setzero_ps();
1012             fjy0             = _mm256_setzero_ps();
1013             fjz0             = _mm256_setzero_ps();
1014
1015             /**************************
1016              * CALCULATE INTERACTIONS *
1017              **************************/
1018
1019             /* Compute parameters for interactions between i and j atoms */
1020             qq00             = _mm256_mul_ps(iq0,jq0);
1021             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1022                                             vdwioffsetptr0+vdwjidx0B,
1023                                             vdwioffsetptr0+vdwjidx0C,
1024                                             vdwioffsetptr0+vdwjidx0D,
1025                                             vdwioffsetptr0+vdwjidx0E,
1026                                             vdwioffsetptr0+vdwjidx0F,
1027                                             vdwioffsetptr0+vdwjidx0G,
1028                                             vdwioffsetptr0+vdwjidx0H,
1029                                             &c6_00,&c12_00);
1030
1031             /* REACTION-FIELD ELECTROSTATICS */
1032             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1033
1034             /* LENNARD-JONES DISPERSION/REPULSION */
1035
1036             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1037             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1038
1039             fscal            = _mm256_add_ps(felec,fvdw);
1040
1041             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1042
1043             /* Calculate temporary vectorial force */
1044             tx               = _mm256_mul_ps(fscal,dx00);
1045             ty               = _mm256_mul_ps(fscal,dy00);
1046             tz               = _mm256_mul_ps(fscal,dz00);
1047
1048             /* Update vectorial force */
1049             fix0             = _mm256_add_ps(fix0,tx);
1050             fiy0             = _mm256_add_ps(fiy0,ty);
1051             fiz0             = _mm256_add_ps(fiz0,tz);
1052
1053             fjx0             = _mm256_add_ps(fjx0,tx);
1054             fjy0             = _mm256_add_ps(fjy0,ty);
1055             fjz0             = _mm256_add_ps(fjz0,tz);
1056
1057             /**************************
1058              * CALCULATE INTERACTIONS *
1059              **************************/
1060
1061             /* Compute parameters for interactions between i and j atoms */
1062             qq10             = _mm256_mul_ps(iq1,jq0);
1063
1064             /* REACTION-FIELD ELECTROSTATICS */
1065             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1066
1067             fscal            = felec;
1068
1069             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1070
1071             /* Calculate temporary vectorial force */
1072             tx               = _mm256_mul_ps(fscal,dx10);
1073             ty               = _mm256_mul_ps(fscal,dy10);
1074             tz               = _mm256_mul_ps(fscal,dz10);
1075
1076             /* Update vectorial force */
1077             fix1             = _mm256_add_ps(fix1,tx);
1078             fiy1             = _mm256_add_ps(fiy1,ty);
1079             fiz1             = _mm256_add_ps(fiz1,tz);
1080
1081             fjx0             = _mm256_add_ps(fjx0,tx);
1082             fjy0             = _mm256_add_ps(fjy0,ty);
1083             fjz0             = _mm256_add_ps(fjz0,tz);
1084
1085             /**************************
1086              * CALCULATE INTERACTIONS *
1087              **************************/
1088
1089             /* Compute parameters for interactions between i and j atoms */
1090             qq20             = _mm256_mul_ps(iq2,jq0);
1091
1092             /* REACTION-FIELD ELECTROSTATICS */
1093             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1094
1095             fscal            = felec;
1096
1097             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1098
1099             /* Calculate temporary vectorial force */
1100             tx               = _mm256_mul_ps(fscal,dx20);
1101             ty               = _mm256_mul_ps(fscal,dy20);
1102             tz               = _mm256_mul_ps(fscal,dz20);
1103
1104             /* Update vectorial force */
1105             fix2             = _mm256_add_ps(fix2,tx);
1106             fiy2             = _mm256_add_ps(fiy2,ty);
1107             fiz2             = _mm256_add_ps(fiz2,tz);
1108
1109             fjx0             = _mm256_add_ps(fjx0,tx);
1110             fjy0             = _mm256_add_ps(fjy0,ty);
1111             fjz0             = _mm256_add_ps(fjz0,tz);
1112
1113             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1114             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1115             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1116             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1117             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1118             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1119             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1120             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1121
1122             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1123
1124             /* Inner loop uses 91 flops */
1125         }
1126
1127         /* End of innermost loop */
1128
1129         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1130                                                  f+i_coord_offset,fshift+i_shift_offset);
1131
1132         /* Increment number of inner iterations */
1133         inneriter                  += j_index_end - j_index_start;
1134
1135         /* Outer loop uses 18 flops */
1136     }
1137
1138     /* Increment number of outer iterations */
1139     outeriter        += nri;
1140
1141     /* Update outer/inner flops */
1142
1143     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*91);
1144 }