src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRF_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_ElecRF_VdwNone_GeomW4P1_VF_avx_256_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            None
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRF_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     krf              = _mm256_set1_ps(fr->ic->k_rf);
 117     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 118     crf              = _mm256_set1_ps(fr->ic->c_rf);
 119
 120     /* Setup water-specific parameters */
 121     inr              = nlist->iinr[0];
 122     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 123     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 124     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 125
 126     /* Avoid stupid compiler warnings */
 127     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 128     j_coord_offsetA = 0;
 129     j_coord_offsetB = 0;
 130     j_coord_offsetC = 0;
 131     j_coord_offsetD = 0;
 132     j_coord_offsetE = 0;
 133     j_coord_offsetF = 0;
 134     j_coord_offsetG = 0;
 135     j_coord_offsetH = 0;
 136
 137     outeriter        = 0;
 138     inneriter        = 0;
 139
 140     for(iidx=0;iidx<4*DIM;iidx++)
 141     {
 142         scratch[iidx] = 0.0;
 143     }
 144
 145     /* Start outer loop over neighborlists */
 146     for(iidx=0; iidx<nri; iidx++)
 147     {
 148         /* Load shift vector for this list */
 149         i_shift_offset   = DIM*shiftidx[iidx];
 150
 151         /* Load limits for loop over neighbors */
 152         j_index_start    = jindex[iidx];
 153         j_index_end      = jindex[iidx+1];
 154
 155         /* Get outer coordinate index */
 156         inr              = iinr[iidx];
 157         i_coord_offset   = DIM*inr;
 158
 159         /* Load i particle coords and add shift vector */
 160         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 161                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 162
 163         fix1             = _mm256_setzero_ps();
 164         fiy1             = _mm256_setzero_ps();
 165         fiz1             = _mm256_setzero_ps();
 166         fix2             = _mm256_setzero_ps();
 167         fiy2             = _mm256_setzero_ps();
 168         fiz2             = _mm256_setzero_ps();
 169         fix3             = _mm256_setzero_ps();
 170         fiy3             = _mm256_setzero_ps();
 171         fiz3             = _mm256_setzero_ps();
 172
 173         /* Reset potential sums */
 174         velecsum         = _mm256_setzero_ps();
 175
 176         /* Start inner kernel loop */
 177         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 178         {
 179
 180             /* Get j neighbor index, and coordinate index */
 181             jnrA             = jjnr[jidx];
 182             jnrB             = jjnr[jidx+1];
 183             jnrC             = jjnr[jidx+2];
 184             jnrD             = jjnr[jidx+3];
 185             jnrE             = jjnr[jidx+4];
 186             jnrF             = jjnr[jidx+5];
 187             jnrG             = jjnr[jidx+6];
 188             jnrH             = jjnr[jidx+7];
 189             j_coord_offsetA  = DIM*jnrA;
 190             j_coord_offsetB  = DIM*jnrB;
 191             j_coord_offsetC  = DIM*jnrC;
 192             j_coord_offsetD  = DIM*jnrD;
 193             j_coord_offsetE  = DIM*jnrE;
 194             j_coord_offsetF  = DIM*jnrF;
 195             j_coord_offsetG  = DIM*jnrG;
 196             j_coord_offsetH  = DIM*jnrH;
 197
 198             /* load j atom coordinates */
 199             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 200                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 201                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 202                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 203                                                  &jx0,&jy0,&jz0);
 204
 205             /* Calculate displacement vector */
 206             dx10             = _mm256_sub_ps(ix1,jx0);
 207             dy10             = _mm256_sub_ps(iy1,jy0);
 208             dz10             = _mm256_sub_ps(iz1,jz0);
 209             dx20             = _mm256_sub_ps(ix2,jx0);
 210             dy20             = _mm256_sub_ps(iy2,jy0);
 211             dz20             = _mm256_sub_ps(iz2,jz0);
 212             dx30             = _mm256_sub_ps(ix3,jx0);
 213             dy30             = _mm256_sub_ps(iy3,jy0);
 214             dz30             = _mm256_sub_ps(iz3,jz0);
 215
 216             /* Calculate squared distance and things based on it */
 217             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 218             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 219             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 220
 221             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 222             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 223             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 224
 225             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 226             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 227             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 228
 229             /* Load parameters for j particles */
 230             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 231                                                                  charge+jnrC+0,charge+jnrD+0,
 232                                                                  charge+jnrE+0,charge+jnrF+0,
 233                                                                  charge+jnrG+0,charge+jnrH+0);
 234
 235             fjx0             = _mm256_setzero_ps();
 236             fjy0             = _mm256_setzero_ps();
 237             fjz0             = _mm256_setzero_ps();
 238
 239             /**************************
 240              * CALCULATE INTERACTIONS *
 241              **************************/
 242
 243             /* Compute parameters for interactions between i and j atoms */
 244             qq10             = _mm256_mul_ps(iq1,jq0);
 245
 246             /* REACTION-FIELD ELECTROSTATICS */
 247             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 248             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 249
 250             /* Update potential sum for this i atom from the interaction with this j atom. */
 251             velecsum         = _mm256_add_ps(velecsum,velec);
 252
 253             fscal            = felec;
 254
 255             /* Calculate temporary vectorial force */
 256             tx               = _mm256_mul_ps(fscal,dx10);
 257             ty               = _mm256_mul_ps(fscal,dy10);
 258             tz               = _mm256_mul_ps(fscal,dz10);
 259
 260             /* Update vectorial force */
 261             fix1             = _mm256_add_ps(fix1,tx);
 262             fiy1             = _mm256_add_ps(fiy1,ty);
 263             fiz1             = _mm256_add_ps(fiz1,tz);
 264
 265             fjx0             = _mm256_add_ps(fjx0,tx);
 266             fjy0             = _mm256_add_ps(fjy0,ty);
 267             fjz0             = _mm256_add_ps(fjz0,tz);
 268
 269             /**************************
 270              * CALCULATE INTERACTIONS *
 271              **************************/
 272
 273             /* Compute parameters for interactions between i and j atoms */
 274             qq20             = _mm256_mul_ps(iq2,jq0);
 275
 276             /* REACTION-FIELD ELECTROSTATICS */
 277             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 278             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 279
 280             /* Update potential sum for this i atom from the interaction with this j atom. */
 281             velecsum         = _mm256_add_ps(velecsum,velec);
 282
 283             fscal            = felec;
 284
 285             /* Calculate temporary vectorial force */
 286             tx               = _mm256_mul_ps(fscal,dx20);
 287             ty               = _mm256_mul_ps(fscal,dy20);
 288             tz               = _mm256_mul_ps(fscal,dz20);
 289
 290             /* Update vectorial force */
 291             fix2             = _mm256_add_ps(fix2,tx);
 292             fiy2             = _mm256_add_ps(fiy2,ty);
 293             fiz2             = _mm256_add_ps(fiz2,tz);
 294
 295             fjx0             = _mm256_add_ps(fjx0,tx);
 296             fjy0             = _mm256_add_ps(fjy0,ty);
 297             fjz0             = _mm256_add_ps(fjz0,tz);
 298
 299             /**************************
 300              * CALCULATE INTERACTIONS *
 301              **************************/
 302
 303             /* Compute parameters for interactions between i and j atoms */
 304             qq30             = _mm256_mul_ps(iq3,jq0);
 305
 306             /* REACTION-FIELD ELECTROSTATICS */
 307             velec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
 308             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 309
 310             /* Update potential sum for this i atom from the interaction with this j atom. */
 311             velecsum         = _mm256_add_ps(velecsum,velec);
 312
 313             fscal            = felec;
 314
 315             /* Calculate temporary vectorial force */
 316             tx               = _mm256_mul_ps(fscal,dx30);
 317             ty               = _mm256_mul_ps(fscal,dy30);
 318             tz               = _mm256_mul_ps(fscal,dz30);
 319
 320             /* Update vectorial force */
 321             fix3             = _mm256_add_ps(fix3,tx);
 322             fiy3             = _mm256_add_ps(fiy3,ty);
 323             fiz3             = _mm256_add_ps(fiz3,tz);
 324
 325             fjx0             = _mm256_add_ps(fjx0,tx);
 326             fjy0             = _mm256_add_ps(fjy0,ty);
 327             fjz0             = _mm256_add_ps(fjz0,tz);
 328
 329             fjptrA             = f+j_coord_offsetA;
 330             fjptrB             = f+j_coord_offsetB;
 331             fjptrC             = f+j_coord_offsetC;
 332             fjptrD             = f+j_coord_offsetD;
 333             fjptrE             = f+j_coord_offsetE;
 334             fjptrF             = f+j_coord_offsetF;
 335             fjptrG             = f+j_coord_offsetG;
 336             fjptrH             = f+j_coord_offsetH;
 337
 338             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 339
 340             /* Inner loop uses 99 flops */
 341         }
 342
 343         if(jidx<j_index_end)
 344         {
 345
 346             /* Get j neighbor index, and coordinate index */
 347             jnrlistA         = jjnr[jidx];
 348             jnrlistB         = jjnr[jidx+1];
 349             jnrlistC         = jjnr[jidx+2];
 350             jnrlistD         = jjnr[jidx+3];
 351             jnrlistE         = jjnr[jidx+4];
 352             jnrlistF         = jjnr[jidx+5];
 353             jnrlistG         = jjnr[jidx+6];
 354             jnrlistH         = jjnr[jidx+7];
 355             /* Sign of each element will be negative for non-real atoms.
 356              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 357              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 358              */
 359             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 360                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 361
 362             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 363             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 364             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 365             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 366             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 367             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 368             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 369             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 370             j_coord_offsetA  = DIM*jnrA;
 371             j_coord_offsetB  = DIM*jnrB;
 372             j_coord_offsetC  = DIM*jnrC;
 373             j_coord_offsetD  = DIM*jnrD;
 374             j_coord_offsetE  = DIM*jnrE;
 375             j_coord_offsetF  = DIM*jnrF;
 376             j_coord_offsetG  = DIM*jnrG;
 377             j_coord_offsetH  = DIM*jnrH;
 378
 379             /* load j atom coordinates */
 380             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 381                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 382                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 383                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 384                                                  &jx0,&jy0,&jz0);
 385
 386             /* Calculate displacement vector */
 387             dx10             = _mm256_sub_ps(ix1,jx0);
 388             dy10             = _mm256_sub_ps(iy1,jy0);
 389             dz10             = _mm256_sub_ps(iz1,jz0);
 390             dx20             = _mm256_sub_ps(ix2,jx0);
 391             dy20             = _mm256_sub_ps(iy2,jy0);
 392             dz20             = _mm256_sub_ps(iz2,jz0);
 393             dx30             = _mm256_sub_ps(ix3,jx0);
 394             dy30             = _mm256_sub_ps(iy3,jy0);
 395             dz30             = _mm256_sub_ps(iz3,jz0);
 396
 397             /* Calculate squared distance and things based on it */
 398             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 399             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 400             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 401
 402             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 403             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 404             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 405
 406             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 407             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 408             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 409
 410             /* Load parameters for j particles */
 411             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 412                                                                  charge+jnrC+0,charge+jnrD+0,
 413                                                                  charge+jnrE+0,charge+jnrF+0,
 414                                                                  charge+jnrG+0,charge+jnrH+0);
 415
 416             fjx0             = _mm256_setzero_ps();
 417             fjy0             = _mm256_setzero_ps();
 418             fjz0             = _mm256_setzero_ps();
 419
 420             /**************************
 421              * CALCULATE INTERACTIONS *
 422              **************************/
 423
 424             /* Compute parameters for interactions between i and j atoms */
 425             qq10             = _mm256_mul_ps(iq1,jq0);
 426
 427             /* REACTION-FIELD ELECTROSTATICS */
 428             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 429             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 430
 431             /* Update potential sum for this i atom from the interaction with this j atom. */
 432             velec            = _mm256_andnot_ps(dummy_mask,velec);
 433             velecsum         = _mm256_add_ps(velecsum,velec);
 434
 435             fscal            = felec;
 436
 437             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 438
 439             /* Calculate temporary vectorial force */
 440             tx               = _mm256_mul_ps(fscal,dx10);
 441             ty               = _mm256_mul_ps(fscal,dy10);
 442             tz               = _mm256_mul_ps(fscal,dz10);
 443
 444             /* Update vectorial force */
 445             fix1             = _mm256_add_ps(fix1,tx);
 446             fiy1             = _mm256_add_ps(fiy1,ty);
 447             fiz1             = _mm256_add_ps(fiz1,tz);
 448
 449             fjx0             = _mm256_add_ps(fjx0,tx);
 450             fjy0             = _mm256_add_ps(fjy0,ty);
 451             fjz0             = _mm256_add_ps(fjz0,tz);
 452
 453             /**************************
 454              * CALCULATE INTERACTIONS *
 455              **************************/
 456
 457             /* Compute parameters for interactions between i and j atoms */
 458             qq20             = _mm256_mul_ps(iq2,jq0);
 459
 460             /* REACTION-FIELD ELECTROSTATICS */
 461             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 462             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 463
 464             /* Update potential sum for this i atom from the interaction with this j atom. */
 465             velec            = _mm256_andnot_ps(dummy_mask,velec);
 466             velecsum         = _mm256_add_ps(velecsum,velec);
 467
 468             fscal            = felec;
 469
 470             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 471
 472             /* Calculate temporary vectorial force */
 473             tx               = _mm256_mul_ps(fscal,dx20);
 474             ty               = _mm256_mul_ps(fscal,dy20);
 475             tz               = _mm256_mul_ps(fscal,dz20);
 476
 477             /* Update vectorial force */
 478             fix2             = _mm256_add_ps(fix2,tx);
 479             fiy2             = _mm256_add_ps(fiy2,ty);
 480             fiz2             = _mm256_add_ps(fiz2,tz);
 481
 482             fjx0             = _mm256_add_ps(fjx0,tx);
 483             fjy0             = _mm256_add_ps(fjy0,ty);
 484             fjz0             = _mm256_add_ps(fjz0,tz);
 485
 486             /**************************
 487              * CALCULATE INTERACTIONS *
 488              **************************/
 489
 490             /* Compute parameters for interactions between i and j atoms */
 491             qq30             = _mm256_mul_ps(iq3,jq0);
 492
 493             /* REACTION-FIELD ELECTROSTATICS */
 494             velec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
 495             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 496
 497             /* Update potential sum for this i atom from the interaction with this j atom. */
 498             velec            = _mm256_andnot_ps(dummy_mask,velec);
 499             velecsum         = _mm256_add_ps(velecsum,velec);
 500
 501             fscal            = felec;
 502
 503             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 504
 505             /* Calculate temporary vectorial force */
 506             tx               = _mm256_mul_ps(fscal,dx30);
 507             ty               = _mm256_mul_ps(fscal,dy30);
 508             tz               = _mm256_mul_ps(fscal,dz30);
 509
 510             /* Update vectorial force */
 511             fix3             = _mm256_add_ps(fix3,tx);
 512             fiy3             = _mm256_add_ps(fiy3,ty);
 513             fiz3             = _mm256_add_ps(fiz3,tz);
 514
 515             fjx0             = _mm256_add_ps(fjx0,tx);
 516             fjy0             = _mm256_add_ps(fjy0,ty);
 517             fjz0             = _mm256_add_ps(fjz0,tz);
 518
 519             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 520             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 521             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 522             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 523             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 524             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 525             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 526             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 527
 528             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 529
 530             /* Inner loop uses 99 flops */
 531         }
 532
 533         /* End of innermost loop */
 534
 535         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 536                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
 537
 538         ggid                        = gid[iidx];
 539         /* Update potential energies */
 540         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 541
 542         /* Increment number of inner iterations */
 543         inneriter                  += j_index_end - j_index_start;
 544
 545         /* Outer loop uses 19 flops */
 546     }
 547
 548     /* Increment number of outer iterations */
 549     outeriter        += nri;
 550
 551     /* Update outer/inner flops */
 552
 553     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*99);
 554 }
 555 /*
 556  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW4P1_F_avx_256_single
 557  * Electrostatics interaction: ReactionField
 558  * VdW interaction:            None
 559  * Geometry:                   Water4-Particle
 560  * Calculate force/pot:        Force
 561  */
 562 void
 563 nb_kernel_ElecRF_VdwNone_GeomW4P1_F_avx_256_single
 564                     (t_nblist                    * gmx_restrict       nlist,
 565                      rvec                        * gmx_restrict          xx,
 566                      rvec                        * gmx_restrict          ff,
 567                      t_forcerec                  * gmx_restrict          fr,
 568                      t_mdatoms                   * gmx_restrict     mdatoms,
 569                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 570                      t_nrnb                      * gmx_restrict        nrnb)
 571 {
 572     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 573      * just 0 for non-waters.
 574      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 575      * jnr indices corresponding to data put in the four positions in the SIMD register.
 576      */
 577     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 578     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 579     int              jnrA,jnrB,jnrC,jnrD;
 580     int              jnrE,jnrF,jnrG,jnrH;
 581     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 582     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 583     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 584     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 585     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 586     real             rcutoff_scalar;
 587     real             *shiftvec,*fshift,*x,*f;
 588     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 589     real             scratch[4*DIM];
 590     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 591     real *           vdwioffsetptr1;
 592     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 593     real *           vdwioffsetptr2;
 594     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 595     real *           vdwioffsetptr3;
 596     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 597     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 598     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 599     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 600     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 601     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 602     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 603     real             *charge;
 604     __m256           dummy_mask,cutoff_mask;
 605     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 606     __m256           one     = _mm256_set1_ps(1.0);
 607     __m256           two     = _mm256_set1_ps(2.0);
 608     x                = xx[0];
 609     f                = ff[0];
 610
 611     nri              = nlist->nri;
 612     iinr             = nlist->iinr;
 613     jindex           = nlist->jindex;
 614     jjnr             = nlist->jjnr;
 615     shiftidx         = nlist->shift;
 616     gid              = nlist->gid;
 617     shiftvec         = fr->shift_vec[0];
 618     fshift           = fr->fshift[0];
 619     facel            = _mm256_set1_ps(fr->epsfac);
 620     charge           = mdatoms->chargeA;
 621     krf              = _mm256_set1_ps(fr->ic->k_rf);
 622     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 623     crf              = _mm256_set1_ps(fr->ic->c_rf);
 624
 625     /* Setup water-specific parameters */
 626     inr              = nlist->iinr[0];
 627     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 628     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 629     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 630
 631     /* Avoid stupid compiler warnings */
 632     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 633     j_coord_offsetA = 0;
 634     j_coord_offsetB = 0;
 635     j_coord_offsetC = 0;
 636     j_coord_offsetD = 0;
 637     j_coord_offsetE = 0;
 638     j_coord_offsetF = 0;
 639     j_coord_offsetG = 0;
 640     j_coord_offsetH = 0;
 641
 642     outeriter        = 0;
 643     inneriter        = 0;
 644
 645     for(iidx=0;iidx<4*DIM;iidx++)
 646     {
 647         scratch[iidx] = 0.0;
 648     }
 649
 650     /* Start outer loop over neighborlists */
 651     for(iidx=0; iidx<nri; iidx++)
 652     {
 653         /* Load shift vector for this list */
 654         i_shift_offset   = DIM*shiftidx[iidx];
 655
 656         /* Load limits for loop over neighbors */
 657         j_index_start    = jindex[iidx];
 658         j_index_end      = jindex[iidx+1];
 659
 660         /* Get outer coordinate index */
 661         inr              = iinr[iidx];
 662         i_coord_offset   = DIM*inr;
 663
 664         /* Load i particle coords and add shift vector */
 665         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 666                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 667
 668         fix1             = _mm256_setzero_ps();
 669         fiy1             = _mm256_setzero_ps();
 670         fiz1             = _mm256_setzero_ps();
 671         fix2             = _mm256_setzero_ps();
 672         fiy2             = _mm256_setzero_ps();
 673         fiz2             = _mm256_setzero_ps();
 674         fix3             = _mm256_setzero_ps();
 675         fiy3             = _mm256_setzero_ps();
 676         fiz3             = _mm256_setzero_ps();
 677
 678         /* Start inner kernel loop */
 679         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 680         {
 681
 682             /* Get j neighbor index, and coordinate index */
 683             jnrA             = jjnr[jidx];
 684             jnrB             = jjnr[jidx+1];
 685             jnrC             = jjnr[jidx+2];
 686             jnrD             = jjnr[jidx+3];
 687             jnrE             = jjnr[jidx+4];
 688             jnrF             = jjnr[jidx+5];
 689             jnrG             = jjnr[jidx+6];
 690             jnrH             = jjnr[jidx+7];
 691             j_coord_offsetA  = DIM*jnrA;
 692             j_coord_offsetB  = DIM*jnrB;
 693             j_coord_offsetC  = DIM*jnrC;
 694             j_coord_offsetD  = DIM*jnrD;
 695             j_coord_offsetE  = DIM*jnrE;
 696             j_coord_offsetF  = DIM*jnrF;
 697             j_coord_offsetG  = DIM*jnrG;
 698             j_coord_offsetH  = DIM*jnrH;
 699
 700             /* load j atom coordinates */
 701             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 702                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 703                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 704                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 705                                                  &jx0,&jy0,&jz0);
 706
 707             /* Calculate displacement vector */
 708             dx10             = _mm256_sub_ps(ix1,jx0);
 709             dy10             = _mm256_sub_ps(iy1,jy0);
 710             dz10             = _mm256_sub_ps(iz1,jz0);
 711             dx20             = _mm256_sub_ps(ix2,jx0);
 712             dy20             = _mm256_sub_ps(iy2,jy0);
 713             dz20             = _mm256_sub_ps(iz2,jz0);
 714             dx30             = _mm256_sub_ps(ix3,jx0);
 715             dy30             = _mm256_sub_ps(iy3,jy0);
 716             dz30             = _mm256_sub_ps(iz3,jz0);
 717
 718             /* Calculate squared distance and things based on it */
 719             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 720             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 721             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 722
 723             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 724             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 725             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 726
 727             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 728             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 729             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 730
 731             /* Load parameters for j particles */
 732             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 733                                                                  charge+jnrC+0,charge+jnrD+0,
 734                                                                  charge+jnrE+0,charge+jnrF+0,
 735                                                                  charge+jnrG+0,charge+jnrH+0);
 736
 737             fjx0             = _mm256_setzero_ps();
 738             fjy0             = _mm256_setzero_ps();
 739             fjz0             = _mm256_setzero_ps();
 740
 741             /**************************
 742              * CALCULATE INTERACTIONS *
 743              **************************/
 744
 745             /* Compute parameters for interactions between i and j atoms */
 746             qq10             = _mm256_mul_ps(iq1,jq0);
 747
 748             /* REACTION-FIELD ELECTROSTATICS */
 749             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 750
 751             fscal            = felec;
 752
 753             /* Calculate temporary vectorial force */
 754             tx               = _mm256_mul_ps(fscal,dx10);
 755             ty               = _mm256_mul_ps(fscal,dy10);
 756             tz               = _mm256_mul_ps(fscal,dz10);
 757
 758             /* Update vectorial force */
 759             fix1             = _mm256_add_ps(fix1,tx);
 760             fiy1             = _mm256_add_ps(fiy1,ty);
 761             fiz1             = _mm256_add_ps(fiz1,tz);
 762
 763             fjx0             = _mm256_add_ps(fjx0,tx);
 764             fjy0             = _mm256_add_ps(fjy0,ty);
 765             fjz0             = _mm256_add_ps(fjz0,tz);
 766
 767             /**************************
 768              * CALCULATE INTERACTIONS *
 769              **************************/
 770
 771             /* Compute parameters for interactions between i and j atoms */
 772             qq20             = _mm256_mul_ps(iq2,jq0);
 773
 774             /* REACTION-FIELD ELECTROSTATICS */
 775             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 776
 777             fscal            = felec;
 778
 779             /* Calculate temporary vectorial force */
 780             tx               = _mm256_mul_ps(fscal,dx20);
 781             ty               = _mm256_mul_ps(fscal,dy20);
 782             tz               = _mm256_mul_ps(fscal,dz20);
 783
 784             /* Update vectorial force */
 785             fix2             = _mm256_add_ps(fix2,tx);
 786             fiy2             = _mm256_add_ps(fiy2,ty);
 787             fiz2             = _mm256_add_ps(fiz2,tz);
 788
 789             fjx0             = _mm256_add_ps(fjx0,tx);
 790             fjy0             = _mm256_add_ps(fjy0,ty);
 791             fjz0             = _mm256_add_ps(fjz0,tz);
 792
 793             /**************************
 794              * CALCULATE INTERACTIONS *
 795              **************************/
 796
 797             /* Compute parameters for interactions between i and j atoms */
 798             qq30             = _mm256_mul_ps(iq3,jq0);
 799
 800             /* REACTION-FIELD ELECTROSTATICS */
 801             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 802
 803             fscal            = felec;
 804
 805             /* Calculate temporary vectorial force */
 806             tx               = _mm256_mul_ps(fscal,dx30);
 807             ty               = _mm256_mul_ps(fscal,dy30);
 808             tz               = _mm256_mul_ps(fscal,dz30);
 809
 810             /* Update vectorial force */
 811             fix3             = _mm256_add_ps(fix3,tx);
 812             fiy3             = _mm256_add_ps(fiy3,ty);
 813             fiz3             = _mm256_add_ps(fiz3,tz);
 814
 815             fjx0             = _mm256_add_ps(fjx0,tx);
 816             fjy0             = _mm256_add_ps(fjy0,ty);
 817             fjz0             = _mm256_add_ps(fjz0,tz);
 818
 819             fjptrA             = f+j_coord_offsetA;
 820             fjptrB             = f+j_coord_offsetB;
 821             fjptrC             = f+j_coord_offsetC;
 822             fjptrD             = f+j_coord_offsetD;
 823             fjptrE             = f+j_coord_offsetE;
 824             fjptrF             = f+j_coord_offsetF;
 825             fjptrG             = f+j_coord_offsetG;
 826             fjptrH             = f+j_coord_offsetH;
 827
 828             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 829
 830             /* Inner loop uses 84 flops */
 831         }
 832
 833         if(jidx<j_index_end)
 834         {
 835
 836             /* Get j neighbor index, and coordinate index */
 837             jnrlistA         = jjnr[jidx];
 838             jnrlistB         = jjnr[jidx+1];
 839             jnrlistC         = jjnr[jidx+2];
 840             jnrlistD         = jjnr[jidx+3];
 841             jnrlistE         = jjnr[jidx+4];
 842             jnrlistF         = jjnr[jidx+5];
 843             jnrlistG         = jjnr[jidx+6];
 844             jnrlistH         = jjnr[jidx+7];
 845             /* Sign of each element will be negative for non-real atoms.
 846              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 847              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 848              */
 849             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 850                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 851
 852             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 853             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 854             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 855             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 856             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 857             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 858             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 859             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 860             j_coord_offsetA  = DIM*jnrA;
 861             j_coord_offsetB  = DIM*jnrB;
 862             j_coord_offsetC  = DIM*jnrC;
 863             j_coord_offsetD  = DIM*jnrD;
 864             j_coord_offsetE  = DIM*jnrE;
 865             j_coord_offsetF  = DIM*jnrF;
 866             j_coord_offsetG  = DIM*jnrG;
 867             j_coord_offsetH  = DIM*jnrH;
 868
 869             /* load j atom coordinates */
 870             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 871                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 872                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 873                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 874                                                  &jx0,&jy0,&jz0);
 875
 876             /* Calculate displacement vector */
 877             dx10             = _mm256_sub_ps(ix1,jx0);
 878             dy10             = _mm256_sub_ps(iy1,jy0);
 879             dz10             = _mm256_sub_ps(iz1,jz0);
 880             dx20             = _mm256_sub_ps(ix2,jx0);
 881             dy20             = _mm256_sub_ps(iy2,jy0);
 882             dz20             = _mm256_sub_ps(iz2,jz0);
 883             dx30             = _mm256_sub_ps(ix3,jx0);
 884             dy30             = _mm256_sub_ps(iy3,jy0);
 885             dz30             = _mm256_sub_ps(iz3,jz0);
 886
 887             /* Calculate squared distance and things based on it */
 888             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 889             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 890             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 891
 892             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 893             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 894             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 895
 896             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 897             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 898             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 899
 900             /* Load parameters for j particles */
 901             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 902                                                                  charge+jnrC+0,charge+jnrD+0,
 903                                                                  charge+jnrE+0,charge+jnrF+0,
 904                                                                  charge+jnrG+0,charge+jnrH+0);
 905
 906             fjx0             = _mm256_setzero_ps();
 907             fjy0             = _mm256_setzero_ps();
 908             fjz0             = _mm256_setzero_ps();
 909
 910             /**************************
 911              * CALCULATE INTERACTIONS *
 912              **************************/
 913
 914             /* Compute parameters for interactions between i and j atoms */
 915             qq10             = _mm256_mul_ps(iq1,jq0);
 916
 917             /* REACTION-FIELD ELECTROSTATICS */
 918             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 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             /* REACTION-FIELD ELECTROSTATICS */
 946             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 947
 948             fscal            = felec;
 949
 950             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 951
 952             /* Calculate temporary vectorial force */
 953             tx               = _mm256_mul_ps(fscal,dx20);
 954             ty               = _mm256_mul_ps(fscal,dy20);
 955             tz               = _mm256_mul_ps(fscal,dz20);
 956
 957             /* Update vectorial force */
 958             fix2             = _mm256_add_ps(fix2,tx);
 959             fiy2             = _mm256_add_ps(fiy2,ty);
 960             fiz2             = _mm256_add_ps(fiz2,tz);
 961
 962             fjx0             = _mm256_add_ps(fjx0,tx);
 963             fjy0             = _mm256_add_ps(fjy0,ty);
 964             fjz0             = _mm256_add_ps(fjz0,tz);
 965
 966             /**************************
 967              * CALCULATE INTERACTIONS *
 968              **************************/
 969
 970             /* Compute parameters for interactions between i and j atoms */
 971             qq30             = _mm256_mul_ps(iq3,jq0);
 972
 973             /* REACTION-FIELD ELECTROSTATICS */
 974             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 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 84 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*84);
1025 }