src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRFCut_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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW4P1_VF_avx_256_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            None
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwNone_GeomW4P1_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr1;
  89     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     real *           vdwioffsetptr3;
  93     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  95     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  97     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  99     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     __m256           dummy_mask,cutoff_mask;
 102     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 103     __m256           one     = _mm256_set1_ps(1.0);
 104     __m256           two     = _mm256_set1_ps(2.0);
 105     x                = xx[0];
 106     f                = ff[0];
 107
 108     nri              = nlist->nri;
 109     iinr             = nlist->iinr;
 110     jindex           = nlist->jindex;
 111     jjnr             = nlist->jjnr;
 112     shiftidx         = nlist->shift;
 113     gid              = nlist->gid;
 114     shiftvec         = fr->shift_vec[0];
 115     fshift           = fr->fshift[0];
 116     facel            = _mm256_set1_ps(fr->epsfac);
 117     charge           = mdatoms->chargeA;
 118     krf              = _mm256_set1_ps(fr->ic->k_rf);
 119     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 120     crf              = _mm256_set1_ps(fr->ic->c_rf);
 121
 122     /* Setup water-specific parameters */
 123     inr              = nlist->iinr[0];
 124     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 125     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 126     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 127
 128     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 129     rcutoff_scalar   = fr->rcoulomb;
 130     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 131     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 132
 133     /* Avoid stupid compiler warnings */
 134     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 135     j_coord_offsetA = 0;
 136     j_coord_offsetB = 0;
 137     j_coord_offsetC = 0;
 138     j_coord_offsetD = 0;
 139     j_coord_offsetE = 0;
 140     j_coord_offsetF = 0;
 141     j_coord_offsetG = 0;
 142     j_coord_offsetH = 0;
 143
 144     outeriter        = 0;
 145     inneriter        = 0;
 146
 147     for(iidx=0;iidx<4*DIM;iidx++)
 148     {
 149         scratch[iidx] = 0.0;
 150     }
 151
 152     /* Start outer loop over neighborlists */
 153     for(iidx=0; iidx<nri; iidx++)
 154     {
 155         /* Load shift vector for this list */
 156         i_shift_offset   = DIM*shiftidx[iidx];
 157
 158         /* Load limits for loop over neighbors */
 159         j_index_start    = jindex[iidx];
 160         j_index_end      = jindex[iidx+1];
 161
 162         /* Get outer coordinate index */
 163         inr              = iinr[iidx];
 164         i_coord_offset   = DIM*inr;
 165
 166         /* Load i particle coords and add shift vector */
 167         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 168                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 169
 170         fix1             = _mm256_setzero_ps();
 171         fiy1             = _mm256_setzero_ps();
 172         fiz1             = _mm256_setzero_ps();
 173         fix2             = _mm256_setzero_ps();
 174         fiy2             = _mm256_setzero_ps();
 175         fiz2             = _mm256_setzero_ps();
 176         fix3             = _mm256_setzero_ps();
 177         fiy3             = _mm256_setzero_ps();
 178         fiz3             = _mm256_setzero_ps();
 179
 180         /* Reset potential sums */
 181         velecsum         = _mm256_setzero_ps();
 182
 183         /* Start inner kernel loop */
 184         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 185         {
 186
 187             /* Get j neighbor index, and coordinate index */
 188             jnrA             = jjnr[jidx];
 189             jnrB             = jjnr[jidx+1];
 190             jnrC             = jjnr[jidx+2];
 191             jnrD             = jjnr[jidx+3];
 192             jnrE             = jjnr[jidx+4];
 193             jnrF             = jjnr[jidx+5];
 194             jnrG             = jjnr[jidx+6];
 195             jnrH             = jjnr[jidx+7];
 196             j_coord_offsetA  = DIM*jnrA;
 197             j_coord_offsetB  = DIM*jnrB;
 198             j_coord_offsetC  = DIM*jnrC;
 199             j_coord_offsetD  = DIM*jnrD;
 200             j_coord_offsetE  = DIM*jnrE;
 201             j_coord_offsetF  = DIM*jnrF;
 202             j_coord_offsetG  = DIM*jnrG;
 203             j_coord_offsetH  = DIM*jnrH;
 204
 205             /* load j atom coordinates */
 206             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 207                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 208                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 209                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 210                                                  &jx0,&jy0,&jz0);
 211
 212             /* Calculate displacement vector */
 213             dx10             = _mm256_sub_ps(ix1,jx0);
 214             dy10             = _mm256_sub_ps(iy1,jy0);
 215             dz10             = _mm256_sub_ps(iz1,jz0);
 216             dx20             = _mm256_sub_ps(ix2,jx0);
 217             dy20             = _mm256_sub_ps(iy2,jy0);
 218             dz20             = _mm256_sub_ps(iz2,jz0);
 219             dx30             = _mm256_sub_ps(ix3,jx0);
 220             dy30             = _mm256_sub_ps(iy3,jy0);
 221             dz30             = _mm256_sub_ps(iz3,jz0);
 222
 223             /* Calculate squared distance and things based on it */
 224             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 225             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 226             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 227
 228             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 229             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 230             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 231
 232             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 233             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 234             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 235
 236             /* Load parameters for j particles */
 237             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 238                                                                  charge+jnrC+0,charge+jnrD+0,
 239                                                                  charge+jnrE+0,charge+jnrF+0,
 240                                                                  charge+jnrG+0,charge+jnrH+0);
 241
 242             fjx0             = _mm256_setzero_ps();
 243             fjy0             = _mm256_setzero_ps();
 244             fjz0             = _mm256_setzero_ps();
 245
 246             /**************************
 247              * CALCULATE INTERACTIONS *
 248              **************************/
 249
 250             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 251             {
 252
 253             /* Compute parameters for interactions between i and j atoms */
 254             qq10             = _mm256_mul_ps(iq1,jq0);
 255
 256             /* REACTION-FIELD ELECTROSTATICS */
 257             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 258             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 259
 260             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 261
 262             /* Update potential sum for this i atom from the interaction with this j atom. */
 263             velec            = _mm256_and_ps(velec,cutoff_mask);
 264             velecsum         = _mm256_add_ps(velecsum,velec);
 265
 266             fscal            = felec;
 267
 268             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 269
 270             /* Calculate temporary vectorial force */
 271             tx               = _mm256_mul_ps(fscal,dx10);
 272             ty               = _mm256_mul_ps(fscal,dy10);
 273             tz               = _mm256_mul_ps(fscal,dz10);
 274
 275             /* Update vectorial force */
 276             fix1             = _mm256_add_ps(fix1,tx);
 277             fiy1             = _mm256_add_ps(fiy1,ty);
 278             fiz1             = _mm256_add_ps(fiz1,tz);
 279
 280             fjx0             = _mm256_add_ps(fjx0,tx);
 281             fjy0             = _mm256_add_ps(fjy0,ty);
 282             fjz0             = _mm256_add_ps(fjz0,tz);
 283
 284             }
 285
 286             /**************************
 287              * CALCULATE INTERACTIONS *
 288              **************************/
 289
 290             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 291             {
 292
 293             /* Compute parameters for interactions between i and j atoms */
 294             qq20             = _mm256_mul_ps(iq2,jq0);
 295
 296             /* REACTION-FIELD ELECTROSTATICS */
 297             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 298             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 299
 300             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 301
 302             /* Update potential sum for this i atom from the interaction with this j atom. */
 303             velec            = _mm256_and_ps(velec,cutoff_mask);
 304             velecsum         = _mm256_add_ps(velecsum,velec);
 305
 306             fscal            = felec;
 307
 308             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 309
 310             /* Calculate temporary vectorial force */
 311             tx               = _mm256_mul_ps(fscal,dx20);
 312             ty               = _mm256_mul_ps(fscal,dy20);
 313             tz               = _mm256_mul_ps(fscal,dz20);
 314
 315             /* Update vectorial force */
 316             fix2             = _mm256_add_ps(fix2,tx);
 317             fiy2             = _mm256_add_ps(fiy2,ty);
 318             fiz2             = _mm256_add_ps(fiz2,tz);
 319
 320             fjx0             = _mm256_add_ps(fjx0,tx);
 321             fjy0             = _mm256_add_ps(fjy0,ty);
 322             fjz0             = _mm256_add_ps(fjz0,tz);
 323
 324             }
 325
 326             /**************************
 327              * CALCULATE INTERACTIONS *
 328              **************************/
 329
 330             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 331             {
 332
 333             /* Compute parameters for interactions between i and j atoms */
 334             qq30             = _mm256_mul_ps(iq3,jq0);
 335
 336             /* REACTION-FIELD ELECTROSTATICS */
 337             velec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
 338             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 339
 340             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 341
 342             /* Update potential sum for this i atom from the interaction with this j atom. */
 343             velec            = _mm256_and_ps(velec,cutoff_mask);
 344             velecsum         = _mm256_add_ps(velecsum,velec);
 345
 346             fscal            = felec;
 347
 348             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 349
 350             /* Calculate temporary vectorial force */
 351             tx               = _mm256_mul_ps(fscal,dx30);
 352             ty               = _mm256_mul_ps(fscal,dy30);
 353             tz               = _mm256_mul_ps(fscal,dz30);
 354
 355             /* Update vectorial force */
 356             fix3             = _mm256_add_ps(fix3,tx);
 357             fiy3             = _mm256_add_ps(fiy3,ty);
 358             fiz3             = _mm256_add_ps(fiz3,tz);
 359
 360             fjx0             = _mm256_add_ps(fjx0,tx);
 361             fjy0             = _mm256_add_ps(fjy0,ty);
 362             fjz0             = _mm256_add_ps(fjz0,tz);
 363
 364             }
 365
 366             fjptrA             = f+j_coord_offsetA;
 367             fjptrB             = f+j_coord_offsetB;
 368             fjptrC             = f+j_coord_offsetC;
 369             fjptrD             = f+j_coord_offsetD;
 370             fjptrE             = f+j_coord_offsetE;
 371             fjptrF             = f+j_coord_offsetF;
 372             fjptrG             = f+j_coord_offsetG;
 373             fjptrH             = f+j_coord_offsetH;
 374
 375             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 376
 377             /* Inner loop uses 111 flops */
 378         }
 379
 380         if(jidx<j_index_end)
 381         {
 382
 383             /* Get j neighbor index, and coordinate index */
 384             jnrlistA         = jjnr[jidx];
 385             jnrlistB         = jjnr[jidx+1];
 386             jnrlistC         = jjnr[jidx+2];
 387             jnrlistD         = jjnr[jidx+3];
 388             jnrlistE         = jjnr[jidx+4];
 389             jnrlistF         = jjnr[jidx+5];
 390             jnrlistG         = jjnr[jidx+6];
 391             jnrlistH         = jjnr[jidx+7];
 392             /* Sign of each element will be negative for non-real atoms.
 393              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 394              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 395              */
 396             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 397                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 398
 399             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 400             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 401             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 402             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 403             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 404             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 405             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 406             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 407             j_coord_offsetA  = DIM*jnrA;
 408             j_coord_offsetB  = DIM*jnrB;
 409             j_coord_offsetC  = DIM*jnrC;
 410             j_coord_offsetD  = DIM*jnrD;
 411             j_coord_offsetE  = DIM*jnrE;
 412             j_coord_offsetF  = DIM*jnrF;
 413             j_coord_offsetG  = DIM*jnrG;
 414             j_coord_offsetH  = DIM*jnrH;
 415
 416             /* load j atom coordinates */
 417             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 418                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 419                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 420                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 421                                                  &jx0,&jy0,&jz0);
 422
 423             /* Calculate displacement vector */
 424             dx10             = _mm256_sub_ps(ix1,jx0);
 425             dy10             = _mm256_sub_ps(iy1,jy0);
 426             dz10             = _mm256_sub_ps(iz1,jz0);
 427             dx20             = _mm256_sub_ps(ix2,jx0);
 428             dy20             = _mm256_sub_ps(iy2,jy0);
 429             dz20             = _mm256_sub_ps(iz2,jz0);
 430             dx30             = _mm256_sub_ps(ix3,jx0);
 431             dy30             = _mm256_sub_ps(iy3,jy0);
 432             dz30             = _mm256_sub_ps(iz3,jz0);
 433
 434             /* Calculate squared distance and things based on it */
 435             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 436             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 437             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 438
 439             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 440             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 441             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 442
 443             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 444             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 445             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 446
 447             /* Load parameters for j particles */
 448             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 449                                                                  charge+jnrC+0,charge+jnrD+0,
 450                                                                  charge+jnrE+0,charge+jnrF+0,
 451                                                                  charge+jnrG+0,charge+jnrH+0);
 452
 453             fjx0             = _mm256_setzero_ps();
 454             fjy0             = _mm256_setzero_ps();
 455             fjz0             = _mm256_setzero_ps();
 456
 457             /**************************
 458              * CALCULATE INTERACTIONS *
 459              **************************/
 460
 461             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 462             {
 463
 464             /* Compute parameters for interactions between i and j atoms */
 465             qq10             = _mm256_mul_ps(iq1,jq0);
 466
 467             /* REACTION-FIELD ELECTROSTATICS */
 468             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 469             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 470
 471             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 472
 473             /* Update potential sum for this i atom from the interaction with this j atom. */
 474             velec            = _mm256_and_ps(velec,cutoff_mask);
 475             velec            = _mm256_andnot_ps(dummy_mask,velec);
 476             velecsum         = _mm256_add_ps(velecsum,velec);
 477
 478             fscal            = felec;
 479
 480             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 481
 482             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 483
 484             /* Calculate temporary vectorial force */
 485             tx               = _mm256_mul_ps(fscal,dx10);
 486             ty               = _mm256_mul_ps(fscal,dy10);
 487             tz               = _mm256_mul_ps(fscal,dz10);
 488
 489             /* Update vectorial force */
 490             fix1             = _mm256_add_ps(fix1,tx);
 491             fiy1             = _mm256_add_ps(fiy1,ty);
 492             fiz1             = _mm256_add_ps(fiz1,tz);
 493
 494             fjx0             = _mm256_add_ps(fjx0,tx);
 495             fjy0             = _mm256_add_ps(fjy0,ty);
 496             fjz0             = _mm256_add_ps(fjz0,tz);
 497
 498             }
 499
 500             /**************************
 501              * CALCULATE INTERACTIONS *
 502              **************************/
 503
 504             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 505             {
 506
 507             /* Compute parameters for interactions between i and j atoms */
 508             qq20             = _mm256_mul_ps(iq2,jq0);
 509
 510             /* REACTION-FIELD ELECTROSTATICS */
 511             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 512             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 513
 514             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 515
 516             /* Update potential sum for this i atom from the interaction with this j atom. */
 517             velec            = _mm256_and_ps(velec,cutoff_mask);
 518             velec            = _mm256_andnot_ps(dummy_mask,velec);
 519             velecsum         = _mm256_add_ps(velecsum,velec);
 520
 521             fscal            = felec;
 522
 523             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 524
 525             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 526
 527             /* Calculate temporary vectorial force */
 528             tx               = _mm256_mul_ps(fscal,dx20);
 529             ty               = _mm256_mul_ps(fscal,dy20);
 530             tz               = _mm256_mul_ps(fscal,dz20);
 531
 532             /* Update vectorial force */
 533             fix2             = _mm256_add_ps(fix2,tx);
 534             fiy2             = _mm256_add_ps(fiy2,ty);
 535             fiz2             = _mm256_add_ps(fiz2,tz);
 536
 537             fjx0             = _mm256_add_ps(fjx0,tx);
 538             fjy0             = _mm256_add_ps(fjy0,ty);
 539             fjz0             = _mm256_add_ps(fjz0,tz);
 540
 541             }
 542
 543             /**************************
 544              * CALCULATE INTERACTIONS *
 545              **************************/
 546
 547             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 548             {
 549
 550             /* Compute parameters for interactions between i and j atoms */
 551             qq30             = _mm256_mul_ps(iq3,jq0);
 552
 553             /* REACTION-FIELD ELECTROSTATICS */
 554             velec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
 555             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 556
 557             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 558
 559             /* Update potential sum for this i atom from the interaction with this j atom. */
 560             velec            = _mm256_and_ps(velec,cutoff_mask);
 561             velec            = _mm256_andnot_ps(dummy_mask,velec);
 562             velecsum         = _mm256_add_ps(velecsum,velec);
 563
 564             fscal            = felec;
 565
 566             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 567
 568             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 569
 570             /* Calculate temporary vectorial force */
 571             tx               = _mm256_mul_ps(fscal,dx30);
 572             ty               = _mm256_mul_ps(fscal,dy30);
 573             tz               = _mm256_mul_ps(fscal,dz30);
 574
 575             /* Update vectorial force */
 576             fix3             = _mm256_add_ps(fix3,tx);
 577             fiy3             = _mm256_add_ps(fiy3,ty);
 578             fiz3             = _mm256_add_ps(fiz3,tz);
 579
 580             fjx0             = _mm256_add_ps(fjx0,tx);
 581             fjy0             = _mm256_add_ps(fjy0,ty);
 582             fjz0             = _mm256_add_ps(fjz0,tz);
 583
 584             }
 585
 586             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 587             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 588             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 589             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 590             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 591             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 592             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 593             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 594
 595             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 596
 597             /* Inner loop uses 111 flops */
 598         }
 599
 600         /* End of innermost loop */
 601
 602         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 603                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
 604
 605         ggid                        = gid[iidx];
 606         /* Update potential energies */
 607         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 608
 609         /* Increment number of inner iterations */
 610         inneriter                  += j_index_end - j_index_start;
 611
 612         /* Outer loop uses 19 flops */
 613     }
 614
 615     /* Increment number of outer iterations */
 616     outeriter        += nri;
 617
 618     /* Update outer/inner flops */
 619
 620     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*111);
 621 }
 622 /*
 623  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW4P1_F_avx_256_single
 624  * Electrostatics interaction: ReactionField
 625  * VdW interaction:            None
 626  * Geometry:                   Water4-Particle
 627  * Calculate force/pot:        Force
 628  */
 629 void
 630 nb_kernel_ElecRFCut_VdwNone_GeomW4P1_F_avx_256_single
 631                     (t_nblist                    * gmx_restrict       nlist,
 632                      rvec                        * gmx_restrict          xx,
 633                      rvec                        * gmx_restrict          ff,
 634                      t_forcerec                  * gmx_restrict          fr,
 635                      t_mdatoms                   * gmx_restrict     mdatoms,
 636                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 637                      t_nrnb                      * gmx_restrict        nrnb)
 638 {
 639     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 640      * just 0 for non-waters.
 641      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 642      * jnr indices corresponding to data put in the four positions in the SIMD register.
 643      */
 644     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 645     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 646     int              jnrA,jnrB,jnrC,jnrD;
 647     int              jnrE,jnrF,jnrG,jnrH;
 648     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 649     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 650     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 651     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 652     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 653     real             rcutoff_scalar;
 654     real             *shiftvec,*fshift,*x,*f;
 655     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 656     real             scratch[4*DIM];
 657     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 658     real *           vdwioffsetptr1;
 659     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 660     real *           vdwioffsetptr2;
 661     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 662     real *           vdwioffsetptr3;
 663     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 664     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 665     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 666     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 667     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 668     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 669     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 670     real             *charge;
 671     __m256           dummy_mask,cutoff_mask;
 672     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 673     __m256           one     = _mm256_set1_ps(1.0);
 674     __m256           two     = _mm256_set1_ps(2.0);
 675     x                = xx[0];
 676     f                = ff[0];
 677
 678     nri              = nlist->nri;
 679     iinr             = nlist->iinr;
 680     jindex           = nlist->jindex;
 681     jjnr             = nlist->jjnr;
 682     shiftidx         = nlist->shift;
 683     gid              = nlist->gid;
 684     shiftvec         = fr->shift_vec[0];
 685     fshift           = fr->fshift[0];
 686     facel            = _mm256_set1_ps(fr->epsfac);
 687     charge           = mdatoms->chargeA;
 688     krf              = _mm256_set1_ps(fr->ic->k_rf);
 689     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 690     crf              = _mm256_set1_ps(fr->ic->c_rf);
 691
 692     /* Setup water-specific parameters */
 693     inr              = nlist->iinr[0];
 694     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 695     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 696     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 697
 698     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 699     rcutoff_scalar   = fr->rcoulomb;
 700     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 701     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 702
 703     /* Avoid stupid compiler warnings */
 704     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 705     j_coord_offsetA = 0;
 706     j_coord_offsetB = 0;
 707     j_coord_offsetC = 0;
 708     j_coord_offsetD = 0;
 709     j_coord_offsetE = 0;
 710     j_coord_offsetF = 0;
 711     j_coord_offsetG = 0;
 712     j_coord_offsetH = 0;
 713
 714     outeriter        = 0;
 715     inneriter        = 0;
 716
 717     for(iidx=0;iidx<4*DIM;iidx++)
 718     {
 719         scratch[iidx] = 0.0;
 720     }
 721
 722     /* Start outer loop over neighborlists */
 723     for(iidx=0; iidx<nri; iidx++)
 724     {
 725         /* Load shift vector for this list */
 726         i_shift_offset   = DIM*shiftidx[iidx];
 727
 728         /* Load limits for loop over neighbors */
 729         j_index_start    = jindex[iidx];
 730         j_index_end      = jindex[iidx+1];
 731
 732         /* Get outer coordinate index */
 733         inr              = iinr[iidx];
 734         i_coord_offset   = DIM*inr;
 735
 736         /* Load i particle coords and add shift vector */
 737         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 738                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 739
 740         fix1             = _mm256_setzero_ps();
 741         fiy1             = _mm256_setzero_ps();
 742         fiz1             = _mm256_setzero_ps();
 743         fix2             = _mm256_setzero_ps();
 744         fiy2             = _mm256_setzero_ps();
 745         fiz2             = _mm256_setzero_ps();
 746         fix3             = _mm256_setzero_ps();
 747         fiy3             = _mm256_setzero_ps();
 748         fiz3             = _mm256_setzero_ps();
 749
 750         /* Start inner kernel loop */
 751         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 752         {
 753
 754             /* Get j neighbor index, and coordinate index */
 755             jnrA             = jjnr[jidx];
 756             jnrB             = jjnr[jidx+1];
 757             jnrC             = jjnr[jidx+2];
 758             jnrD             = jjnr[jidx+3];
 759             jnrE             = jjnr[jidx+4];
 760             jnrF             = jjnr[jidx+5];
 761             jnrG             = jjnr[jidx+6];
 762             jnrH             = jjnr[jidx+7];
 763             j_coord_offsetA  = DIM*jnrA;
 764             j_coord_offsetB  = DIM*jnrB;
 765             j_coord_offsetC  = DIM*jnrC;
 766             j_coord_offsetD  = DIM*jnrD;
 767             j_coord_offsetE  = DIM*jnrE;
 768             j_coord_offsetF  = DIM*jnrF;
 769             j_coord_offsetG  = DIM*jnrG;
 770             j_coord_offsetH  = DIM*jnrH;
 771
 772             /* load j atom coordinates */
 773             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 774                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 775                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 776                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 777                                                  &jx0,&jy0,&jz0);
 778
 779             /* Calculate displacement vector */
 780             dx10             = _mm256_sub_ps(ix1,jx0);
 781             dy10             = _mm256_sub_ps(iy1,jy0);
 782             dz10             = _mm256_sub_ps(iz1,jz0);
 783             dx20             = _mm256_sub_ps(ix2,jx0);
 784             dy20             = _mm256_sub_ps(iy2,jy0);
 785             dz20             = _mm256_sub_ps(iz2,jz0);
 786             dx30             = _mm256_sub_ps(ix3,jx0);
 787             dy30             = _mm256_sub_ps(iy3,jy0);
 788             dz30             = _mm256_sub_ps(iz3,jz0);
 789
 790             /* Calculate squared distance and things based on it */
 791             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 792             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 793             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 794
 795             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 796             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 797             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 798
 799             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 800             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 801             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 802
 803             /* Load parameters for j particles */
 804             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 805                                                                  charge+jnrC+0,charge+jnrD+0,
 806                                                                  charge+jnrE+0,charge+jnrF+0,
 807                                                                  charge+jnrG+0,charge+jnrH+0);
 808
 809             fjx0             = _mm256_setzero_ps();
 810             fjy0             = _mm256_setzero_ps();
 811             fjz0             = _mm256_setzero_ps();
 812
 813             /**************************
 814              * CALCULATE INTERACTIONS *
 815              **************************/
 816
 817             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 818             {
 819
 820             /* Compute parameters for interactions between i and j atoms */
 821             qq10             = _mm256_mul_ps(iq1,jq0);
 822
 823             /* REACTION-FIELD ELECTROSTATICS */
 824             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 825
 826             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 827
 828             fscal            = felec;
 829
 830             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 831
 832             /* Calculate temporary vectorial force */
 833             tx               = _mm256_mul_ps(fscal,dx10);
 834             ty               = _mm256_mul_ps(fscal,dy10);
 835             tz               = _mm256_mul_ps(fscal,dz10);
 836
 837             /* Update vectorial force */
 838             fix1             = _mm256_add_ps(fix1,tx);
 839             fiy1             = _mm256_add_ps(fiy1,ty);
 840             fiz1             = _mm256_add_ps(fiz1,tz);
 841
 842             fjx0             = _mm256_add_ps(fjx0,tx);
 843             fjy0             = _mm256_add_ps(fjy0,ty);
 844             fjz0             = _mm256_add_ps(fjz0,tz);
 845
 846             }
 847
 848             /**************************
 849              * CALCULATE INTERACTIONS *
 850              **************************/
 851
 852             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 853             {
 854
 855             /* Compute parameters for interactions between i and j atoms */
 856             qq20             = _mm256_mul_ps(iq2,jq0);
 857
 858             /* REACTION-FIELD ELECTROSTATICS */
 859             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 860
 861             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 862
 863             fscal            = felec;
 864
 865             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 866
 867             /* Calculate temporary vectorial force */
 868             tx               = _mm256_mul_ps(fscal,dx20);
 869             ty               = _mm256_mul_ps(fscal,dy20);
 870             tz               = _mm256_mul_ps(fscal,dz20);
 871
 872             /* Update vectorial force */
 873             fix2             = _mm256_add_ps(fix2,tx);
 874             fiy2             = _mm256_add_ps(fiy2,ty);
 875             fiz2             = _mm256_add_ps(fiz2,tz);
 876
 877             fjx0             = _mm256_add_ps(fjx0,tx);
 878             fjy0             = _mm256_add_ps(fjy0,ty);
 879             fjz0             = _mm256_add_ps(fjz0,tz);
 880
 881             }
 882
 883             /**************************
 884              * CALCULATE INTERACTIONS *
 885              **************************/
 886
 887             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 888             {
 889
 890             /* Compute parameters for interactions between i and j atoms */
 891             qq30             = _mm256_mul_ps(iq3,jq0);
 892
 893             /* REACTION-FIELD ELECTROSTATICS */
 894             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 895
 896             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 897
 898             fscal            = felec;
 899
 900             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 901
 902             /* Calculate temporary vectorial force */
 903             tx               = _mm256_mul_ps(fscal,dx30);
 904             ty               = _mm256_mul_ps(fscal,dy30);
 905             tz               = _mm256_mul_ps(fscal,dz30);
 906
 907             /* Update vectorial force */
 908             fix3             = _mm256_add_ps(fix3,tx);
 909             fiy3             = _mm256_add_ps(fiy3,ty);
 910             fiz3             = _mm256_add_ps(fiz3,tz);
 911
 912             fjx0             = _mm256_add_ps(fjx0,tx);
 913             fjy0             = _mm256_add_ps(fjy0,ty);
 914             fjz0             = _mm256_add_ps(fjz0,tz);
 915
 916             }
 917
 918             fjptrA             = f+j_coord_offsetA;
 919             fjptrB             = f+j_coord_offsetB;
 920             fjptrC             = f+j_coord_offsetC;
 921             fjptrD             = f+j_coord_offsetD;
 922             fjptrE             = f+j_coord_offsetE;
 923             fjptrF             = f+j_coord_offsetF;
 924             fjptrG             = f+j_coord_offsetG;
 925             fjptrH             = f+j_coord_offsetH;
 926
 927             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 928
 929             /* Inner loop uses 93 flops */
 930         }
 931
 932         if(jidx<j_index_end)
 933         {
 934
 935             /* Get j neighbor index, and coordinate index */
 936             jnrlistA         = jjnr[jidx];
 937             jnrlistB         = jjnr[jidx+1];
 938             jnrlistC         = jjnr[jidx+2];
 939             jnrlistD         = jjnr[jidx+3];
 940             jnrlistE         = jjnr[jidx+4];
 941             jnrlistF         = jjnr[jidx+5];
 942             jnrlistG         = jjnr[jidx+6];
 943             jnrlistH         = jjnr[jidx+7];
 944             /* Sign of each element will be negative for non-real atoms.
 945              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 946              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 947              */
 948             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 949                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 950
 951             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 952             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 953             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 954             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 955             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 956             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 957             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 958             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 959             j_coord_offsetA  = DIM*jnrA;
 960             j_coord_offsetB  = DIM*jnrB;
 961             j_coord_offsetC  = DIM*jnrC;
 962             j_coord_offsetD  = DIM*jnrD;
 963             j_coord_offsetE  = DIM*jnrE;
 964             j_coord_offsetF  = DIM*jnrF;
 965             j_coord_offsetG  = DIM*jnrG;
 966             j_coord_offsetH  = DIM*jnrH;
 967
 968             /* load j atom coordinates */
 969             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 970                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 971                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 972                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 973                                                  &jx0,&jy0,&jz0);
 974
 975             /* Calculate displacement vector */
 976             dx10             = _mm256_sub_ps(ix1,jx0);
 977             dy10             = _mm256_sub_ps(iy1,jy0);
 978             dz10             = _mm256_sub_ps(iz1,jz0);
 979             dx20             = _mm256_sub_ps(ix2,jx0);
 980             dy20             = _mm256_sub_ps(iy2,jy0);
 981             dz20             = _mm256_sub_ps(iz2,jz0);
 982             dx30             = _mm256_sub_ps(ix3,jx0);
 983             dy30             = _mm256_sub_ps(iy3,jy0);
 984             dz30             = _mm256_sub_ps(iz3,jz0);
 985
 986             /* Calculate squared distance and things based on it */
 987             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 988             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 989             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 990
 991             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 992             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 993             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 994
 995             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 996             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 997             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 998
 999             /* Load parameters for j particles */
1000             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1001                                                                  charge+jnrC+0,charge+jnrD+0,
1002                                                                  charge+jnrE+0,charge+jnrF+0,
1003                                                                  charge+jnrG+0,charge+jnrH+0);
1004
1005             fjx0             = _mm256_setzero_ps();
1006             fjy0             = _mm256_setzero_ps();
1007             fjz0             = _mm256_setzero_ps();
1008
1009             /**************************
1010              * CALCULATE INTERACTIONS *
1011              **************************/
1012
1013             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1014             {
1015
1016             /* Compute parameters for interactions between i and j atoms */
1017             qq10             = _mm256_mul_ps(iq1,jq0);
1018
1019             /* REACTION-FIELD ELECTROSTATICS */
1020             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1021
1022             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1023
1024             fscal            = felec;
1025
1026             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1027
1028             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1029
1030             /* Calculate temporary vectorial force */
1031             tx               = _mm256_mul_ps(fscal,dx10);
1032             ty               = _mm256_mul_ps(fscal,dy10);
1033             tz               = _mm256_mul_ps(fscal,dz10);
1034
1035             /* Update vectorial force */
1036             fix1             = _mm256_add_ps(fix1,tx);
1037             fiy1             = _mm256_add_ps(fiy1,ty);
1038             fiz1             = _mm256_add_ps(fiz1,tz);
1039
1040             fjx0             = _mm256_add_ps(fjx0,tx);
1041             fjy0             = _mm256_add_ps(fjy0,ty);
1042             fjz0             = _mm256_add_ps(fjz0,tz);
1043
1044             }
1045
1046             /**************************
1047              * CALCULATE INTERACTIONS *
1048              **************************/
1049
1050             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1051             {
1052
1053             /* Compute parameters for interactions between i and j atoms */
1054             qq20             = _mm256_mul_ps(iq2,jq0);
1055
1056             /* REACTION-FIELD ELECTROSTATICS */
1057             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1058
1059             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1060
1061             fscal            = felec;
1062
1063             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1064
1065             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1066
1067             /* Calculate temporary vectorial force */
1068             tx               = _mm256_mul_ps(fscal,dx20);
1069             ty               = _mm256_mul_ps(fscal,dy20);
1070             tz               = _mm256_mul_ps(fscal,dz20);
1071
1072             /* Update vectorial force */
1073             fix2             = _mm256_add_ps(fix2,tx);
1074             fiy2             = _mm256_add_ps(fiy2,ty);
1075             fiz2             = _mm256_add_ps(fiz2,tz);
1076
1077             fjx0             = _mm256_add_ps(fjx0,tx);
1078             fjy0             = _mm256_add_ps(fjy0,ty);
1079             fjz0             = _mm256_add_ps(fjz0,tz);
1080
1081             }
1082
1083             /**************************
1084              * CALCULATE INTERACTIONS *
1085              **************************/
1086
1087             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1088             {
1089
1090             /* Compute parameters for interactions between i and j atoms */
1091             qq30             = _mm256_mul_ps(iq3,jq0);
1092
1093             /* REACTION-FIELD ELECTROSTATICS */
1094             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
1095
1096             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1097
1098             fscal            = felec;
1099
1100             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1101
1102             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1103
1104             /* Calculate temporary vectorial force */
1105             tx               = _mm256_mul_ps(fscal,dx30);
1106             ty               = _mm256_mul_ps(fscal,dy30);
1107             tz               = _mm256_mul_ps(fscal,dz30);
1108
1109             /* Update vectorial force */
1110             fix3             = _mm256_add_ps(fix3,tx);
1111             fiy3             = _mm256_add_ps(fiy3,ty);
1112             fiz3             = _mm256_add_ps(fiz3,tz);
1113
1114             fjx0             = _mm256_add_ps(fjx0,tx);
1115             fjy0             = _mm256_add_ps(fjy0,ty);
1116             fjz0             = _mm256_add_ps(fjz0,tz);
1117
1118             }
1119
1120             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1121             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1122             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1123             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1124             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1125             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1126             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1127             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1128
1129             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1130
1131             /* Inner loop uses 93 flops */
1132         }
1133
1134         /* End of innermost loop */
1135
1136         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1137                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1138
1139         /* Increment number of inner iterations */
1140         inneriter                  += j_index_end - j_index_start;
1141
1142         /* Outer loop uses 18 flops */
1143     }
1144
1145     /* Increment number of outer iterations */
1146     outeriter        += nri;
1147
1148     /* Update outer/inner flops */
1149
1150     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*93);
1151 }