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) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW4P1_VF_avx_256_single
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            None
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_VdwNone_GeomW4P1_VF_avx_256_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr1;
  87     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  88     real *           vdwioffsetptr2;
  89     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  90     real *           vdwioffsetptr3;
  91     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  93     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  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           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  97     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     __m256           dummy_mask,cutoff_mask;
 100     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 101     __m256           one     = _mm256_set1_ps(1.0);
 102     __m256           two     = _mm256_set1_ps(2.0);
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     facel            = _mm256_set1_ps(fr->epsfac);
 115     charge           = mdatoms->chargeA;
 116
 117     /* Setup water-specific parameters */
 118     inr              = nlist->iinr[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     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 122
 123     /* Avoid stupid compiler warnings */
 124     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 125     j_coord_offsetA = 0;
 126     j_coord_offsetB = 0;
 127     j_coord_offsetC = 0;
 128     j_coord_offsetD = 0;
 129     j_coord_offsetE = 0;
 130     j_coord_offsetF = 0;
 131     j_coord_offsetG = 0;
 132     j_coord_offsetH = 0;
 133
 134     outeriter        = 0;
 135     inneriter        = 0;
 136
 137     for(iidx=0;iidx<4*DIM;iidx++)
 138     {
 139         scratch[iidx] = 0.0;
 140     }
 141
 142     /* Start outer loop over neighborlists */
 143     for(iidx=0; iidx<nri; iidx++)
 144     {
 145         /* Load shift vector for this list */
 146         i_shift_offset   = DIM*shiftidx[iidx];
 147
 148         /* Load limits for loop over neighbors */
 149         j_index_start    = jindex[iidx];
 150         j_index_end      = jindex[iidx+1];
 151
 152         /* Get outer coordinate index */
 153         inr              = iinr[iidx];
 154         i_coord_offset   = DIM*inr;
 155
 156         /* Load i particle coords and add shift vector */
 157         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 158                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 159
 160         fix1             = _mm256_setzero_ps();
 161         fiy1             = _mm256_setzero_ps();
 162         fiz1             = _mm256_setzero_ps();
 163         fix2             = _mm256_setzero_ps();
 164         fiy2             = _mm256_setzero_ps();
 165         fiz2             = _mm256_setzero_ps();
 166         fix3             = _mm256_setzero_ps();
 167         fiy3             = _mm256_setzero_ps();
 168         fiz3             = _mm256_setzero_ps();
 169
 170         /* Reset potential sums */
 171         velecsum         = _mm256_setzero_ps();
 172
 173         /* Start inner kernel loop */
 174         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 175         {
 176
 177             /* Get j neighbor index, and coordinate index */
 178             jnrA             = jjnr[jidx];
 179             jnrB             = jjnr[jidx+1];
 180             jnrC             = jjnr[jidx+2];
 181             jnrD             = jjnr[jidx+3];
 182             jnrE             = jjnr[jidx+4];
 183             jnrF             = jjnr[jidx+5];
 184             jnrG             = jjnr[jidx+6];
 185             jnrH             = jjnr[jidx+7];
 186             j_coord_offsetA  = DIM*jnrA;
 187             j_coord_offsetB  = DIM*jnrB;
 188             j_coord_offsetC  = DIM*jnrC;
 189             j_coord_offsetD  = DIM*jnrD;
 190             j_coord_offsetE  = DIM*jnrE;
 191             j_coord_offsetF  = DIM*jnrF;
 192             j_coord_offsetG  = DIM*jnrG;
 193             j_coord_offsetH  = DIM*jnrH;
 194
 195             /* load j atom coordinates */
 196             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 197                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 198                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 199                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 200                                                  &jx0,&jy0,&jz0);
 201
 202             /* Calculate displacement vector */
 203             dx10             = _mm256_sub_ps(ix1,jx0);
 204             dy10             = _mm256_sub_ps(iy1,jy0);
 205             dz10             = _mm256_sub_ps(iz1,jz0);
 206             dx20             = _mm256_sub_ps(ix2,jx0);
 207             dy20             = _mm256_sub_ps(iy2,jy0);
 208             dz20             = _mm256_sub_ps(iz2,jz0);
 209             dx30             = _mm256_sub_ps(ix3,jx0);
 210             dy30             = _mm256_sub_ps(iy3,jy0);
 211             dz30             = _mm256_sub_ps(iz3,jz0);
 212
 213             /* Calculate squared distance and things based on it */
 214             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 215             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 216             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 217
 218             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 219             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 220             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 221
 222             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 223             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 224             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 225
 226             /* Load parameters for j particles */
 227             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 228                                                                  charge+jnrC+0,charge+jnrD+0,
 229                                                                  charge+jnrE+0,charge+jnrF+0,
 230                                                                  charge+jnrG+0,charge+jnrH+0);
 231
 232             fjx0             = _mm256_setzero_ps();
 233             fjy0             = _mm256_setzero_ps();
 234             fjz0             = _mm256_setzero_ps();
 235
 236             /**************************
 237              * CALCULATE INTERACTIONS *
 238              **************************/
 239
 240             /* Compute parameters for interactions between i and j atoms */
 241             qq10             = _mm256_mul_ps(iq1,jq0);
 242
 243             /* COULOMB ELECTROSTATICS */
 244             velec            = _mm256_mul_ps(qq10,rinv10);
 245             felec            = _mm256_mul_ps(velec,rinvsq10);
 246
 247             /* Update potential sum for this i atom from the interaction with this j atom. */
 248             velecsum         = _mm256_add_ps(velecsum,velec);
 249
 250             fscal            = felec;
 251
 252             /* Calculate temporary vectorial force */
 253             tx               = _mm256_mul_ps(fscal,dx10);
 254             ty               = _mm256_mul_ps(fscal,dy10);
 255             tz               = _mm256_mul_ps(fscal,dz10);
 256
 257             /* Update vectorial force */
 258             fix1             = _mm256_add_ps(fix1,tx);
 259             fiy1             = _mm256_add_ps(fiy1,ty);
 260             fiz1             = _mm256_add_ps(fiz1,tz);
 261
 262             fjx0             = _mm256_add_ps(fjx0,tx);
 263             fjy0             = _mm256_add_ps(fjy0,ty);
 264             fjz0             = _mm256_add_ps(fjz0,tz);
 265
 266             /**************************
 267              * CALCULATE INTERACTIONS *
 268              **************************/
 269
 270             /* Compute parameters for interactions between i and j atoms */
 271             qq20             = _mm256_mul_ps(iq2,jq0);
 272
 273             /* COULOMB ELECTROSTATICS */
 274             velec            = _mm256_mul_ps(qq20,rinv20);
 275             felec            = _mm256_mul_ps(velec,rinvsq20);
 276
 277             /* Update potential sum for this i atom from the interaction with this j atom. */
 278             velecsum         = _mm256_add_ps(velecsum,velec);
 279
 280             fscal            = felec;
 281
 282             /* Calculate temporary vectorial force */
 283             tx               = _mm256_mul_ps(fscal,dx20);
 284             ty               = _mm256_mul_ps(fscal,dy20);
 285             tz               = _mm256_mul_ps(fscal,dz20);
 286
 287             /* Update vectorial force */
 288             fix2             = _mm256_add_ps(fix2,tx);
 289             fiy2             = _mm256_add_ps(fiy2,ty);
 290             fiz2             = _mm256_add_ps(fiz2,tz);
 291
 292             fjx0             = _mm256_add_ps(fjx0,tx);
 293             fjy0             = _mm256_add_ps(fjy0,ty);
 294             fjz0             = _mm256_add_ps(fjz0,tz);
 295
 296             /**************************
 297              * CALCULATE INTERACTIONS *
 298              **************************/
 299
 300             /* Compute parameters for interactions between i and j atoms */
 301             qq30             = _mm256_mul_ps(iq3,jq0);
 302
 303             /* COULOMB ELECTROSTATICS */
 304             velec            = _mm256_mul_ps(qq30,rinv30);
 305             felec            = _mm256_mul_ps(velec,rinvsq30);
 306
 307             /* Update potential sum for this i atom from the interaction with this j atom. */
 308             velecsum         = _mm256_add_ps(velecsum,velec);
 309
 310             fscal            = felec;
 311
 312             /* Calculate temporary vectorial force */
 313             tx               = _mm256_mul_ps(fscal,dx30);
 314             ty               = _mm256_mul_ps(fscal,dy30);
 315             tz               = _mm256_mul_ps(fscal,dz30);
 316
 317             /* Update vectorial force */
 318             fix3             = _mm256_add_ps(fix3,tx);
 319             fiy3             = _mm256_add_ps(fiy3,ty);
 320             fiz3             = _mm256_add_ps(fiz3,tz);
 321
 322             fjx0             = _mm256_add_ps(fjx0,tx);
 323             fjy0             = _mm256_add_ps(fjy0,ty);
 324             fjz0             = _mm256_add_ps(fjz0,tz);
 325
 326             fjptrA             = f+j_coord_offsetA;
 327             fjptrB             = f+j_coord_offsetB;
 328             fjptrC             = f+j_coord_offsetC;
 329             fjptrD             = f+j_coord_offsetD;
 330             fjptrE             = f+j_coord_offsetE;
 331             fjptrF             = f+j_coord_offsetF;
 332             fjptrG             = f+j_coord_offsetG;
 333             fjptrH             = f+j_coord_offsetH;
 334
 335             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 336
 337             /* Inner loop uses 84 flops */
 338         }
 339
 340         if(jidx<j_index_end)
 341         {
 342
 343             /* Get j neighbor index, and coordinate index */
 344             jnrlistA         = jjnr[jidx];
 345             jnrlistB         = jjnr[jidx+1];
 346             jnrlistC         = jjnr[jidx+2];
 347             jnrlistD         = jjnr[jidx+3];
 348             jnrlistE         = jjnr[jidx+4];
 349             jnrlistF         = jjnr[jidx+5];
 350             jnrlistG         = jjnr[jidx+6];
 351             jnrlistH         = jjnr[jidx+7];
 352             /* Sign of each element will be negative for non-real atoms.
 353              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 354              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 355              */
 356             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 357                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 358
 359             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 360             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 361             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 362             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 363             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 364             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 365             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 366             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 367             j_coord_offsetA  = DIM*jnrA;
 368             j_coord_offsetB  = DIM*jnrB;
 369             j_coord_offsetC  = DIM*jnrC;
 370             j_coord_offsetD  = DIM*jnrD;
 371             j_coord_offsetE  = DIM*jnrE;
 372             j_coord_offsetF  = DIM*jnrF;
 373             j_coord_offsetG  = DIM*jnrG;
 374             j_coord_offsetH  = DIM*jnrH;
 375
 376             /* load j atom coordinates */
 377             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 378                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 379                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 380                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 381                                                  &jx0,&jy0,&jz0);
 382
 383             /* Calculate displacement vector */
 384             dx10             = _mm256_sub_ps(ix1,jx0);
 385             dy10             = _mm256_sub_ps(iy1,jy0);
 386             dz10             = _mm256_sub_ps(iz1,jz0);
 387             dx20             = _mm256_sub_ps(ix2,jx0);
 388             dy20             = _mm256_sub_ps(iy2,jy0);
 389             dz20             = _mm256_sub_ps(iz2,jz0);
 390             dx30             = _mm256_sub_ps(ix3,jx0);
 391             dy30             = _mm256_sub_ps(iy3,jy0);
 392             dz30             = _mm256_sub_ps(iz3,jz0);
 393
 394             /* Calculate squared distance and things based on it */
 395             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 396             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 397             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 398
 399             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 400             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 401             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 402
 403             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 404             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 405             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 406
 407             /* Load parameters for j particles */
 408             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 409                                                                  charge+jnrC+0,charge+jnrD+0,
 410                                                                  charge+jnrE+0,charge+jnrF+0,
 411                                                                  charge+jnrG+0,charge+jnrH+0);
 412
 413             fjx0             = _mm256_setzero_ps();
 414             fjy0             = _mm256_setzero_ps();
 415             fjz0             = _mm256_setzero_ps();
 416
 417             /**************************
 418              * CALCULATE INTERACTIONS *
 419              **************************/
 420
 421             /* Compute parameters for interactions between i and j atoms */
 422             qq10             = _mm256_mul_ps(iq1,jq0);
 423
 424             /* COULOMB ELECTROSTATICS */
 425             velec            = _mm256_mul_ps(qq10,rinv10);
 426             felec            = _mm256_mul_ps(velec,rinvsq10);
 427
 428             /* Update potential sum for this i atom from the interaction with this j atom. */
 429             velec            = _mm256_andnot_ps(dummy_mask,velec);
 430             velecsum         = _mm256_add_ps(velecsum,velec);
 431
 432             fscal            = felec;
 433
 434             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 435
 436             /* Calculate temporary vectorial force */
 437             tx               = _mm256_mul_ps(fscal,dx10);
 438             ty               = _mm256_mul_ps(fscal,dy10);
 439             tz               = _mm256_mul_ps(fscal,dz10);
 440
 441             /* Update vectorial force */
 442             fix1             = _mm256_add_ps(fix1,tx);
 443             fiy1             = _mm256_add_ps(fiy1,ty);
 444             fiz1             = _mm256_add_ps(fiz1,tz);
 445
 446             fjx0             = _mm256_add_ps(fjx0,tx);
 447             fjy0             = _mm256_add_ps(fjy0,ty);
 448             fjz0             = _mm256_add_ps(fjz0,tz);
 449
 450             /**************************
 451              * CALCULATE INTERACTIONS *
 452              **************************/
 453
 454             /* Compute parameters for interactions between i and j atoms */
 455             qq20             = _mm256_mul_ps(iq2,jq0);
 456
 457             /* COULOMB ELECTROSTATICS */
 458             velec            = _mm256_mul_ps(qq20,rinv20);
 459             felec            = _mm256_mul_ps(velec,rinvsq20);
 460
 461             /* Update potential sum for this i atom from the interaction with this j atom. */
 462             velec            = _mm256_andnot_ps(dummy_mask,velec);
 463             velecsum         = _mm256_add_ps(velecsum,velec);
 464
 465             fscal            = felec;
 466
 467             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 468
 469             /* Calculate temporary vectorial force */
 470             tx               = _mm256_mul_ps(fscal,dx20);
 471             ty               = _mm256_mul_ps(fscal,dy20);
 472             tz               = _mm256_mul_ps(fscal,dz20);
 473
 474             /* Update vectorial force */
 475             fix2             = _mm256_add_ps(fix2,tx);
 476             fiy2             = _mm256_add_ps(fiy2,ty);
 477             fiz2             = _mm256_add_ps(fiz2,tz);
 478
 479             fjx0             = _mm256_add_ps(fjx0,tx);
 480             fjy0             = _mm256_add_ps(fjy0,ty);
 481             fjz0             = _mm256_add_ps(fjz0,tz);
 482
 483             /**************************
 484              * CALCULATE INTERACTIONS *
 485              **************************/
 486
 487             /* Compute parameters for interactions between i and j atoms */
 488             qq30             = _mm256_mul_ps(iq3,jq0);
 489
 490             /* COULOMB ELECTROSTATICS */
 491             velec            = _mm256_mul_ps(qq30,rinv30);
 492             felec            = _mm256_mul_ps(velec,rinvsq30);
 493
 494             /* Update potential sum for this i atom from the interaction with this j atom. */
 495             velec            = _mm256_andnot_ps(dummy_mask,velec);
 496             velecsum         = _mm256_add_ps(velecsum,velec);
 497
 498             fscal            = felec;
 499
 500             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 501
 502             /* Calculate temporary vectorial force */
 503             tx               = _mm256_mul_ps(fscal,dx30);
 504             ty               = _mm256_mul_ps(fscal,dy30);
 505             tz               = _mm256_mul_ps(fscal,dz30);
 506
 507             /* Update vectorial force */
 508             fix3             = _mm256_add_ps(fix3,tx);
 509             fiy3             = _mm256_add_ps(fiy3,ty);
 510             fiz3             = _mm256_add_ps(fiz3,tz);
 511
 512             fjx0             = _mm256_add_ps(fjx0,tx);
 513             fjy0             = _mm256_add_ps(fjy0,ty);
 514             fjz0             = _mm256_add_ps(fjz0,tz);
 515
 516             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 517             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 518             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 519             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 520             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 521             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 522             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 523             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 524
 525             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 526
 527             /* Inner loop uses 84 flops */
 528         }
 529
 530         /* End of innermost loop */
 531
 532         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 533                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
 534
 535         ggid                        = gid[iidx];
 536         /* Update potential energies */
 537         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 538
 539         /* Increment number of inner iterations */
 540         inneriter                  += j_index_end - j_index_start;
 541
 542         /* Outer loop uses 19 flops */
 543     }
 544
 545     /* Increment number of outer iterations */
 546     outeriter        += nri;
 547
 548     /* Update outer/inner flops */
 549
 550     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*84);
 551 }
 552 /*
 553  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW4P1_F_avx_256_single
 554  * Electrostatics interaction: Coulomb
 555  * VdW interaction:            None
 556  * Geometry:                   Water4-Particle
 557  * Calculate force/pot:        Force
 558  */
 559 void
 560 nb_kernel_ElecCoul_VdwNone_GeomW4P1_F_avx_256_single
 561                     (t_nblist                    * gmx_restrict       nlist,
 562                      rvec                        * gmx_restrict          xx,
 563                      rvec                        * gmx_restrict          ff,
 564                      t_forcerec                  * gmx_restrict          fr,
 565                      t_mdatoms                   * gmx_restrict     mdatoms,
 566                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 567                      t_nrnb                      * gmx_restrict        nrnb)
 568 {
 569     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 570      * just 0 for non-waters.
 571      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 572      * jnr indices corresponding to data put in the four positions in the SIMD register.
 573      */
 574     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 575     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 576     int              jnrA,jnrB,jnrC,jnrD;
 577     int              jnrE,jnrF,jnrG,jnrH;
 578     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 579     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 580     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 581     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 582     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 583     real             rcutoff_scalar;
 584     real             *shiftvec,*fshift,*x,*f;
 585     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 586     real             scratch[4*DIM];
 587     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 588     real *           vdwioffsetptr1;
 589     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 590     real *           vdwioffsetptr2;
 591     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 592     real *           vdwioffsetptr3;
 593     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 594     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 595     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 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           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 599     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 600     real             *charge;
 601     __m256           dummy_mask,cutoff_mask;
 602     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 603     __m256           one     = _mm256_set1_ps(1.0);
 604     __m256           two     = _mm256_set1_ps(2.0);
 605     x                = xx[0];
 606     f                = ff[0];
 607
 608     nri              = nlist->nri;
 609     iinr             = nlist->iinr;
 610     jindex           = nlist->jindex;
 611     jjnr             = nlist->jjnr;
 612     shiftidx         = nlist->shift;
 613     gid              = nlist->gid;
 614     shiftvec         = fr->shift_vec[0];
 615     fshift           = fr->fshift[0];
 616     facel            = _mm256_set1_ps(fr->epsfac);
 617     charge           = mdatoms->chargeA;
 618
 619     /* Setup water-specific parameters */
 620     inr              = nlist->iinr[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     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 624
 625     /* Avoid stupid compiler warnings */
 626     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 627     j_coord_offsetA = 0;
 628     j_coord_offsetB = 0;
 629     j_coord_offsetC = 0;
 630     j_coord_offsetD = 0;
 631     j_coord_offsetE = 0;
 632     j_coord_offsetF = 0;
 633     j_coord_offsetG = 0;
 634     j_coord_offsetH = 0;
 635
 636     outeriter        = 0;
 637     inneriter        = 0;
 638
 639     for(iidx=0;iidx<4*DIM;iidx++)
 640     {
 641         scratch[iidx] = 0.0;
 642     }
 643
 644     /* Start outer loop over neighborlists */
 645     for(iidx=0; iidx<nri; iidx++)
 646     {
 647         /* Load shift vector for this list */
 648         i_shift_offset   = DIM*shiftidx[iidx];
 649
 650         /* Load limits for loop over neighbors */
 651         j_index_start    = jindex[iidx];
 652         j_index_end      = jindex[iidx+1];
 653
 654         /* Get outer coordinate index */
 655         inr              = iinr[iidx];
 656         i_coord_offset   = DIM*inr;
 657
 658         /* Load i particle coords and add shift vector */
 659         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 660                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 661
 662         fix1             = _mm256_setzero_ps();
 663         fiy1             = _mm256_setzero_ps();
 664         fiz1             = _mm256_setzero_ps();
 665         fix2             = _mm256_setzero_ps();
 666         fiy2             = _mm256_setzero_ps();
 667         fiz2             = _mm256_setzero_ps();
 668         fix3             = _mm256_setzero_ps();
 669         fiy3             = _mm256_setzero_ps();
 670         fiz3             = _mm256_setzero_ps();
 671
 672         /* Start inner kernel loop */
 673         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 674         {
 675
 676             /* Get j neighbor index, and coordinate index */
 677             jnrA             = jjnr[jidx];
 678             jnrB             = jjnr[jidx+1];
 679             jnrC             = jjnr[jidx+2];
 680             jnrD             = jjnr[jidx+3];
 681             jnrE             = jjnr[jidx+4];
 682             jnrF             = jjnr[jidx+5];
 683             jnrG             = jjnr[jidx+6];
 684             jnrH             = jjnr[jidx+7];
 685             j_coord_offsetA  = DIM*jnrA;
 686             j_coord_offsetB  = DIM*jnrB;
 687             j_coord_offsetC  = DIM*jnrC;
 688             j_coord_offsetD  = DIM*jnrD;
 689             j_coord_offsetE  = DIM*jnrE;
 690             j_coord_offsetF  = DIM*jnrF;
 691             j_coord_offsetG  = DIM*jnrG;
 692             j_coord_offsetH  = DIM*jnrH;
 693
 694             /* load j atom coordinates */
 695             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 696                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 697                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 698                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 699                                                  &jx0,&jy0,&jz0);
 700
 701             /* Calculate displacement vector */
 702             dx10             = _mm256_sub_ps(ix1,jx0);
 703             dy10             = _mm256_sub_ps(iy1,jy0);
 704             dz10             = _mm256_sub_ps(iz1,jz0);
 705             dx20             = _mm256_sub_ps(ix2,jx0);
 706             dy20             = _mm256_sub_ps(iy2,jy0);
 707             dz20             = _mm256_sub_ps(iz2,jz0);
 708             dx30             = _mm256_sub_ps(ix3,jx0);
 709             dy30             = _mm256_sub_ps(iy3,jy0);
 710             dz30             = _mm256_sub_ps(iz3,jz0);
 711
 712             /* Calculate squared distance and things based on it */
 713             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 714             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 715             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 716
 717             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 718             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 719             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 720
 721             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 722             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 723             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 724
 725             /* Load parameters for j particles */
 726             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 727                                                                  charge+jnrC+0,charge+jnrD+0,
 728                                                                  charge+jnrE+0,charge+jnrF+0,
 729                                                                  charge+jnrG+0,charge+jnrH+0);
 730
 731             fjx0             = _mm256_setzero_ps();
 732             fjy0             = _mm256_setzero_ps();
 733             fjz0             = _mm256_setzero_ps();
 734
 735             /**************************
 736              * CALCULATE INTERACTIONS *
 737              **************************/
 738
 739             /* Compute parameters for interactions between i and j atoms */
 740             qq10             = _mm256_mul_ps(iq1,jq0);
 741
 742             /* COULOMB ELECTROSTATICS */
 743             velec            = _mm256_mul_ps(qq10,rinv10);
 744             felec            = _mm256_mul_ps(velec,rinvsq10);
 745
 746             fscal            = felec;
 747
 748             /* Calculate temporary vectorial force */
 749             tx               = _mm256_mul_ps(fscal,dx10);
 750             ty               = _mm256_mul_ps(fscal,dy10);
 751             tz               = _mm256_mul_ps(fscal,dz10);
 752
 753             /* Update vectorial force */
 754             fix1             = _mm256_add_ps(fix1,tx);
 755             fiy1             = _mm256_add_ps(fiy1,ty);
 756             fiz1             = _mm256_add_ps(fiz1,tz);
 757
 758             fjx0             = _mm256_add_ps(fjx0,tx);
 759             fjy0             = _mm256_add_ps(fjy0,ty);
 760             fjz0             = _mm256_add_ps(fjz0,tz);
 761
 762             /**************************
 763              * CALCULATE INTERACTIONS *
 764              **************************/
 765
 766             /* Compute parameters for interactions between i and j atoms */
 767             qq20             = _mm256_mul_ps(iq2,jq0);
 768
 769             /* COULOMB ELECTROSTATICS */
 770             velec            = _mm256_mul_ps(qq20,rinv20);
 771             felec            = _mm256_mul_ps(velec,rinvsq20);
 772
 773             fscal            = felec;
 774
 775             /* Calculate temporary vectorial force */
 776             tx               = _mm256_mul_ps(fscal,dx20);
 777             ty               = _mm256_mul_ps(fscal,dy20);
 778             tz               = _mm256_mul_ps(fscal,dz20);
 779
 780             /* Update vectorial force */
 781             fix2             = _mm256_add_ps(fix2,tx);
 782             fiy2             = _mm256_add_ps(fiy2,ty);
 783             fiz2             = _mm256_add_ps(fiz2,tz);
 784
 785             fjx0             = _mm256_add_ps(fjx0,tx);
 786             fjy0             = _mm256_add_ps(fjy0,ty);
 787             fjz0             = _mm256_add_ps(fjz0,tz);
 788
 789             /**************************
 790              * CALCULATE INTERACTIONS *
 791              **************************/
 792
 793             /* Compute parameters for interactions between i and j atoms */
 794             qq30             = _mm256_mul_ps(iq3,jq0);
 795
 796             /* COULOMB ELECTROSTATICS */
 797             velec            = _mm256_mul_ps(qq30,rinv30);
 798             felec            = _mm256_mul_ps(velec,rinvsq30);
 799
 800             fscal            = felec;
 801
 802             /* Calculate temporary vectorial force */
 803             tx               = _mm256_mul_ps(fscal,dx30);
 804             ty               = _mm256_mul_ps(fscal,dy30);
 805             tz               = _mm256_mul_ps(fscal,dz30);
 806
 807             /* Update vectorial force */
 808             fix3             = _mm256_add_ps(fix3,tx);
 809             fiy3             = _mm256_add_ps(fiy3,ty);
 810             fiz3             = _mm256_add_ps(fiz3,tz);
 811
 812             fjx0             = _mm256_add_ps(fjx0,tx);
 813             fjy0             = _mm256_add_ps(fjy0,ty);
 814             fjz0             = _mm256_add_ps(fjz0,tz);
 815
 816             fjptrA             = f+j_coord_offsetA;
 817             fjptrB             = f+j_coord_offsetB;
 818             fjptrC             = f+j_coord_offsetC;
 819             fjptrD             = f+j_coord_offsetD;
 820             fjptrE             = f+j_coord_offsetE;
 821             fjptrF             = f+j_coord_offsetF;
 822             fjptrG             = f+j_coord_offsetG;
 823             fjptrH             = f+j_coord_offsetH;
 824
 825             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 826
 827             /* Inner loop uses 81 flops */
 828         }
 829
 830         if(jidx<j_index_end)
 831         {
 832
 833             /* Get j neighbor index, and coordinate index */
 834             jnrlistA         = jjnr[jidx];
 835             jnrlistB         = jjnr[jidx+1];
 836             jnrlistC         = jjnr[jidx+2];
 837             jnrlistD         = jjnr[jidx+3];
 838             jnrlistE         = jjnr[jidx+4];
 839             jnrlistF         = jjnr[jidx+5];
 840             jnrlistG         = jjnr[jidx+6];
 841             jnrlistH         = jjnr[jidx+7];
 842             /* Sign of each element will be negative for non-real atoms.
 843              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 844              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 845              */
 846             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 847                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 848
 849             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 850             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 851             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 852             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 853             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 854             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 855             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 856             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 857             j_coord_offsetA  = DIM*jnrA;
 858             j_coord_offsetB  = DIM*jnrB;
 859             j_coord_offsetC  = DIM*jnrC;
 860             j_coord_offsetD  = DIM*jnrD;
 861             j_coord_offsetE  = DIM*jnrE;
 862             j_coord_offsetF  = DIM*jnrF;
 863             j_coord_offsetG  = DIM*jnrG;
 864             j_coord_offsetH  = DIM*jnrH;
 865
 866             /* load j atom coordinates */
 867             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 868                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 869                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 870                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 871                                                  &jx0,&jy0,&jz0);
 872
 873             /* Calculate displacement vector */
 874             dx10             = _mm256_sub_ps(ix1,jx0);
 875             dy10             = _mm256_sub_ps(iy1,jy0);
 876             dz10             = _mm256_sub_ps(iz1,jz0);
 877             dx20             = _mm256_sub_ps(ix2,jx0);
 878             dy20             = _mm256_sub_ps(iy2,jy0);
 879             dz20             = _mm256_sub_ps(iz2,jz0);
 880             dx30             = _mm256_sub_ps(ix3,jx0);
 881             dy30             = _mm256_sub_ps(iy3,jy0);
 882             dz30             = _mm256_sub_ps(iz3,jz0);
 883
 884             /* Calculate squared distance and things based on it */
 885             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 886             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 887             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 888
 889             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 890             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 891             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 892
 893             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 894             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 895             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 896
 897             /* Load parameters for j particles */
 898             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 899                                                                  charge+jnrC+0,charge+jnrD+0,
 900                                                                  charge+jnrE+0,charge+jnrF+0,
 901                                                                  charge+jnrG+0,charge+jnrH+0);
 902
 903             fjx0             = _mm256_setzero_ps();
 904             fjy0             = _mm256_setzero_ps();
 905             fjz0             = _mm256_setzero_ps();
 906
 907             /**************************
 908              * CALCULATE INTERACTIONS *
 909              **************************/
 910
 911             /* Compute parameters for interactions between i and j atoms */
 912             qq10             = _mm256_mul_ps(iq1,jq0);
 913
 914             /* COULOMB ELECTROSTATICS */
 915             velec            = _mm256_mul_ps(qq10,rinv10);
 916             felec            = _mm256_mul_ps(velec,rinvsq10);
 917
 918             fscal            = felec;
 919
 920             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 921
 922             /* Calculate temporary vectorial force */
 923             tx               = _mm256_mul_ps(fscal,dx10);
 924             ty               = _mm256_mul_ps(fscal,dy10);
 925             tz               = _mm256_mul_ps(fscal,dz10);
 926
 927             /* Update vectorial force */
 928             fix1             = _mm256_add_ps(fix1,tx);
 929             fiy1             = _mm256_add_ps(fiy1,ty);
 930             fiz1             = _mm256_add_ps(fiz1,tz);
 931
 932             fjx0             = _mm256_add_ps(fjx0,tx);
 933             fjy0             = _mm256_add_ps(fjy0,ty);
 934             fjz0             = _mm256_add_ps(fjz0,tz);
 935
 936             /**************************
 937              * CALCULATE INTERACTIONS *
 938              **************************/
 939
 940             /* Compute parameters for interactions between i and j atoms */
 941             qq20             = _mm256_mul_ps(iq2,jq0);
 942
 943             /* COULOMB ELECTROSTATICS */
 944             velec            = _mm256_mul_ps(qq20,rinv20);
 945             felec            = _mm256_mul_ps(velec,rinvsq20);
 946
 947             fscal            = felec;
 948
 949             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 950
 951             /* Calculate temporary vectorial force */
 952             tx               = _mm256_mul_ps(fscal,dx20);
 953             ty               = _mm256_mul_ps(fscal,dy20);
 954             tz               = _mm256_mul_ps(fscal,dz20);
 955
 956             /* Update vectorial force */
 957             fix2             = _mm256_add_ps(fix2,tx);
 958             fiy2             = _mm256_add_ps(fiy2,ty);
 959             fiz2             = _mm256_add_ps(fiz2,tz);
 960
 961             fjx0             = _mm256_add_ps(fjx0,tx);
 962             fjy0             = _mm256_add_ps(fjy0,ty);
 963             fjz0             = _mm256_add_ps(fjz0,tz);
 964
 965             /**************************
 966              * CALCULATE INTERACTIONS *
 967              **************************/
 968
 969             /* Compute parameters for interactions between i and j atoms */
 970             qq30             = _mm256_mul_ps(iq3,jq0);
 971
 972             /* COULOMB ELECTROSTATICS */
 973             velec            = _mm256_mul_ps(qq30,rinv30);
 974             felec            = _mm256_mul_ps(velec,rinvsq30);
 975
 976             fscal            = felec;
 977
 978             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 979
 980             /* Calculate temporary vectorial force */
 981             tx               = _mm256_mul_ps(fscal,dx30);
 982             ty               = _mm256_mul_ps(fscal,dy30);
 983             tz               = _mm256_mul_ps(fscal,dz30);
 984
 985             /* Update vectorial force */
 986             fix3             = _mm256_add_ps(fix3,tx);
 987             fiy3             = _mm256_add_ps(fiy3,ty);
 988             fiz3             = _mm256_add_ps(fiz3,tz);
 989
 990             fjx0             = _mm256_add_ps(fjx0,tx);
 991             fjy0             = _mm256_add_ps(fjy0,ty);
 992             fjz0             = _mm256_add_ps(fjz0,tz);
 993
 994             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 995             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 996             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 997             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 998             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 999             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1000             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1001             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1002
1003             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1004
1005             /* Inner loop uses 81 flops */
1006         }
1007
1008         /* End of innermost loop */
1009
1010         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1011                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1012
1013         /* Increment number of inner iterations */
1014         inneriter                  += j_index_end - j_index_start;
1015
1016         /* Outer loop uses 18 flops */
1017     }
1018
1019     /* Increment number of outer iterations */
1020     outeriter        += nri;
1021
1022     /* Update outer/inner flops */
1023
1024     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*81);
1025 }