src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCoul_VdwNone_GeomW4P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW4P1_VF_avx_256_single
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            None
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_VdwNone_GeomW4P1_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  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     real *           vdwioffsetptr3;
  93     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  95     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  97     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  99     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     __m256           dummy_mask,cutoff_mask;
 102     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 103     __m256           one     = _mm256_set1_ps(1.0);
 104     __m256           two     = _mm256_set1_ps(2.0);
 105     x                = xx[0];
 106     f                = ff[0];
 107
 108     nri              = nlist->nri;
 109     iinr             = nlist->iinr;
 110     jindex           = nlist->jindex;
 111     jjnr             = nlist->jjnr;
 112     shiftidx         = nlist->shift;
 113     gid              = nlist->gid;
 114     shiftvec         = fr->shift_vec[0];
 115     fshift           = fr->fshift[0];
 116     facel            = _mm256_set1_ps(fr->epsfac);
 117     charge           = mdatoms->chargeA;
 118
 119     /* Setup water-specific parameters */
 120     inr              = nlist->iinr[0];
 121     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 122     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 123     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 124
 125     /* Avoid stupid compiler warnings */
 126     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 127     j_coord_offsetA = 0;
 128     j_coord_offsetB = 0;
 129     j_coord_offsetC = 0;
 130     j_coord_offsetD = 0;
 131     j_coord_offsetE = 0;
 132     j_coord_offsetF = 0;
 133     j_coord_offsetG = 0;
 134     j_coord_offsetH = 0;
 135
 136     outeriter        = 0;
 137     inneriter        = 0;
 138
 139     for(iidx=0;iidx<4*DIM;iidx++)
 140     {
 141         scratch[iidx] = 0.0;
 142     }
 143
 144     /* Start outer loop over neighborlists */
 145     for(iidx=0; iidx<nri; iidx++)
 146     {
 147         /* Load shift vector for this list */
 148         i_shift_offset   = DIM*shiftidx[iidx];
 149
 150         /* Load limits for loop over neighbors */
 151         j_index_start    = jindex[iidx];
 152         j_index_end      = jindex[iidx+1];
 153
 154         /* Get outer coordinate index */
 155         inr              = iinr[iidx];
 156         i_coord_offset   = DIM*inr;
 157
 158         /* Load i particle coords and add shift vector */
 159         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 160                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 161
 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         fix3             = _mm256_setzero_ps();
 169         fiy3             = _mm256_setzero_ps();
 170         fiz3             = _mm256_setzero_ps();
 171
 172         /* Reset potential sums */
 173         velecsum         = _mm256_setzero_ps();
 174
 175         /* Start inner kernel loop */
 176         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 177         {
 178
 179             /* Get j neighbor index, and coordinate index */
 180             jnrA             = jjnr[jidx];
 181             jnrB             = jjnr[jidx+1];
 182             jnrC             = jjnr[jidx+2];
 183             jnrD             = jjnr[jidx+3];
 184             jnrE             = jjnr[jidx+4];
 185             jnrF             = jjnr[jidx+5];
 186             jnrG             = jjnr[jidx+6];
 187             jnrH             = jjnr[jidx+7];
 188             j_coord_offsetA  = DIM*jnrA;
 189             j_coord_offsetB  = DIM*jnrB;
 190             j_coord_offsetC  = DIM*jnrC;
 191             j_coord_offsetD  = DIM*jnrD;
 192             j_coord_offsetE  = DIM*jnrE;
 193             j_coord_offsetF  = DIM*jnrF;
 194             j_coord_offsetG  = DIM*jnrG;
 195             j_coord_offsetH  = DIM*jnrH;
 196
 197             /* load j atom coordinates */
 198             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 199                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 200                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 201                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 202                                                  &jx0,&jy0,&jz0);
 203
 204             /* Calculate displacement vector */
 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             dx30             = _mm256_sub_ps(ix3,jx0);
 212             dy30             = _mm256_sub_ps(iy3,jy0);
 213             dz30             = _mm256_sub_ps(iz3,jz0);
 214
 215             /* Calculate squared distance and things based on it */
 216             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 217             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 218             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 219
 220             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 221             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 222             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 223
 224             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 225             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 226             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 227
 228             /* Load parameters for j particles */
 229             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 230                                                                  charge+jnrC+0,charge+jnrD+0,
 231                                                                  charge+jnrE+0,charge+jnrF+0,
 232                                                                  charge+jnrG+0,charge+jnrH+0);
 233
 234             fjx0             = _mm256_setzero_ps();
 235             fjy0             = _mm256_setzero_ps();
 236             fjz0             = _mm256_setzero_ps();
 237
 238             /**************************
 239              * CALCULATE INTERACTIONS *
 240              **************************/
 241
 242             /* Compute parameters for interactions between i and j atoms */
 243             qq10             = _mm256_mul_ps(iq1,jq0);
 244
 245             /* COULOMB ELECTROSTATICS */
 246             velec            = _mm256_mul_ps(qq10,rinv10);
 247             felec            = _mm256_mul_ps(velec,rinvsq10);
 248
 249             /* Update potential sum for this i atom from the interaction with this j atom. */
 250             velecsum         = _mm256_add_ps(velecsum,velec);
 251
 252             fscal            = felec;
 253
 254             /* Calculate temporary vectorial force */
 255             tx               = _mm256_mul_ps(fscal,dx10);
 256             ty               = _mm256_mul_ps(fscal,dy10);
 257             tz               = _mm256_mul_ps(fscal,dz10);
 258
 259             /* Update vectorial force */
 260             fix1             = _mm256_add_ps(fix1,tx);
 261             fiy1             = _mm256_add_ps(fiy1,ty);
 262             fiz1             = _mm256_add_ps(fiz1,tz);
 263
 264             fjx0             = _mm256_add_ps(fjx0,tx);
 265             fjy0             = _mm256_add_ps(fjy0,ty);
 266             fjz0             = _mm256_add_ps(fjz0,tz);
 267
 268             /**************************
 269              * CALCULATE INTERACTIONS *
 270              **************************/
 271
 272             /* Compute parameters for interactions between i and j atoms */
 273             qq20             = _mm256_mul_ps(iq2,jq0);
 274
 275             /* COULOMB ELECTROSTATICS */
 276             velec            = _mm256_mul_ps(qq20,rinv20);
 277             felec            = _mm256_mul_ps(velec,rinvsq20);
 278
 279             /* Update potential sum for this i atom from the interaction with this j atom. */
 280             velecsum         = _mm256_add_ps(velecsum,velec);
 281
 282             fscal            = felec;
 283
 284             /* Calculate temporary vectorial force */
 285             tx               = _mm256_mul_ps(fscal,dx20);
 286             ty               = _mm256_mul_ps(fscal,dy20);
 287             tz               = _mm256_mul_ps(fscal,dz20);
 288
 289             /* Update vectorial force */
 290             fix2             = _mm256_add_ps(fix2,tx);
 291             fiy2             = _mm256_add_ps(fiy2,ty);
 292             fiz2             = _mm256_add_ps(fiz2,tz);
 293
 294             fjx0             = _mm256_add_ps(fjx0,tx);
 295             fjy0             = _mm256_add_ps(fjy0,ty);
 296             fjz0             = _mm256_add_ps(fjz0,tz);
 297
 298             /**************************
 299              * CALCULATE INTERACTIONS *
 300              **************************/
 301
 302             /* Compute parameters for interactions between i and j atoms */
 303             qq30             = _mm256_mul_ps(iq3,jq0);
 304
 305             /* COULOMB ELECTROSTATICS */
 306             velec            = _mm256_mul_ps(qq30,rinv30);
 307             felec            = _mm256_mul_ps(velec,rinvsq30);
 308
 309             /* Update potential sum for this i atom from the interaction with this j atom. */
 310             velecsum         = _mm256_add_ps(velecsum,velec);
 311
 312             fscal            = felec;
 313
 314             /* Calculate temporary vectorial force */
 315             tx               = _mm256_mul_ps(fscal,dx30);
 316             ty               = _mm256_mul_ps(fscal,dy30);
 317             tz               = _mm256_mul_ps(fscal,dz30);
 318
 319             /* Update vectorial force */
 320             fix3             = _mm256_add_ps(fix3,tx);
 321             fiy3             = _mm256_add_ps(fiy3,ty);
 322             fiz3             = _mm256_add_ps(fiz3,tz);
 323
 324             fjx0             = _mm256_add_ps(fjx0,tx);
 325             fjy0             = _mm256_add_ps(fjy0,ty);
 326             fjz0             = _mm256_add_ps(fjz0,tz);
 327
 328             fjptrA             = f+j_coord_offsetA;
 329             fjptrB             = f+j_coord_offsetB;
 330             fjptrC             = f+j_coord_offsetC;
 331             fjptrD             = f+j_coord_offsetD;
 332             fjptrE             = f+j_coord_offsetE;
 333             fjptrF             = f+j_coord_offsetF;
 334             fjptrG             = f+j_coord_offsetG;
 335             fjptrH             = f+j_coord_offsetH;
 336
 337             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 338
 339             /* Inner loop uses 84 flops */
 340         }
 341
 342         if(jidx<j_index_end)
 343         {
 344
 345             /* Get j neighbor index, and coordinate index */
 346             jnrlistA         = jjnr[jidx];
 347             jnrlistB         = jjnr[jidx+1];
 348             jnrlistC         = jjnr[jidx+2];
 349             jnrlistD         = jjnr[jidx+3];
 350             jnrlistE         = jjnr[jidx+4];
 351             jnrlistF         = jjnr[jidx+5];
 352             jnrlistG         = jjnr[jidx+6];
 353             jnrlistH         = jjnr[jidx+7];
 354             /* Sign of each element will be negative for non-real atoms.
 355              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 356              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 357              */
 358             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 359                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 360
 361             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 362             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 363             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 364             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 365             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 366             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 367             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 368             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 369             j_coord_offsetA  = DIM*jnrA;
 370             j_coord_offsetB  = DIM*jnrB;
 371             j_coord_offsetC  = DIM*jnrC;
 372             j_coord_offsetD  = DIM*jnrD;
 373             j_coord_offsetE  = DIM*jnrE;
 374             j_coord_offsetF  = DIM*jnrF;
 375             j_coord_offsetG  = DIM*jnrG;
 376             j_coord_offsetH  = DIM*jnrH;
 377
 378             /* load j atom coordinates */
 379             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 380                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 381                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 382                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 383                                                  &jx0,&jy0,&jz0);
 384
 385             /* Calculate displacement vector */
 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             dx30             = _mm256_sub_ps(ix3,jx0);
 393             dy30             = _mm256_sub_ps(iy3,jy0);
 394             dz30             = _mm256_sub_ps(iz3,jz0);
 395
 396             /* Calculate squared distance and things based on it */
 397             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 398             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 399             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 400
 401             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 402             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 403             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 404
 405             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 406             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 407             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 408
 409             /* Load parameters for j particles */
 410             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 411                                                                  charge+jnrC+0,charge+jnrD+0,
 412                                                                  charge+jnrE+0,charge+jnrF+0,
 413                                                                  charge+jnrG+0,charge+jnrH+0);
 414
 415             fjx0             = _mm256_setzero_ps();
 416             fjy0             = _mm256_setzero_ps();
 417             fjz0             = _mm256_setzero_ps();
 418
 419             /**************************
 420              * CALCULATE INTERACTIONS *
 421              **************************/
 422
 423             /* Compute parameters for interactions between i and j atoms */
 424             qq10             = _mm256_mul_ps(iq1,jq0);
 425
 426             /* COULOMB ELECTROSTATICS */
 427             velec            = _mm256_mul_ps(qq10,rinv10);
 428             felec            = _mm256_mul_ps(velec,rinvsq10);
 429
 430             /* Update potential sum for this i atom from the interaction with this j atom. */
 431             velec            = _mm256_andnot_ps(dummy_mask,velec);
 432             velecsum         = _mm256_add_ps(velecsum,velec);
 433
 434             fscal            = felec;
 435
 436             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 437
 438             /* Calculate temporary vectorial force */
 439             tx               = _mm256_mul_ps(fscal,dx10);
 440             ty               = _mm256_mul_ps(fscal,dy10);
 441             tz               = _mm256_mul_ps(fscal,dz10);
 442
 443             /* Update vectorial force */
 444             fix1             = _mm256_add_ps(fix1,tx);
 445             fiy1             = _mm256_add_ps(fiy1,ty);
 446             fiz1             = _mm256_add_ps(fiz1,tz);
 447
 448             fjx0             = _mm256_add_ps(fjx0,tx);
 449             fjy0             = _mm256_add_ps(fjy0,ty);
 450             fjz0             = _mm256_add_ps(fjz0,tz);
 451
 452             /**************************
 453              * CALCULATE INTERACTIONS *
 454              **************************/
 455
 456             /* Compute parameters for interactions between i and j atoms */
 457             qq20             = _mm256_mul_ps(iq2,jq0);
 458
 459             /* COULOMB ELECTROSTATICS */
 460             velec            = _mm256_mul_ps(qq20,rinv20);
 461             felec            = _mm256_mul_ps(velec,rinvsq20);
 462
 463             /* Update potential sum for this i atom from the interaction with this j atom. */
 464             velec            = _mm256_andnot_ps(dummy_mask,velec);
 465             velecsum         = _mm256_add_ps(velecsum,velec);
 466
 467             fscal            = felec;
 468
 469             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 470
 471             /* Calculate temporary vectorial force */
 472             tx               = _mm256_mul_ps(fscal,dx20);
 473             ty               = _mm256_mul_ps(fscal,dy20);
 474             tz               = _mm256_mul_ps(fscal,dz20);
 475
 476             /* Update vectorial force */
 477             fix2             = _mm256_add_ps(fix2,tx);
 478             fiy2             = _mm256_add_ps(fiy2,ty);
 479             fiz2             = _mm256_add_ps(fiz2,tz);
 480
 481             fjx0             = _mm256_add_ps(fjx0,tx);
 482             fjy0             = _mm256_add_ps(fjy0,ty);
 483             fjz0             = _mm256_add_ps(fjz0,tz);
 484
 485             /**************************
 486              * CALCULATE INTERACTIONS *
 487              **************************/
 488
 489             /* Compute parameters for interactions between i and j atoms */
 490             qq30             = _mm256_mul_ps(iq3,jq0);
 491
 492             /* COULOMB ELECTROSTATICS */
 493             velec            = _mm256_mul_ps(qq30,rinv30);
 494             felec            = _mm256_mul_ps(velec,rinvsq30);
 495
 496             /* Update potential sum for this i atom from the interaction with this j atom. */
 497             velec            = _mm256_andnot_ps(dummy_mask,velec);
 498             velecsum         = _mm256_add_ps(velecsum,velec);
 499
 500             fscal            = felec;
 501
 502             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 503
 504             /* Calculate temporary vectorial force */
 505             tx               = _mm256_mul_ps(fscal,dx30);
 506             ty               = _mm256_mul_ps(fscal,dy30);
 507             tz               = _mm256_mul_ps(fscal,dz30);
 508
 509             /* Update vectorial force */
 510             fix3             = _mm256_add_ps(fix3,tx);
 511             fiy3             = _mm256_add_ps(fiy3,ty);
 512             fiz3             = _mm256_add_ps(fiz3,tz);
 513
 514             fjx0             = _mm256_add_ps(fjx0,tx);
 515             fjy0             = _mm256_add_ps(fjy0,ty);
 516             fjz0             = _mm256_add_ps(fjz0,tz);
 517
 518             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 519             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 520             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 521             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 522             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 523             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 524             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 525             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 526
 527             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 528
 529             /* Inner loop uses 84 flops */
 530         }
 531
 532         /* End of innermost loop */
 533
 534         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 535                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
 536
 537         ggid                        = gid[iidx];
 538         /* Update potential energies */
 539         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 540
 541         /* Increment number of inner iterations */
 542         inneriter                  += j_index_end - j_index_start;
 543
 544         /* Outer loop uses 19 flops */
 545     }
 546
 547     /* Increment number of outer iterations */
 548     outeriter        += nri;
 549
 550     /* Update outer/inner flops */
 551
 552     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*84);
 553 }
 554 /*
 555  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW4P1_F_avx_256_single
 556  * Electrostatics interaction: Coulomb
 557  * VdW interaction:            None
 558  * Geometry:                   Water4-Particle
 559  * Calculate force/pot:        Force
 560  */
 561 void
 562 nb_kernel_ElecCoul_VdwNone_GeomW4P1_F_avx_256_single
 563                     (t_nblist                    * gmx_restrict       nlist,
 564                      rvec                        * gmx_restrict          xx,
 565                      rvec                        * gmx_restrict          ff,
 566                      t_forcerec                  * gmx_restrict          fr,
 567                      t_mdatoms                   * gmx_restrict     mdatoms,
 568                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 569                      t_nrnb                      * gmx_restrict        nrnb)
 570 {
 571     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 572      * just 0 for non-waters.
 573      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 574      * jnr indices corresponding to data put in the four positions in the SIMD register.
 575      */
 576     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 577     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 578     int              jnrA,jnrB,jnrC,jnrD;
 579     int              jnrE,jnrF,jnrG,jnrH;
 580     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 581     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 582     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 583     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 584     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 585     real             rcutoff_scalar;
 586     real             *shiftvec,*fshift,*x,*f;
 587     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 588     real             scratch[4*DIM];
 589     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 590     real *           vdwioffsetptr1;
 591     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 592     real *           vdwioffsetptr2;
 593     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 594     real *           vdwioffsetptr3;
 595     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 596     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 597     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 598     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 599     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 600     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 601     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 602     real             *charge;
 603     __m256           dummy_mask,cutoff_mask;
 604     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 605     __m256           one     = _mm256_set1_ps(1.0);
 606     __m256           two     = _mm256_set1_ps(2.0);
 607     x                = xx[0];
 608     f                = ff[0];
 609
 610     nri              = nlist->nri;
 611     iinr             = nlist->iinr;
 612     jindex           = nlist->jindex;
 613     jjnr             = nlist->jjnr;
 614     shiftidx         = nlist->shift;
 615     gid              = nlist->gid;
 616     shiftvec         = fr->shift_vec[0];
 617     fshift           = fr->fshift[0];
 618     facel            = _mm256_set1_ps(fr->epsfac);
 619     charge           = mdatoms->chargeA;
 620
 621     /* Setup water-specific parameters */
 622     inr              = nlist->iinr[0];
 623     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 624     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 625     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 626
 627     /* Avoid stupid compiler warnings */
 628     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 629     j_coord_offsetA = 0;
 630     j_coord_offsetB = 0;
 631     j_coord_offsetC = 0;
 632     j_coord_offsetD = 0;
 633     j_coord_offsetE = 0;
 634     j_coord_offsetF = 0;
 635     j_coord_offsetG = 0;
 636     j_coord_offsetH = 0;
 637
 638     outeriter        = 0;
 639     inneriter        = 0;
 640
 641     for(iidx=0;iidx<4*DIM;iidx++)
 642     {
 643         scratch[iidx] = 0.0;
 644     }
 645
 646     /* Start outer loop over neighborlists */
 647     for(iidx=0; iidx<nri; iidx++)
 648     {
 649         /* Load shift vector for this list */
 650         i_shift_offset   = DIM*shiftidx[iidx];
 651
 652         /* Load limits for loop over neighbors */
 653         j_index_start    = jindex[iidx];
 654         j_index_end      = jindex[iidx+1];
 655
 656         /* Get outer coordinate index */
 657         inr              = iinr[iidx];
 658         i_coord_offset   = DIM*inr;
 659
 660         /* Load i particle coords and add shift vector */
 661         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 662                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 663
 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         fix3             = _mm256_setzero_ps();
 671         fiy3             = _mm256_setzero_ps();
 672         fiz3             = _mm256_setzero_ps();
 673
 674         /* Start inner kernel loop */
 675         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 676         {
 677
 678             /* Get j neighbor index, and coordinate index */
 679             jnrA             = jjnr[jidx];
 680             jnrB             = jjnr[jidx+1];
 681             jnrC             = jjnr[jidx+2];
 682             jnrD             = jjnr[jidx+3];
 683             jnrE             = jjnr[jidx+4];
 684             jnrF             = jjnr[jidx+5];
 685             jnrG             = jjnr[jidx+6];
 686             jnrH             = jjnr[jidx+7];
 687             j_coord_offsetA  = DIM*jnrA;
 688             j_coord_offsetB  = DIM*jnrB;
 689             j_coord_offsetC  = DIM*jnrC;
 690             j_coord_offsetD  = DIM*jnrD;
 691             j_coord_offsetE  = DIM*jnrE;
 692             j_coord_offsetF  = DIM*jnrF;
 693             j_coord_offsetG  = DIM*jnrG;
 694             j_coord_offsetH  = DIM*jnrH;
 695
 696             /* load j atom coordinates */
 697             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 698                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 699                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 700                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 701                                                  &jx0,&jy0,&jz0);
 702
 703             /* Calculate displacement vector */
 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             dx30             = _mm256_sub_ps(ix3,jx0);
 711             dy30             = _mm256_sub_ps(iy3,jy0);
 712             dz30             = _mm256_sub_ps(iz3,jz0);
 713
 714             /* Calculate squared distance and things based on it */
 715             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 716             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 717             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 718
 719             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 720             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 721             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 722
 723             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 724             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 725             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 726
 727             /* Load parameters for j particles */
 728             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 729                                                                  charge+jnrC+0,charge+jnrD+0,
 730                                                                  charge+jnrE+0,charge+jnrF+0,
 731                                                                  charge+jnrG+0,charge+jnrH+0);
 732
 733             fjx0             = _mm256_setzero_ps();
 734             fjy0             = _mm256_setzero_ps();
 735             fjz0             = _mm256_setzero_ps();
 736
 737             /**************************
 738              * CALCULATE INTERACTIONS *
 739              **************************/
 740
 741             /* Compute parameters for interactions between i and j atoms */
 742             qq10             = _mm256_mul_ps(iq1,jq0);
 743
 744             /* COULOMB ELECTROSTATICS */
 745             velec            = _mm256_mul_ps(qq10,rinv10);
 746             felec            = _mm256_mul_ps(velec,rinvsq10);
 747
 748             fscal            = felec;
 749
 750             /* Calculate temporary vectorial force */
 751             tx               = _mm256_mul_ps(fscal,dx10);
 752             ty               = _mm256_mul_ps(fscal,dy10);
 753             tz               = _mm256_mul_ps(fscal,dz10);
 754
 755             /* Update vectorial force */
 756             fix1             = _mm256_add_ps(fix1,tx);
 757             fiy1             = _mm256_add_ps(fiy1,ty);
 758             fiz1             = _mm256_add_ps(fiz1,tz);
 759
 760             fjx0             = _mm256_add_ps(fjx0,tx);
 761             fjy0             = _mm256_add_ps(fjy0,ty);
 762             fjz0             = _mm256_add_ps(fjz0,tz);
 763
 764             /**************************
 765              * CALCULATE INTERACTIONS *
 766              **************************/
 767
 768             /* Compute parameters for interactions between i and j atoms */
 769             qq20             = _mm256_mul_ps(iq2,jq0);
 770
 771             /* COULOMB ELECTROSTATICS */
 772             velec            = _mm256_mul_ps(qq20,rinv20);
 773             felec            = _mm256_mul_ps(velec,rinvsq20);
 774
 775             fscal            = felec;
 776
 777             /* Calculate temporary vectorial force */
 778             tx               = _mm256_mul_ps(fscal,dx20);
 779             ty               = _mm256_mul_ps(fscal,dy20);
 780             tz               = _mm256_mul_ps(fscal,dz20);
 781
 782             /* Update vectorial force */
 783             fix2             = _mm256_add_ps(fix2,tx);
 784             fiy2             = _mm256_add_ps(fiy2,ty);
 785             fiz2             = _mm256_add_ps(fiz2,tz);
 786
 787             fjx0             = _mm256_add_ps(fjx0,tx);
 788             fjy0             = _mm256_add_ps(fjy0,ty);
 789             fjz0             = _mm256_add_ps(fjz0,tz);
 790
 791             /**************************
 792              * CALCULATE INTERACTIONS *
 793              **************************/
 794
 795             /* Compute parameters for interactions between i and j atoms */
 796             qq30             = _mm256_mul_ps(iq3,jq0);
 797
 798             /* COULOMB ELECTROSTATICS */
 799             velec            = _mm256_mul_ps(qq30,rinv30);
 800             felec            = _mm256_mul_ps(velec,rinvsq30);
 801
 802             fscal            = felec;
 803
 804             /* Calculate temporary vectorial force */
 805             tx               = _mm256_mul_ps(fscal,dx30);
 806             ty               = _mm256_mul_ps(fscal,dy30);
 807             tz               = _mm256_mul_ps(fscal,dz30);
 808
 809             /* Update vectorial force */
 810             fix3             = _mm256_add_ps(fix3,tx);
 811             fiy3             = _mm256_add_ps(fiy3,ty);
 812             fiz3             = _mm256_add_ps(fiz3,tz);
 813
 814             fjx0             = _mm256_add_ps(fjx0,tx);
 815             fjy0             = _mm256_add_ps(fjy0,ty);
 816             fjz0             = _mm256_add_ps(fjz0,tz);
 817
 818             fjptrA             = f+j_coord_offsetA;
 819             fjptrB             = f+j_coord_offsetB;
 820             fjptrC             = f+j_coord_offsetC;
 821             fjptrD             = f+j_coord_offsetD;
 822             fjptrE             = f+j_coord_offsetE;
 823             fjptrF             = f+j_coord_offsetF;
 824             fjptrG             = f+j_coord_offsetG;
 825             fjptrH             = f+j_coord_offsetH;
 826
 827             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 828
 829             /* Inner loop uses 81 flops */
 830         }
 831
 832         if(jidx<j_index_end)
 833         {
 834
 835             /* Get j neighbor index, and coordinate index */
 836             jnrlistA         = jjnr[jidx];
 837             jnrlistB         = jjnr[jidx+1];
 838             jnrlistC         = jjnr[jidx+2];
 839             jnrlistD         = jjnr[jidx+3];
 840             jnrlistE         = jjnr[jidx+4];
 841             jnrlistF         = jjnr[jidx+5];
 842             jnrlistG         = jjnr[jidx+6];
 843             jnrlistH         = jjnr[jidx+7];
 844             /* Sign of each element will be negative for non-real atoms.
 845              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 846              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 847              */
 848             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 849                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 850
 851             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 852             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 853             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 854             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 855             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 856             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 857             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 858             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 859             j_coord_offsetA  = DIM*jnrA;
 860             j_coord_offsetB  = DIM*jnrB;
 861             j_coord_offsetC  = DIM*jnrC;
 862             j_coord_offsetD  = DIM*jnrD;
 863             j_coord_offsetE  = DIM*jnrE;
 864             j_coord_offsetF  = DIM*jnrF;
 865             j_coord_offsetG  = DIM*jnrG;
 866             j_coord_offsetH  = DIM*jnrH;
 867
 868             /* load j atom coordinates */
 869             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 870                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 871                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 872                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 873                                                  &jx0,&jy0,&jz0);
 874
 875             /* Calculate displacement vector */
 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             dx30             = _mm256_sub_ps(ix3,jx0);
 883             dy30             = _mm256_sub_ps(iy3,jy0);
 884             dz30             = _mm256_sub_ps(iz3,jz0);
 885
 886             /* Calculate squared distance and things based on it */
 887             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 888             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 889             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 890
 891             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 892             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 893             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 894
 895             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 896             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 897             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 898
 899             /* Load parameters for j particles */
 900             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 901                                                                  charge+jnrC+0,charge+jnrD+0,
 902                                                                  charge+jnrE+0,charge+jnrF+0,
 903                                                                  charge+jnrG+0,charge+jnrH+0);
 904
 905             fjx0             = _mm256_setzero_ps();
 906             fjy0             = _mm256_setzero_ps();
 907             fjz0             = _mm256_setzero_ps();
 908
 909             /**************************
 910              * CALCULATE INTERACTIONS *
 911              **************************/
 912
 913             /* Compute parameters for interactions between i and j atoms */
 914             qq10             = _mm256_mul_ps(iq1,jq0);
 915
 916             /* COULOMB ELECTROSTATICS */
 917             velec            = _mm256_mul_ps(qq10,rinv10);
 918             felec            = _mm256_mul_ps(velec,rinvsq10);
 919
 920             fscal            = felec;
 921
 922             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 923
 924             /* Calculate temporary vectorial force */
 925             tx               = _mm256_mul_ps(fscal,dx10);
 926             ty               = _mm256_mul_ps(fscal,dy10);
 927             tz               = _mm256_mul_ps(fscal,dz10);
 928
 929             /* Update vectorial force */
 930             fix1             = _mm256_add_ps(fix1,tx);
 931             fiy1             = _mm256_add_ps(fiy1,ty);
 932             fiz1             = _mm256_add_ps(fiz1,tz);
 933
 934             fjx0             = _mm256_add_ps(fjx0,tx);
 935             fjy0             = _mm256_add_ps(fjy0,ty);
 936             fjz0             = _mm256_add_ps(fjz0,tz);
 937
 938             /**************************
 939              * CALCULATE INTERACTIONS *
 940              **************************/
 941
 942             /* Compute parameters for interactions between i and j atoms */
 943             qq20             = _mm256_mul_ps(iq2,jq0);
 944
 945             /* COULOMB ELECTROSTATICS */
 946             velec            = _mm256_mul_ps(qq20,rinv20);
 947             felec            = _mm256_mul_ps(velec,rinvsq20);
 948
 949             fscal            = felec;
 950
 951             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 952
 953             /* Calculate temporary vectorial force */
 954             tx               = _mm256_mul_ps(fscal,dx20);
 955             ty               = _mm256_mul_ps(fscal,dy20);
 956             tz               = _mm256_mul_ps(fscal,dz20);
 957
 958             /* Update vectorial force */
 959             fix2             = _mm256_add_ps(fix2,tx);
 960             fiy2             = _mm256_add_ps(fiy2,ty);
 961             fiz2             = _mm256_add_ps(fiz2,tz);
 962
 963             fjx0             = _mm256_add_ps(fjx0,tx);
 964             fjy0             = _mm256_add_ps(fjy0,ty);
 965             fjz0             = _mm256_add_ps(fjz0,tz);
 966
 967             /**************************
 968              * CALCULATE INTERACTIONS *
 969              **************************/
 970
 971             /* Compute parameters for interactions between i and j atoms */
 972             qq30             = _mm256_mul_ps(iq3,jq0);
 973
 974             /* COULOMB ELECTROSTATICS */
 975             velec            = _mm256_mul_ps(qq30,rinv30);
 976             felec            = _mm256_mul_ps(velec,rinvsq30);
 977
 978             fscal            = felec;
 979
 980             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 981
 982             /* Calculate temporary vectorial force */
 983             tx               = _mm256_mul_ps(fscal,dx30);
 984             ty               = _mm256_mul_ps(fscal,dy30);
 985             tz               = _mm256_mul_ps(fscal,dz30);
 986
 987             /* Update vectorial force */
 988             fix3             = _mm256_add_ps(fix3,tx);
 989             fiy3             = _mm256_add_ps(fiy3,ty);
 990             fiz3             = _mm256_add_ps(fiz3,tz);
 991
 992             fjx0             = _mm256_add_ps(fjx0,tx);
 993             fjy0             = _mm256_add_ps(fjy0,ty);
 994             fjz0             = _mm256_add_ps(fjz0,tz);
 995
 996             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 997             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 998             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 999             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1000             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1001             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1002             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1003             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1004
1005             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1006
1007             /* Inner loop uses 81 flops */
1008         }
1009
1010         /* End of innermost loop */
1011
1012         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1013                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1014
1015         /* Increment number of inner iterations */
1016         inneriter                  += j_index_end - j_index_start;
1017
1018         /* Outer loop uses 18 flops */
1019     }
1020
1021     /* Increment number of outer iterations */
1022     outeriter        += nri;
1023
1024     /* Update outer/inner flops */
1025
1026     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*81);
1027 }