src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCoul_VdwNone_GeomW3P1_avx_256_single.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW3P1_VF_avx_256_single
  51  * Electrostatics interaction: Coulomb
  52  * VdW interaction:            None
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCoul_VdwNone_GeomW3P1_VF_avx_256_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrE,jnrF,jnrG,jnrH;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  83     real             scratch[4*DIM];
  84     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     real *           vdwioffsetptr0;
  86     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     real *           vdwioffsetptr1;
  88     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     real *           vdwioffsetptr2;
  90     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  92     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     __m256           dummy_mask,cutoff_mask;
  99     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 100     __m256           one     = _mm256_set1_ps(1.0);
 101     __m256           two     = _mm256_set1_ps(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm256_set1_ps(fr->ic->epsfac);
 114     charge           = mdatoms->chargeA;
 115
 116     /* Setup water-specific parameters */
 117     inr              = nlist->iinr[0];
 118     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 119     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 120     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 121
 122     /* Avoid stupid compiler warnings */
 123     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 124     j_coord_offsetA = 0;
 125     j_coord_offsetB = 0;
 126     j_coord_offsetC = 0;
 127     j_coord_offsetD = 0;
 128     j_coord_offsetE = 0;
 129     j_coord_offsetF = 0;
 130     j_coord_offsetG = 0;
 131     j_coord_offsetH = 0;
 132
 133     outeriter        = 0;
 134     inneriter        = 0;
 135
 136     for(iidx=0;iidx<4*DIM;iidx++)
 137     {
 138         scratch[iidx] = 0.0;
 139     }
 140
 141     /* Start outer loop over neighborlists */
 142     for(iidx=0; iidx<nri; iidx++)
 143     {
 144         /* Load shift vector for this list */
 145         i_shift_offset   = DIM*shiftidx[iidx];
 146
 147         /* Load limits for loop over neighbors */
 148         j_index_start    = jindex[iidx];
 149         j_index_end      = jindex[iidx+1];
 150
 151         /* Get outer coordinate index */
 152         inr              = iinr[iidx];
 153         i_coord_offset   = DIM*inr;
 154
 155         /* Load i particle coords and add shift vector */
 156         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 157                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 158
 159         fix0             = _mm256_setzero_ps();
 160         fiy0             = _mm256_setzero_ps();
 161         fiz0             = _mm256_setzero_ps();
 162         fix1             = _mm256_setzero_ps();
 163         fiy1             = _mm256_setzero_ps();
 164         fiz1             = _mm256_setzero_ps();
 165         fix2             = _mm256_setzero_ps();
 166         fiy2             = _mm256_setzero_ps();
 167         fiz2             = _mm256_setzero_ps();
 168
 169         /* Reset potential sums */
 170         velecsum         = _mm256_setzero_ps();
 171
 172         /* Start inner kernel loop */
 173         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 174         {
 175
 176             /* Get j neighbor index, and coordinate index */
 177             jnrA             = jjnr[jidx];
 178             jnrB             = jjnr[jidx+1];
 179             jnrC             = jjnr[jidx+2];
 180             jnrD             = jjnr[jidx+3];
 181             jnrE             = jjnr[jidx+4];
 182             jnrF             = jjnr[jidx+5];
 183             jnrG             = jjnr[jidx+6];
 184             jnrH             = jjnr[jidx+7];
 185             j_coord_offsetA  = DIM*jnrA;
 186             j_coord_offsetB  = DIM*jnrB;
 187             j_coord_offsetC  = DIM*jnrC;
 188             j_coord_offsetD  = DIM*jnrD;
 189             j_coord_offsetE  = DIM*jnrE;
 190             j_coord_offsetF  = DIM*jnrF;
 191             j_coord_offsetG  = DIM*jnrG;
 192             j_coord_offsetH  = DIM*jnrH;
 193
 194             /* load j atom coordinates */
 195             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 196                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 197                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 198                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 199                                                  &jx0,&jy0,&jz0);
 200
 201             /* Calculate displacement vector */
 202             dx00             = _mm256_sub_ps(ix0,jx0);
 203             dy00             = _mm256_sub_ps(iy0,jy0);
 204             dz00             = _mm256_sub_ps(iz0,jz0);
 205             dx10             = _mm256_sub_ps(ix1,jx0);
 206             dy10             = _mm256_sub_ps(iy1,jy0);
 207             dz10             = _mm256_sub_ps(iz1,jz0);
 208             dx20             = _mm256_sub_ps(ix2,jx0);
 209             dy20             = _mm256_sub_ps(iy2,jy0);
 210             dz20             = _mm256_sub_ps(iz2,jz0);
 211
 212             /* Calculate squared distance and things based on it */
 213             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 214             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 215             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 216
 217             rinv00           = avx256_invsqrt_f(rsq00);
 218             rinv10           = avx256_invsqrt_f(rsq10);
 219             rinv20           = avx256_invsqrt_f(rsq20);
 220
 221             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 222             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 223             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 224
 225             /* Load parameters for j particles */
 226             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 227                                                                  charge+jnrC+0,charge+jnrD+0,
 228                                                                  charge+jnrE+0,charge+jnrF+0,
 229                                                                  charge+jnrG+0,charge+jnrH+0);
 230
 231             fjx0             = _mm256_setzero_ps();
 232             fjy0             = _mm256_setzero_ps();
 233             fjz0             = _mm256_setzero_ps();
 234
 235             /**************************
 236              * CALCULATE INTERACTIONS *
 237              **************************/
 238
 239             /* Compute parameters for interactions between i and j atoms */
 240             qq00             = _mm256_mul_ps(iq0,jq0);
 241
 242             /* COULOMB ELECTROSTATICS */
 243             velec            = _mm256_mul_ps(qq00,rinv00);
 244             felec            = _mm256_mul_ps(velec,rinvsq00);
 245
 246             /* Update potential sum for this i atom from the interaction with this j atom. */
 247             velecsum         = _mm256_add_ps(velecsum,velec);
 248
 249             fscal            = felec;
 250
 251             /* Calculate temporary vectorial force */
 252             tx               = _mm256_mul_ps(fscal,dx00);
 253             ty               = _mm256_mul_ps(fscal,dy00);
 254             tz               = _mm256_mul_ps(fscal,dz00);
 255
 256             /* Update vectorial force */
 257             fix0             = _mm256_add_ps(fix0,tx);
 258             fiy0             = _mm256_add_ps(fiy0,ty);
 259             fiz0             = _mm256_add_ps(fiz0,tz);
 260
 261             fjx0             = _mm256_add_ps(fjx0,tx);
 262             fjy0             = _mm256_add_ps(fjy0,ty);
 263             fjz0             = _mm256_add_ps(fjz0,tz);
 264
 265             /**************************
 266              * CALCULATE INTERACTIONS *
 267              **************************/
 268
 269             /* Compute parameters for interactions between i and j atoms */
 270             qq10             = _mm256_mul_ps(iq1,jq0);
 271
 272             /* COULOMB ELECTROSTATICS */
 273             velec            = _mm256_mul_ps(qq10,rinv10);
 274             felec            = _mm256_mul_ps(velec,rinvsq10);
 275
 276             /* Update potential sum for this i atom from the interaction with this j atom. */
 277             velecsum         = _mm256_add_ps(velecsum,velec);
 278
 279             fscal            = felec;
 280
 281             /* Calculate temporary vectorial force */
 282             tx               = _mm256_mul_ps(fscal,dx10);
 283             ty               = _mm256_mul_ps(fscal,dy10);
 284             tz               = _mm256_mul_ps(fscal,dz10);
 285
 286             /* Update vectorial force */
 287             fix1             = _mm256_add_ps(fix1,tx);
 288             fiy1             = _mm256_add_ps(fiy1,ty);
 289             fiz1             = _mm256_add_ps(fiz1,tz);
 290
 291             fjx0             = _mm256_add_ps(fjx0,tx);
 292             fjy0             = _mm256_add_ps(fjy0,ty);
 293             fjz0             = _mm256_add_ps(fjz0,tz);
 294
 295             /**************************
 296              * CALCULATE INTERACTIONS *
 297              **************************/
 298
 299             /* Compute parameters for interactions between i and j atoms */
 300             qq20             = _mm256_mul_ps(iq2,jq0);
 301
 302             /* COULOMB ELECTROSTATICS */
 303             velec            = _mm256_mul_ps(qq20,rinv20);
 304             felec            = _mm256_mul_ps(velec,rinvsq20);
 305
 306             /* Update potential sum for this i atom from the interaction with this j atom. */
 307             velecsum         = _mm256_add_ps(velecsum,velec);
 308
 309             fscal            = felec;
 310
 311             /* Calculate temporary vectorial force */
 312             tx               = _mm256_mul_ps(fscal,dx20);
 313             ty               = _mm256_mul_ps(fscal,dy20);
 314             tz               = _mm256_mul_ps(fscal,dz20);
 315
 316             /* Update vectorial force */
 317             fix2             = _mm256_add_ps(fix2,tx);
 318             fiy2             = _mm256_add_ps(fiy2,ty);
 319             fiz2             = _mm256_add_ps(fiz2,tz);
 320
 321             fjx0             = _mm256_add_ps(fjx0,tx);
 322             fjy0             = _mm256_add_ps(fjy0,ty);
 323             fjz0             = _mm256_add_ps(fjz0,tz);
 324
 325             fjptrA             = f+j_coord_offsetA;
 326             fjptrB             = f+j_coord_offsetB;
 327             fjptrC             = f+j_coord_offsetC;
 328             fjptrD             = f+j_coord_offsetD;
 329             fjptrE             = f+j_coord_offsetE;
 330             fjptrF             = f+j_coord_offsetF;
 331             fjptrG             = f+j_coord_offsetG;
 332             fjptrH             = f+j_coord_offsetH;
 333
 334             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 335
 336             /* Inner loop uses 84 flops */
 337         }
 338
 339         if(jidx<j_index_end)
 340         {
 341
 342             /* Get j neighbor index, and coordinate index */
 343             jnrlistA         = jjnr[jidx];
 344             jnrlistB         = jjnr[jidx+1];
 345             jnrlistC         = jjnr[jidx+2];
 346             jnrlistD         = jjnr[jidx+3];
 347             jnrlistE         = jjnr[jidx+4];
 348             jnrlistF         = jjnr[jidx+5];
 349             jnrlistG         = jjnr[jidx+6];
 350             jnrlistH         = jjnr[jidx+7];
 351             /* Sign of each element will be negative for non-real atoms.
 352              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 353              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 354              */
 355             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 356                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 357
 358             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 359             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 360             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 361             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 362             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 363             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 364             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 365             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 366             j_coord_offsetA  = DIM*jnrA;
 367             j_coord_offsetB  = DIM*jnrB;
 368             j_coord_offsetC  = DIM*jnrC;
 369             j_coord_offsetD  = DIM*jnrD;
 370             j_coord_offsetE  = DIM*jnrE;
 371             j_coord_offsetF  = DIM*jnrF;
 372             j_coord_offsetG  = DIM*jnrG;
 373             j_coord_offsetH  = DIM*jnrH;
 374
 375             /* load j atom coordinates */
 376             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 377                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 378                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 379                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 380                                                  &jx0,&jy0,&jz0);
 381
 382             /* Calculate displacement vector */
 383             dx00             = _mm256_sub_ps(ix0,jx0);
 384             dy00             = _mm256_sub_ps(iy0,jy0);
 385             dz00             = _mm256_sub_ps(iz0,jz0);
 386             dx10             = _mm256_sub_ps(ix1,jx0);
 387             dy10             = _mm256_sub_ps(iy1,jy0);
 388             dz10             = _mm256_sub_ps(iz1,jz0);
 389             dx20             = _mm256_sub_ps(ix2,jx0);
 390             dy20             = _mm256_sub_ps(iy2,jy0);
 391             dz20             = _mm256_sub_ps(iz2,jz0);
 392
 393             /* Calculate squared distance and things based on it */
 394             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 395             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 396             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 397
 398             rinv00           = avx256_invsqrt_f(rsq00);
 399             rinv10           = avx256_invsqrt_f(rsq10);
 400             rinv20           = avx256_invsqrt_f(rsq20);
 401
 402             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 403             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 404             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 405
 406             /* Load parameters for j particles */
 407             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 408                                                                  charge+jnrC+0,charge+jnrD+0,
 409                                                                  charge+jnrE+0,charge+jnrF+0,
 410                                                                  charge+jnrG+0,charge+jnrH+0);
 411
 412             fjx0             = _mm256_setzero_ps();
 413             fjy0             = _mm256_setzero_ps();
 414             fjz0             = _mm256_setzero_ps();
 415
 416             /**************************
 417              * CALCULATE INTERACTIONS *
 418              **************************/
 419
 420             /* Compute parameters for interactions between i and j atoms */
 421             qq00             = _mm256_mul_ps(iq0,jq0);
 422
 423             /* COULOMB ELECTROSTATICS */
 424             velec            = _mm256_mul_ps(qq00,rinv00);
 425             felec            = _mm256_mul_ps(velec,rinvsq00);
 426
 427             /* Update potential sum for this i atom from the interaction with this j atom. */
 428             velec            = _mm256_andnot_ps(dummy_mask,velec);
 429             velecsum         = _mm256_add_ps(velecsum,velec);
 430
 431             fscal            = felec;
 432
 433             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 434
 435             /* Calculate temporary vectorial force */
 436             tx               = _mm256_mul_ps(fscal,dx00);
 437             ty               = _mm256_mul_ps(fscal,dy00);
 438             tz               = _mm256_mul_ps(fscal,dz00);
 439
 440             /* Update vectorial force */
 441             fix0             = _mm256_add_ps(fix0,tx);
 442             fiy0             = _mm256_add_ps(fiy0,ty);
 443             fiz0             = _mm256_add_ps(fiz0,tz);
 444
 445             fjx0             = _mm256_add_ps(fjx0,tx);
 446             fjy0             = _mm256_add_ps(fjy0,ty);
 447             fjz0             = _mm256_add_ps(fjz0,tz);
 448
 449             /**************************
 450              * CALCULATE INTERACTIONS *
 451              **************************/
 452
 453             /* Compute parameters for interactions between i and j atoms */
 454             qq10             = _mm256_mul_ps(iq1,jq0);
 455
 456             /* COULOMB ELECTROSTATICS */
 457             velec            = _mm256_mul_ps(qq10,rinv10);
 458             felec            = _mm256_mul_ps(velec,rinvsq10);
 459
 460             /* Update potential sum for this i atom from the interaction with this j atom. */
 461             velec            = _mm256_andnot_ps(dummy_mask,velec);
 462             velecsum         = _mm256_add_ps(velecsum,velec);
 463
 464             fscal            = felec;
 465
 466             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 467
 468             /* Calculate temporary vectorial force */
 469             tx               = _mm256_mul_ps(fscal,dx10);
 470             ty               = _mm256_mul_ps(fscal,dy10);
 471             tz               = _mm256_mul_ps(fscal,dz10);
 472
 473             /* Update vectorial force */
 474             fix1             = _mm256_add_ps(fix1,tx);
 475             fiy1             = _mm256_add_ps(fiy1,ty);
 476             fiz1             = _mm256_add_ps(fiz1,tz);
 477
 478             fjx0             = _mm256_add_ps(fjx0,tx);
 479             fjy0             = _mm256_add_ps(fjy0,ty);
 480             fjz0             = _mm256_add_ps(fjz0,tz);
 481
 482             /**************************
 483              * CALCULATE INTERACTIONS *
 484              **************************/
 485
 486             /* Compute parameters for interactions between i and j atoms */
 487             qq20             = _mm256_mul_ps(iq2,jq0);
 488
 489             /* COULOMB ELECTROSTATICS */
 490             velec            = _mm256_mul_ps(qq20,rinv20);
 491             felec            = _mm256_mul_ps(velec,rinvsq20);
 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
 497             fscal            = felec;
 498
 499             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 500
 501             /* Calculate temporary vectorial force */
 502             tx               = _mm256_mul_ps(fscal,dx20);
 503             ty               = _mm256_mul_ps(fscal,dy20);
 504             tz               = _mm256_mul_ps(fscal,dz20);
 505
 506             /* Update vectorial force */
 507             fix2             = _mm256_add_ps(fix2,tx);
 508             fiy2             = _mm256_add_ps(fiy2,ty);
 509             fiz2             = _mm256_add_ps(fiz2,tz);
 510
 511             fjx0             = _mm256_add_ps(fjx0,tx);
 512             fjy0             = _mm256_add_ps(fjy0,ty);
 513             fjz0             = _mm256_add_ps(fjz0,tz);
 514
 515             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 516             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 517             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 518             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 519             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 520             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 521             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 522             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 523
 524             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 525
 526             /* Inner loop uses 84 flops */
 527         }
 528
 529         /* End of innermost loop */
 530
 531         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 532                                                  f+i_coord_offset,fshift+i_shift_offset);
 533
 534         ggid                        = gid[iidx];
 535         /* Update potential energies */
 536         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 537
 538         /* Increment number of inner iterations */
 539         inneriter                  += j_index_end - j_index_start;
 540
 541         /* Outer loop uses 19 flops */
 542     }
 543
 544     /* Increment number of outer iterations */
 545     outeriter        += nri;
 546
 547     /* Update outer/inner flops */
 548
 549     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*84);
 550 }
 551 /*
 552  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW3P1_F_avx_256_single
 553  * Electrostatics interaction: Coulomb
 554  * VdW interaction:            None
 555  * Geometry:                   Water3-Particle
 556  * Calculate force/pot:        Force
 557  */
 558 void
 559 nb_kernel_ElecCoul_VdwNone_GeomW3P1_F_avx_256_single
 560                     (t_nblist                    * gmx_restrict       nlist,
 561                      rvec                        * gmx_restrict          xx,
 562                      rvec                        * gmx_restrict          ff,
 563                      struct t_forcerec           * gmx_restrict          fr,
 564                      t_mdatoms                   * gmx_restrict     mdatoms,
 565                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 566                      t_nrnb                      * gmx_restrict        nrnb)
 567 {
 568     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 569      * just 0 for non-waters.
 570      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 571      * jnr indices corresponding to data put in the four positions in the SIMD register.
 572      */
 573     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 574     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 575     int              jnrA,jnrB,jnrC,jnrD;
 576     int              jnrE,jnrF,jnrG,jnrH;
 577     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 578     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 579     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 580     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 581     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 582     real             rcutoff_scalar;
 583     real             *shiftvec,*fshift,*x,*f;
 584     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 585     real             scratch[4*DIM];
 586     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 587     real *           vdwioffsetptr0;
 588     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 589     real *           vdwioffsetptr1;
 590     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 591     real *           vdwioffsetptr2;
 592     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 593     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 594     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 595     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 596     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 597     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 598     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 599     real             *charge;
 600     __m256           dummy_mask,cutoff_mask;
 601     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 602     __m256           one     = _mm256_set1_ps(1.0);
 603     __m256           two     = _mm256_set1_ps(2.0);
 604     x                = xx[0];
 605     f                = ff[0];
 606
 607     nri              = nlist->nri;
 608     iinr             = nlist->iinr;
 609     jindex           = nlist->jindex;
 610     jjnr             = nlist->jjnr;
 611     shiftidx         = nlist->shift;
 612     gid              = nlist->gid;
 613     shiftvec         = fr->shift_vec[0];
 614     fshift           = fr->fshift[0];
 615     facel            = _mm256_set1_ps(fr->ic->epsfac);
 616     charge           = mdatoms->chargeA;
 617
 618     /* Setup water-specific parameters */
 619     inr              = nlist->iinr[0];
 620     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 621     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 622     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 623
 624     /* Avoid stupid compiler warnings */
 625     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 626     j_coord_offsetA = 0;
 627     j_coord_offsetB = 0;
 628     j_coord_offsetC = 0;
 629     j_coord_offsetD = 0;
 630     j_coord_offsetE = 0;
 631     j_coord_offsetF = 0;
 632     j_coord_offsetG = 0;
 633     j_coord_offsetH = 0;
 634
 635     outeriter        = 0;
 636     inneriter        = 0;
 637
 638     for(iidx=0;iidx<4*DIM;iidx++)
 639     {
 640         scratch[iidx] = 0.0;
 641     }
 642
 643     /* Start outer loop over neighborlists */
 644     for(iidx=0; iidx<nri; iidx++)
 645     {
 646         /* Load shift vector for this list */
 647         i_shift_offset   = DIM*shiftidx[iidx];
 648
 649         /* Load limits for loop over neighbors */
 650         j_index_start    = jindex[iidx];
 651         j_index_end      = jindex[iidx+1];
 652
 653         /* Get outer coordinate index */
 654         inr              = iinr[iidx];
 655         i_coord_offset   = DIM*inr;
 656
 657         /* Load i particle coords and add shift vector */
 658         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 659                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 660
 661         fix0             = _mm256_setzero_ps();
 662         fiy0             = _mm256_setzero_ps();
 663         fiz0             = _mm256_setzero_ps();
 664         fix1             = _mm256_setzero_ps();
 665         fiy1             = _mm256_setzero_ps();
 666         fiz1             = _mm256_setzero_ps();
 667         fix2             = _mm256_setzero_ps();
 668         fiy2             = _mm256_setzero_ps();
 669         fiz2             = _mm256_setzero_ps();
 670
 671         /* Start inner kernel loop */
 672         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 673         {
 674
 675             /* Get j neighbor index, and coordinate index */
 676             jnrA             = jjnr[jidx];
 677             jnrB             = jjnr[jidx+1];
 678             jnrC             = jjnr[jidx+2];
 679             jnrD             = jjnr[jidx+3];
 680             jnrE             = jjnr[jidx+4];
 681             jnrF             = jjnr[jidx+5];
 682             jnrG             = jjnr[jidx+6];
 683             jnrH             = jjnr[jidx+7];
 684             j_coord_offsetA  = DIM*jnrA;
 685             j_coord_offsetB  = DIM*jnrB;
 686             j_coord_offsetC  = DIM*jnrC;
 687             j_coord_offsetD  = DIM*jnrD;
 688             j_coord_offsetE  = DIM*jnrE;
 689             j_coord_offsetF  = DIM*jnrF;
 690             j_coord_offsetG  = DIM*jnrG;
 691             j_coord_offsetH  = DIM*jnrH;
 692
 693             /* load j atom coordinates */
 694             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 695                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 696                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 697                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 698                                                  &jx0,&jy0,&jz0);
 699
 700             /* Calculate displacement vector */
 701             dx00             = _mm256_sub_ps(ix0,jx0);
 702             dy00             = _mm256_sub_ps(iy0,jy0);
 703             dz00             = _mm256_sub_ps(iz0,jz0);
 704             dx10             = _mm256_sub_ps(ix1,jx0);
 705             dy10             = _mm256_sub_ps(iy1,jy0);
 706             dz10             = _mm256_sub_ps(iz1,jz0);
 707             dx20             = _mm256_sub_ps(ix2,jx0);
 708             dy20             = _mm256_sub_ps(iy2,jy0);
 709             dz20             = _mm256_sub_ps(iz2,jz0);
 710
 711             /* Calculate squared distance and things based on it */
 712             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 713             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 714             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 715
 716             rinv00           = avx256_invsqrt_f(rsq00);
 717             rinv10           = avx256_invsqrt_f(rsq10);
 718             rinv20           = avx256_invsqrt_f(rsq20);
 719
 720             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 721             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 722             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 723
 724             /* Load parameters for j particles */
 725             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 726                                                                  charge+jnrC+0,charge+jnrD+0,
 727                                                                  charge+jnrE+0,charge+jnrF+0,
 728                                                                  charge+jnrG+0,charge+jnrH+0);
 729
 730             fjx0             = _mm256_setzero_ps();
 731             fjy0             = _mm256_setzero_ps();
 732             fjz0             = _mm256_setzero_ps();
 733
 734             /**************************
 735              * CALCULATE INTERACTIONS *
 736              **************************/
 737
 738             /* Compute parameters for interactions between i and j atoms */
 739             qq00             = _mm256_mul_ps(iq0,jq0);
 740
 741             /* COULOMB ELECTROSTATICS */
 742             velec            = _mm256_mul_ps(qq00,rinv00);
 743             felec            = _mm256_mul_ps(velec,rinvsq00);
 744
 745             fscal            = felec;
 746
 747             /* Calculate temporary vectorial force */
 748             tx               = _mm256_mul_ps(fscal,dx00);
 749             ty               = _mm256_mul_ps(fscal,dy00);
 750             tz               = _mm256_mul_ps(fscal,dz00);
 751
 752             /* Update vectorial force */
 753             fix0             = _mm256_add_ps(fix0,tx);
 754             fiy0             = _mm256_add_ps(fiy0,ty);
 755             fiz0             = _mm256_add_ps(fiz0,tz);
 756
 757             fjx0             = _mm256_add_ps(fjx0,tx);
 758             fjy0             = _mm256_add_ps(fjy0,ty);
 759             fjz0             = _mm256_add_ps(fjz0,tz);
 760
 761             /**************************
 762              * CALCULATE INTERACTIONS *
 763              **************************/
 764
 765             /* Compute parameters for interactions between i and j atoms */
 766             qq10             = _mm256_mul_ps(iq1,jq0);
 767
 768             /* COULOMB ELECTROSTATICS */
 769             velec            = _mm256_mul_ps(qq10,rinv10);
 770             felec            = _mm256_mul_ps(velec,rinvsq10);
 771
 772             fscal            = felec;
 773
 774             /* Calculate temporary vectorial force */
 775             tx               = _mm256_mul_ps(fscal,dx10);
 776             ty               = _mm256_mul_ps(fscal,dy10);
 777             tz               = _mm256_mul_ps(fscal,dz10);
 778
 779             /* Update vectorial force */
 780             fix1             = _mm256_add_ps(fix1,tx);
 781             fiy1             = _mm256_add_ps(fiy1,ty);
 782             fiz1             = _mm256_add_ps(fiz1,tz);
 783
 784             fjx0             = _mm256_add_ps(fjx0,tx);
 785             fjy0             = _mm256_add_ps(fjy0,ty);
 786             fjz0             = _mm256_add_ps(fjz0,tz);
 787
 788             /**************************
 789              * CALCULATE INTERACTIONS *
 790              **************************/
 791
 792             /* Compute parameters for interactions between i and j atoms */
 793             qq20             = _mm256_mul_ps(iq2,jq0);
 794
 795             /* COULOMB ELECTROSTATICS */
 796             velec            = _mm256_mul_ps(qq20,rinv20);
 797             felec            = _mm256_mul_ps(velec,rinvsq20);
 798
 799             fscal            = felec;
 800
 801             /* Calculate temporary vectorial force */
 802             tx               = _mm256_mul_ps(fscal,dx20);
 803             ty               = _mm256_mul_ps(fscal,dy20);
 804             tz               = _mm256_mul_ps(fscal,dz20);
 805
 806             /* Update vectorial force */
 807             fix2             = _mm256_add_ps(fix2,tx);
 808             fiy2             = _mm256_add_ps(fiy2,ty);
 809             fiz2             = _mm256_add_ps(fiz2,tz);
 810
 811             fjx0             = _mm256_add_ps(fjx0,tx);
 812             fjy0             = _mm256_add_ps(fjy0,ty);
 813             fjz0             = _mm256_add_ps(fjz0,tz);
 814
 815             fjptrA             = f+j_coord_offsetA;
 816             fjptrB             = f+j_coord_offsetB;
 817             fjptrC             = f+j_coord_offsetC;
 818             fjptrD             = f+j_coord_offsetD;
 819             fjptrE             = f+j_coord_offsetE;
 820             fjptrF             = f+j_coord_offsetF;
 821             fjptrG             = f+j_coord_offsetG;
 822             fjptrH             = f+j_coord_offsetH;
 823
 824             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 825
 826             /* Inner loop uses 81 flops */
 827         }
 828
 829         if(jidx<j_index_end)
 830         {
 831
 832             /* Get j neighbor index, and coordinate index */
 833             jnrlistA         = jjnr[jidx];
 834             jnrlistB         = jjnr[jidx+1];
 835             jnrlistC         = jjnr[jidx+2];
 836             jnrlistD         = jjnr[jidx+3];
 837             jnrlistE         = jjnr[jidx+4];
 838             jnrlistF         = jjnr[jidx+5];
 839             jnrlistG         = jjnr[jidx+6];
 840             jnrlistH         = jjnr[jidx+7];
 841             /* Sign of each element will be negative for non-real atoms.
 842              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 843              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 844              */
 845             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 846                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 847
 848             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 849             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 850             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 851             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 852             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 853             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 854             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 855             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 856             j_coord_offsetA  = DIM*jnrA;
 857             j_coord_offsetB  = DIM*jnrB;
 858             j_coord_offsetC  = DIM*jnrC;
 859             j_coord_offsetD  = DIM*jnrD;
 860             j_coord_offsetE  = DIM*jnrE;
 861             j_coord_offsetF  = DIM*jnrF;
 862             j_coord_offsetG  = DIM*jnrG;
 863             j_coord_offsetH  = DIM*jnrH;
 864
 865             /* load j atom coordinates */
 866             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 867                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 868                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 869                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 870                                                  &jx0,&jy0,&jz0);
 871
 872             /* Calculate displacement vector */
 873             dx00             = _mm256_sub_ps(ix0,jx0);
 874             dy00             = _mm256_sub_ps(iy0,jy0);
 875             dz00             = _mm256_sub_ps(iz0,jz0);
 876             dx10             = _mm256_sub_ps(ix1,jx0);
 877             dy10             = _mm256_sub_ps(iy1,jy0);
 878             dz10             = _mm256_sub_ps(iz1,jz0);
 879             dx20             = _mm256_sub_ps(ix2,jx0);
 880             dy20             = _mm256_sub_ps(iy2,jy0);
 881             dz20             = _mm256_sub_ps(iz2,jz0);
 882
 883             /* Calculate squared distance and things based on it */
 884             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 885             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 886             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 887
 888             rinv00           = avx256_invsqrt_f(rsq00);
 889             rinv10           = avx256_invsqrt_f(rsq10);
 890             rinv20           = avx256_invsqrt_f(rsq20);
 891
 892             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 893             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 894             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 895
 896             /* Load parameters for j particles */
 897             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 898                                                                  charge+jnrC+0,charge+jnrD+0,
 899                                                                  charge+jnrE+0,charge+jnrF+0,
 900                                                                  charge+jnrG+0,charge+jnrH+0);
 901
 902             fjx0             = _mm256_setzero_ps();
 903             fjy0             = _mm256_setzero_ps();
 904             fjz0             = _mm256_setzero_ps();
 905
 906             /**************************
 907              * CALCULATE INTERACTIONS *
 908              **************************/
 909
 910             /* Compute parameters for interactions between i and j atoms */
 911             qq00             = _mm256_mul_ps(iq0,jq0);
 912
 913             /* COULOMB ELECTROSTATICS */
 914             velec            = _mm256_mul_ps(qq00,rinv00);
 915             felec            = _mm256_mul_ps(velec,rinvsq00);
 916
 917             fscal            = felec;
 918
 919             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 920
 921             /* Calculate temporary vectorial force */
 922             tx               = _mm256_mul_ps(fscal,dx00);
 923             ty               = _mm256_mul_ps(fscal,dy00);
 924             tz               = _mm256_mul_ps(fscal,dz00);
 925
 926             /* Update vectorial force */
 927             fix0             = _mm256_add_ps(fix0,tx);
 928             fiy0             = _mm256_add_ps(fiy0,ty);
 929             fiz0             = _mm256_add_ps(fiz0,tz);
 930
 931             fjx0             = _mm256_add_ps(fjx0,tx);
 932             fjy0             = _mm256_add_ps(fjy0,ty);
 933             fjz0             = _mm256_add_ps(fjz0,tz);
 934
 935             /**************************
 936              * CALCULATE INTERACTIONS *
 937              **************************/
 938
 939             /* Compute parameters for interactions between i and j atoms */
 940             qq10             = _mm256_mul_ps(iq1,jq0);
 941
 942             /* COULOMB ELECTROSTATICS */
 943             velec            = _mm256_mul_ps(qq10,rinv10);
 944             felec            = _mm256_mul_ps(velec,rinvsq10);
 945
 946             fscal            = felec;
 947
 948             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 949
 950             /* Calculate temporary vectorial force */
 951             tx               = _mm256_mul_ps(fscal,dx10);
 952             ty               = _mm256_mul_ps(fscal,dy10);
 953             tz               = _mm256_mul_ps(fscal,dz10);
 954
 955             /* Update vectorial force */
 956             fix1             = _mm256_add_ps(fix1,tx);
 957             fiy1             = _mm256_add_ps(fiy1,ty);
 958             fiz1             = _mm256_add_ps(fiz1,tz);
 959
 960             fjx0             = _mm256_add_ps(fjx0,tx);
 961             fjy0             = _mm256_add_ps(fjy0,ty);
 962             fjz0             = _mm256_add_ps(fjz0,tz);
 963
 964             /**************************
 965              * CALCULATE INTERACTIONS *
 966              **************************/
 967
 968             /* Compute parameters for interactions between i and j atoms */
 969             qq20             = _mm256_mul_ps(iq2,jq0);
 970
 971             /* COULOMB ELECTROSTATICS */
 972             velec            = _mm256_mul_ps(qq20,rinv20);
 973             felec            = _mm256_mul_ps(velec,rinvsq20);
 974
 975             fscal            = felec;
 976
 977             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 978
 979             /* Calculate temporary vectorial force */
 980             tx               = _mm256_mul_ps(fscal,dx20);
 981             ty               = _mm256_mul_ps(fscal,dy20);
 982             tz               = _mm256_mul_ps(fscal,dz20);
 983
 984             /* Update vectorial force */
 985             fix2             = _mm256_add_ps(fix2,tx);
 986             fiy2             = _mm256_add_ps(fiy2,ty);
 987             fiz2             = _mm256_add_ps(fiz2,tz);
 988
 989             fjx0             = _mm256_add_ps(fjx0,tx);
 990             fjy0             = _mm256_add_ps(fjy0,ty);
 991             fjz0             = _mm256_add_ps(fjz0,tz);
 992
 993             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 994             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 995             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 996             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 997             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 998             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 999             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1000             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1001
1002             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1003
1004             /* Inner loop uses 81 flops */
1005         }
1006
1007         /* End of innermost loop */
1008
1009         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1010                                                  f+i_coord_offset,fshift+i_shift_offset);
1011
1012         /* Increment number of inner iterations */
1013         inneriter                  += j_index_end - j_index_start;
1014
1015         /* Outer loop uses 18 flops */
1016     }
1017
1018     /* Increment number of outer iterations */
1019     outeriter        += nri;
1020
1021     /* Update outer/inner flops */
1022
1023     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*81);
1024 }