src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRF_VdwNone_GeomW4P1_avx_256_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_256_single.h"
  34 #include "kernelutil_x86_avx_256_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW4P1_VF_avx_256_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            None
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_VdwNone_GeomW4P1_VF_avx_256_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrE,jnrF,jnrG,jnrH;
  62     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  63     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  64     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  65     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  66     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  67     real             rcutoff_scalar;
  68     real             *shiftvec,*fshift,*x,*f;
  69     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  70     real             scratch[4*DIM];
  71     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  72     real *           vdwioffsetptr1;
  73     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  74     real *           vdwioffsetptr2;
  75     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  76     real *           vdwioffsetptr3;
  77     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  78     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  79     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  80     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  81     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  82     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  83     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     __m256           dummy_mask,cutoff_mask;
  86     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  87     __m256           one     = _mm256_set1_ps(1.0);
  88     __m256           two     = _mm256_set1_ps(2.0);
  89     x                = xx[0];
  90     f                = ff[0];
  91
  92     nri              = nlist->nri;
  93     iinr             = nlist->iinr;
  94     jindex           = nlist->jindex;
  95     jjnr             = nlist->jjnr;
  96     shiftidx         = nlist->shift;
  97     gid              = nlist->gid;
  98     shiftvec         = fr->shift_vec[0];
  99     fshift           = fr->fshift[0];
 100     facel            = _mm256_set1_ps(fr->epsfac);
 101     charge           = mdatoms->chargeA;
 102     krf              = _mm256_set1_ps(fr->ic->k_rf);
 103     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 104     crf              = _mm256_set1_ps(fr->ic->c_rf);
 105
 106     /* Setup water-specific parameters */
 107     inr              = nlist->iinr[0];
 108     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 109     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 110     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 111
 112     /* Avoid stupid compiler warnings */
 113     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 114     j_coord_offsetA = 0;
 115     j_coord_offsetB = 0;
 116     j_coord_offsetC = 0;
 117     j_coord_offsetD = 0;
 118     j_coord_offsetE = 0;
 119     j_coord_offsetF = 0;
 120     j_coord_offsetG = 0;
 121     j_coord_offsetH = 0;
 122
 123     outeriter        = 0;
 124     inneriter        = 0;
 125
 126     for(iidx=0;iidx<4*DIM;iidx++)
 127     {
 128         scratch[iidx] = 0.0;
 129     }
 130
 131     /* Start outer loop over neighborlists */
 132     for(iidx=0; iidx<nri; iidx++)
 133     {
 134         /* Load shift vector for this list */
 135         i_shift_offset   = DIM*shiftidx[iidx];
 136
 137         /* Load limits for loop over neighbors */
 138         j_index_start    = jindex[iidx];
 139         j_index_end      = jindex[iidx+1];
 140
 141         /* Get outer coordinate index */
 142         inr              = iinr[iidx];
 143         i_coord_offset   = DIM*inr;
 144
 145         /* Load i particle coords and add shift vector */
 146         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 147                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 148
 149         fix1             = _mm256_setzero_ps();
 150         fiy1             = _mm256_setzero_ps();
 151         fiz1             = _mm256_setzero_ps();
 152         fix2             = _mm256_setzero_ps();
 153         fiy2             = _mm256_setzero_ps();
 154         fiz2             = _mm256_setzero_ps();
 155         fix3             = _mm256_setzero_ps();
 156         fiy3             = _mm256_setzero_ps();
 157         fiz3             = _mm256_setzero_ps();
 158
 159         /* Reset potential sums */
 160         velecsum         = _mm256_setzero_ps();
 161
 162         /* Start inner kernel loop */
 163         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 164         {
 165
 166             /* Get j neighbor index, and coordinate index */
 167             jnrA             = jjnr[jidx];
 168             jnrB             = jjnr[jidx+1];
 169             jnrC             = jjnr[jidx+2];
 170             jnrD             = jjnr[jidx+3];
 171             jnrE             = jjnr[jidx+4];
 172             jnrF             = jjnr[jidx+5];
 173             jnrG             = jjnr[jidx+6];
 174             jnrH             = jjnr[jidx+7];
 175             j_coord_offsetA  = DIM*jnrA;
 176             j_coord_offsetB  = DIM*jnrB;
 177             j_coord_offsetC  = DIM*jnrC;
 178             j_coord_offsetD  = DIM*jnrD;
 179             j_coord_offsetE  = DIM*jnrE;
 180             j_coord_offsetF  = DIM*jnrF;
 181             j_coord_offsetG  = DIM*jnrG;
 182             j_coord_offsetH  = DIM*jnrH;
 183
 184             /* load j atom coordinates */
 185             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 186                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 187                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 188                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 189                                                  &jx0,&jy0,&jz0);
 190
 191             /* Calculate displacement vector */
 192             dx10             = _mm256_sub_ps(ix1,jx0);
 193             dy10             = _mm256_sub_ps(iy1,jy0);
 194             dz10             = _mm256_sub_ps(iz1,jz0);
 195             dx20             = _mm256_sub_ps(ix2,jx0);
 196             dy20             = _mm256_sub_ps(iy2,jy0);
 197             dz20             = _mm256_sub_ps(iz2,jz0);
 198             dx30             = _mm256_sub_ps(ix3,jx0);
 199             dy30             = _mm256_sub_ps(iy3,jy0);
 200             dz30             = _mm256_sub_ps(iz3,jz0);
 201
 202             /* Calculate squared distance and things based on it */
 203             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 204             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 205             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 206
 207             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 208             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 209             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 210
 211             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 212             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 213             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 214
 215             /* Load parameters for j particles */
 216             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 217                                                                  charge+jnrC+0,charge+jnrD+0,
 218                                                                  charge+jnrE+0,charge+jnrF+0,
 219                                                                  charge+jnrG+0,charge+jnrH+0);
 220
 221             fjx0             = _mm256_setzero_ps();
 222             fjy0             = _mm256_setzero_ps();
 223             fjz0             = _mm256_setzero_ps();
 224
 225             /**************************
 226              * CALCULATE INTERACTIONS *
 227              **************************/
 228
 229             /* Compute parameters for interactions between i and j atoms */
 230             qq10             = _mm256_mul_ps(iq1,jq0);
 231
 232             /* REACTION-FIELD ELECTROSTATICS */
 233             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 234             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 235
 236             /* Update potential sum for this i atom from the interaction with this j atom. */
 237             velecsum         = _mm256_add_ps(velecsum,velec);
 238
 239             fscal            = felec;
 240
 241             /* Calculate temporary vectorial force */
 242             tx               = _mm256_mul_ps(fscal,dx10);
 243             ty               = _mm256_mul_ps(fscal,dy10);
 244             tz               = _mm256_mul_ps(fscal,dz10);
 245
 246             /* Update vectorial force */
 247             fix1             = _mm256_add_ps(fix1,tx);
 248             fiy1             = _mm256_add_ps(fiy1,ty);
 249             fiz1             = _mm256_add_ps(fiz1,tz);
 250
 251             fjx0             = _mm256_add_ps(fjx0,tx);
 252             fjy0             = _mm256_add_ps(fjy0,ty);
 253             fjz0             = _mm256_add_ps(fjz0,tz);
 254
 255             /**************************
 256              * CALCULATE INTERACTIONS *
 257              **************************/
 258
 259             /* Compute parameters for interactions between i and j atoms */
 260             qq20             = _mm256_mul_ps(iq2,jq0);
 261
 262             /* REACTION-FIELD ELECTROSTATICS */
 263             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 264             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 265
 266             /* Update potential sum for this i atom from the interaction with this j atom. */
 267             velecsum         = _mm256_add_ps(velecsum,velec);
 268
 269             fscal            = felec;
 270
 271             /* Calculate temporary vectorial force */
 272             tx               = _mm256_mul_ps(fscal,dx20);
 273             ty               = _mm256_mul_ps(fscal,dy20);
 274             tz               = _mm256_mul_ps(fscal,dz20);
 275
 276             /* Update vectorial force */
 277             fix2             = _mm256_add_ps(fix2,tx);
 278             fiy2             = _mm256_add_ps(fiy2,ty);
 279             fiz2             = _mm256_add_ps(fiz2,tz);
 280
 281             fjx0             = _mm256_add_ps(fjx0,tx);
 282             fjy0             = _mm256_add_ps(fjy0,ty);
 283             fjz0             = _mm256_add_ps(fjz0,tz);
 284
 285             /**************************
 286              * CALCULATE INTERACTIONS *
 287              **************************/
 288
 289             /* Compute parameters for interactions between i and j atoms */
 290             qq30             = _mm256_mul_ps(iq3,jq0);
 291
 292             /* REACTION-FIELD ELECTROSTATICS */
 293             velec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
 294             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 295
 296             /* Update potential sum for this i atom from the interaction with this j atom. */
 297             velecsum         = _mm256_add_ps(velecsum,velec);
 298
 299             fscal            = felec;
 300
 301             /* Calculate temporary vectorial force */
 302             tx               = _mm256_mul_ps(fscal,dx30);
 303             ty               = _mm256_mul_ps(fscal,dy30);
 304             tz               = _mm256_mul_ps(fscal,dz30);
 305
 306             /* Update vectorial force */
 307             fix3             = _mm256_add_ps(fix3,tx);
 308             fiy3             = _mm256_add_ps(fiy3,ty);
 309             fiz3             = _mm256_add_ps(fiz3,tz);
 310
 311             fjx0             = _mm256_add_ps(fjx0,tx);
 312             fjy0             = _mm256_add_ps(fjy0,ty);
 313             fjz0             = _mm256_add_ps(fjz0,tz);
 314
 315             fjptrA             = f+j_coord_offsetA;
 316             fjptrB             = f+j_coord_offsetB;
 317             fjptrC             = f+j_coord_offsetC;
 318             fjptrD             = f+j_coord_offsetD;
 319             fjptrE             = f+j_coord_offsetE;
 320             fjptrF             = f+j_coord_offsetF;
 321             fjptrG             = f+j_coord_offsetG;
 322             fjptrH             = f+j_coord_offsetH;
 323
 324             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 325
 326             /* Inner loop uses 99 flops */
 327         }
 328
 329         if(jidx<j_index_end)
 330         {
 331
 332             /* Get j neighbor index, and coordinate index */
 333             jnrlistA         = jjnr[jidx];
 334             jnrlistB         = jjnr[jidx+1];
 335             jnrlistC         = jjnr[jidx+2];
 336             jnrlistD         = jjnr[jidx+3];
 337             jnrlistE         = jjnr[jidx+4];
 338             jnrlistF         = jjnr[jidx+5];
 339             jnrlistG         = jjnr[jidx+6];
 340             jnrlistH         = jjnr[jidx+7];
 341             /* Sign of each element will be negative for non-real atoms.
 342              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 343              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 344              */
 345             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 346                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 347
 348             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 349             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 350             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 351             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 352             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 353             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 354             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 355             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 356             j_coord_offsetA  = DIM*jnrA;
 357             j_coord_offsetB  = DIM*jnrB;
 358             j_coord_offsetC  = DIM*jnrC;
 359             j_coord_offsetD  = DIM*jnrD;
 360             j_coord_offsetE  = DIM*jnrE;
 361             j_coord_offsetF  = DIM*jnrF;
 362             j_coord_offsetG  = DIM*jnrG;
 363             j_coord_offsetH  = DIM*jnrH;
 364
 365             /* load j atom coordinates */
 366             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 367                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 368                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 369                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 370                                                  &jx0,&jy0,&jz0);
 371
 372             /* Calculate displacement vector */
 373             dx10             = _mm256_sub_ps(ix1,jx0);
 374             dy10             = _mm256_sub_ps(iy1,jy0);
 375             dz10             = _mm256_sub_ps(iz1,jz0);
 376             dx20             = _mm256_sub_ps(ix2,jx0);
 377             dy20             = _mm256_sub_ps(iy2,jy0);
 378             dz20             = _mm256_sub_ps(iz2,jz0);
 379             dx30             = _mm256_sub_ps(ix3,jx0);
 380             dy30             = _mm256_sub_ps(iy3,jy0);
 381             dz30             = _mm256_sub_ps(iz3,jz0);
 382
 383             /* Calculate squared distance and things based on it */
 384             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 385             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 386             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 387
 388             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 389             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 390             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 391
 392             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 393             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 394             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 395
 396             /* Load parameters for j particles */
 397             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 398                                                                  charge+jnrC+0,charge+jnrD+0,
 399                                                                  charge+jnrE+0,charge+jnrF+0,
 400                                                                  charge+jnrG+0,charge+jnrH+0);
 401
 402             fjx0             = _mm256_setzero_ps();
 403             fjy0             = _mm256_setzero_ps();
 404             fjz0             = _mm256_setzero_ps();
 405
 406             /**************************
 407              * CALCULATE INTERACTIONS *
 408              **************************/
 409
 410             /* Compute parameters for interactions between i and j atoms */
 411             qq10             = _mm256_mul_ps(iq1,jq0);
 412
 413             /* REACTION-FIELD ELECTROSTATICS */
 414             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 415             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 416
 417             /* Update potential sum for this i atom from the interaction with this j atom. */
 418             velec            = _mm256_andnot_ps(dummy_mask,velec);
 419             velecsum         = _mm256_add_ps(velecsum,velec);
 420
 421             fscal            = felec;
 422
 423             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 424
 425             /* Calculate temporary vectorial force */
 426             tx               = _mm256_mul_ps(fscal,dx10);
 427             ty               = _mm256_mul_ps(fscal,dy10);
 428             tz               = _mm256_mul_ps(fscal,dz10);
 429
 430             /* Update vectorial force */
 431             fix1             = _mm256_add_ps(fix1,tx);
 432             fiy1             = _mm256_add_ps(fiy1,ty);
 433             fiz1             = _mm256_add_ps(fiz1,tz);
 434
 435             fjx0             = _mm256_add_ps(fjx0,tx);
 436             fjy0             = _mm256_add_ps(fjy0,ty);
 437             fjz0             = _mm256_add_ps(fjz0,tz);
 438
 439             /**************************
 440              * CALCULATE INTERACTIONS *
 441              **************************/
 442
 443             /* Compute parameters for interactions between i and j atoms */
 444             qq20             = _mm256_mul_ps(iq2,jq0);
 445
 446             /* REACTION-FIELD ELECTROSTATICS */
 447             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 448             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 449
 450             /* Update potential sum for this i atom from the interaction with this j atom. */
 451             velec            = _mm256_andnot_ps(dummy_mask,velec);
 452             velecsum         = _mm256_add_ps(velecsum,velec);
 453
 454             fscal            = felec;
 455
 456             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 457
 458             /* Calculate temporary vectorial force */
 459             tx               = _mm256_mul_ps(fscal,dx20);
 460             ty               = _mm256_mul_ps(fscal,dy20);
 461             tz               = _mm256_mul_ps(fscal,dz20);
 462
 463             /* Update vectorial force */
 464             fix2             = _mm256_add_ps(fix2,tx);
 465             fiy2             = _mm256_add_ps(fiy2,ty);
 466             fiz2             = _mm256_add_ps(fiz2,tz);
 467
 468             fjx0             = _mm256_add_ps(fjx0,tx);
 469             fjy0             = _mm256_add_ps(fjy0,ty);
 470             fjz0             = _mm256_add_ps(fjz0,tz);
 471
 472             /**************************
 473              * CALCULATE INTERACTIONS *
 474              **************************/
 475
 476             /* Compute parameters for interactions between i and j atoms */
 477             qq30             = _mm256_mul_ps(iq3,jq0);
 478
 479             /* REACTION-FIELD ELECTROSTATICS */
 480             velec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
 481             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 482
 483             /* Update potential sum for this i atom from the interaction with this j atom. */
 484             velec            = _mm256_andnot_ps(dummy_mask,velec);
 485             velecsum         = _mm256_add_ps(velecsum,velec);
 486
 487             fscal            = felec;
 488
 489             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 490
 491             /* Calculate temporary vectorial force */
 492             tx               = _mm256_mul_ps(fscal,dx30);
 493             ty               = _mm256_mul_ps(fscal,dy30);
 494             tz               = _mm256_mul_ps(fscal,dz30);
 495
 496             /* Update vectorial force */
 497             fix3             = _mm256_add_ps(fix3,tx);
 498             fiy3             = _mm256_add_ps(fiy3,ty);
 499             fiz3             = _mm256_add_ps(fiz3,tz);
 500
 501             fjx0             = _mm256_add_ps(fjx0,tx);
 502             fjy0             = _mm256_add_ps(fjy0,ty);
 503             fjz0             = _mm256_add_ps(fjz0,tz);
 504
 505             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 506             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 507             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 508             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 509             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 510             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 511             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 512             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 513
 514             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 515
 516             /* Inner loop uses 99 flops */
 517         }
 518
 519         /* End of innermost loop */
 520
 521         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 522                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
 523
 524         ggid                        = gid[iidx];
 525         /* Update potential energies */
 526         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 527
 528         /* Increment number of inner iterations */
 529         inneriter                  += j_index_end - j_index_start;
 530
 531         /* Outer loop uses 19 flops */
 532     }
 533
 534     /* Increment number of outer iterations */
 535     outeriter        += nri;
 536
 537     /* Update outer/inner flops */
 538
 539     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*99);
 540 }
 541 /*
 542  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW4P1_F_avx_256_single
 543  * Electrostatics interaction: ReactionField
 544  * VdW interaction:            None
 545  * Geometry:                   Water4-Particle
 546  * Calculate force/pot:        Force
 547  */
 548 void
 549 nb_kernel_ElecRF_VdwNone_GeomW4P1_F_avx_256_single
 550                     (t_nblist * gmx_restrict                nlist,
 551                      rvec * gmx_restrict                    xx,
 552                      rvec * gmx_restrict                    ff,
 553                      t_forcerec * gmx_restrict              fr,
 554                      t_mdatoms * gmx_restrict               mdatoms,
 555                      nb_kernel_data_t * gmx_restrict        kernel_data,
 556                      t_nrnb * gmx_restrict                  nrnb)
 557 {
 558     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 559      * just 0 for non-waters.
 560      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 561      * jnr indices corresponding to data put in the four positions in the SIMD register.
 562      */
 563     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 564     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 565     int              jnrA,jnrB,jnrC,jnrD;
 566     int              jnrE,jnrF,jnrG,jnrH;
 567     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 568     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 569     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 570     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 571     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 572     real             rcutoff_scalar;
 573     real             *shiftvec,*fshift,*x,*f;
 574     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 575     real             scratch[4*DIM];
 576     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 577     real *           vdwioffsetptr1;
 578     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 579     real *           vdwioffsetptr2;
 580     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 581     real *           vdwioffsetptr3;
 582     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 583     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 584     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 585     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 586     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 587     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 588     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 589     real             *charge;
 590     __m256           dummy_mask,cutoff_mask;
 591     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 592     __m256           one     = _mm256_set1_ps(1.0);
 593     __m256           two     = _mm256_set1_ps(2.0);
 594     x                = xx[0];
 595     f                = ff[0];
 596
 597     nri              = nlist->nri;
 598     iinr             = nlist->iinr;
 599     jindex           = nlist->jindex;
 600     jjnr             = nlist->jjnr;
 601     shiftidx         = nlist->shift;
 602     gid              = nlist->gid;
 603     shiftvec         = fr->shift_vec[0];
 604     fshift           = fr->fshift[0];
 605     facel            = _mm256_set1_ps(fr->epsfac);
 606     charge           = mdatoms->chargeA;
 607     krf              = _mm256_set1_ps(fr->ic->k_rf);
 608     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 609     crf              = _mm256_set1_ps(fr->ic->c_rf);
 610
 611     /* Setup water-specific parameters */
 612     inr              = nlist->iinr[0];
 613     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 614     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 615     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 616
 617     /* Avoid stupid compiler warnings */
 618     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 619     j_coord_offsetA = 0;
 620     j_coord_offsetB = 0;
 621     j_coord_offsetC = 0;
 622     j_coord_offsetD = 0;
 623     j_coord_offsetE = 0;
 624     j_coord_offsetF = 0;
 625     j_coord_offsetG = 0;
 626     j_coord_offsetH = 0;
 627
 628     outeriter        = 0;
 629     inneriter        = 0;
 630
 631     for(iidx=0;iidx<4*DIM;iidx++)
 632     {
 633         scratch[iidx] = 0.0;
 634     }
 635
 636     /* Start outer loop over neighborlists */
 637     for(iidx=0; iidx<nri; iidx++)
 638     {
 639         /* Load shift vector for this list */
 640         i_shift_offset   = DIM*shiftidx[iidx];
 641
 642         /* Load limits for loop over neighbors */
 643         j_index_start    = jindex[iidx];
 644         j_index_end      = jindex[iidx+1];
 645
 646         /* Get outer coordinate index */
 647         inr              = iinr[iidx];
 648         i_coord_offset   = DIM*inr;
 649
 650         /* Load i particle coords and add shift vector */
 651         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 652                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 653
 654         fix1             = _mm256_setzero_ps();
 655         fiy1             = _mm256_setzero_ps();
 656         fiz1             = _mm256_setzero_ps();
 657         fix2             = _mm256_setzero_ps();
 658         fiy2             = _mm256_setzero_ps();
 659         fiz2             = _mm256_setzero_ps();
 660         fix3             = _mm256_setzero_ps();
 661         fiy3             = _mm256_setzero_ps();
 662         fiz3             = _mm256_setzero_ps();
 663
 664         /* Start inner kernel loop */
 665         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 666         {
 667
 668             /* Get j neighbor index, and coordinate index */
 669             jnrA             = jjnr[jidx];
 670             jnrB             = jjnr[jidx+1];
 671             jnrC             = jjnr[jidx+2];
 672             jnrD             = jjnr[jidx+3];
 673             jnrE             = jjnr[jidx+4];
 674             jnrF             = jjnr[jidx+5];
 675             jnrG             = jjnr[jidx+6];
 676             jnrH             = jjnr[jidx+7];
 677             j_coord_offsetA  = DIM*jnrA;
 678             j_coord_offsetB  = DIM*jnrB;
 679             j_coord_offsetC  = DIM*jnrC;
 680             j_coord_offsetD  = DIM*jnrD;
 681             j_coord_offsetE  = DIM*jnrE;
 682             j_coord_offsetF  = DIM*jnrF;
 683             j_coord_offsetG  = DIM*jnrG;
 684             j_coord_offsetH  = DIM*jnrH;
 685
 686             /* load j atom coordinates */
 687             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 688                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 689                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 690                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 691                                                  &jx0,&jy0,&jz0);
 692
 693             /* Calculate displacement vector */
 694             dx10             = _mm256_sub_ps(ix1,jx0);
 695             dy10             = _mm256_sub_ps(iy1,jy0);
 696             dz10             = _mm256_sub_ps(iz1,jz0);
 697             dx20             = _mm256_sub_ps(ix2,jx0);
 698             dy20             = _mm256_sub_ps(iy2,jy0);
 699             dz20             = _mm256_sub_ps(iz2,jz0);
 700             dx30             = _mm256_sub_ps(ix3,jx0);
 701             dy30             = _mm256_sub_ps(iy3,jy0);
 702             dz30             = _mm256_sub_ps(iz3,jz0);
 703
 704             /* Calculate squared distance and things based on it */
 705             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 706             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 707             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 708
 709             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 710             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 711             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 712
 713             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 714             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 715             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 716
 717             /* Load parameters for j particles */
 718             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 719                                                                  charge+jnrC+0,charge+jnrD+0,
 720                                                                  charge+jnrE+0,charge+jnrF+0,
 721                                                                  charge+jnrG+0,charge+jnrH+0);
 722
 723             fjx0             = _mm256_setzero_ps();
 724             fjy0             = _mm256_setzero_ps();
 725             fjz0             = _mm256_setzero_ps();
 726
 727             /**************************
 728              * CALCULATE INTERACTIONS *
 729              **************************/
 730
 731             /* Compute parameters for interactions between i and j atoms */
 732             qq10             = _mm256_mul_ps(iq1,jq0);
 733
 734             /* REACTION-FIELD ELECTROSTATICS */
 735             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 736
 737             fscal            = felec;
 738
 739             /* Calculate temporary vectorial force */
 740             tx               = _mm256_mul_ps(fscal,dx10);
 741             ty               = _mm256_mul_ps(fscal,dy10);
 742             tz               = _mm256_mul_ps(fscal,dz10);
 743
 744             /* Update vectorial force */
 745             fix1             = _mm256_add_ps(fix1,tx);
 746             fiy1             = _mm256_add_ps(fiy1,ty);
 747             fiz1             = _mm256_add_ps(fiz1,tz);
 748
 749             fjx0             = _mm256_add_ps(fjx0,tx);
 750             fjy0             = _mm256_add_ps(fjy0,ty);
 751             fjz0             = _mm256_add_ps(fjz0,tz);
 752
 753             /**************************
 754              * CALCULATE INTERACTIONS *
 755              **************************/
 756
 757             /* Compute parameters for interactions between i and j atoms */
 758             qq20             = _mm256_mul_ps(iq2,jq0);
 759
 760             /* REACTION-FIELD ELECTROSTATICS */
 761             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 762
 763             fscal            = felec;
 764
 765             /* Calculate temporary vectorial force */
 766             tx               = _mm256_mul_ps(fscal,dx20);
 767             ty               = _mm256_mul_ps(fscal,dy20);
 768             tz               = _mm256_mul_ps(fscal,dz20);
 769
 770             /* Update vectorial force */
 771             fix2             = _mm256_add_ps(fix2,tx);
 772             fiy2             = _mm256_add_ps(fiy2,ty);
 773             fiz2             = _mm256_add_ps(fiz2,tz);
 774
 775             fjx0             = _mm256_add_ps(fjx0,tx);
 776             fjy0             = _mm256_add_ps(fjy0,ty);
 777             fjz0             = _mm256_add_ps(fjz0,tz);
 778
 779             /**************************
 780              * CALCULATE INTERACTIONS *
 781              **************************/
 782
 783             /* Compute parameters for interactions between i and j atoms */
 784             qq30             = _mm256_mul_ps(iq3,jq0);
 785
 786             /* REACTION-FIELD ELECTROSTATICS */
 787             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 788
 789             fscal            = felec;
 790
 791             /* Calculate temporary vectorial force */
 792             tx               = _mm256_mul_ps(fscal,dx30);
 793             ty               = _mm256_mul_ps(fscal,dy30);
 794             tz               = _mm256_mul_ps(fscal,dz30);
 795
 796             /* Update vectorial force */
 797             fix3             = _mm256_add_ps(fix3,tx);
 798             fiy3             = _mm256_add_ps(fiy3,ty);
 799             fiz3             = _mm256_add_ps(fiz3,tz);
 800
 801             fjx0             = _mm256_add_ps(fjx0,tx);
 802             fjy0             = _mm256_add_ps(fjy0,ty);
 803             fjz0             = _mm256_add_ps(fjz0,tz);
 804
 805             fjptrA             = f+j_coord_offsetA;
 806             fjptrB             = f+j_coord_offsetB;
 807             fjptrC             = f+j_coord_offsetC;
 808             fjptrD             = f+j_coord_offsetD;
 809             fjptrE             = f+j_coord_offsetE;
 810             fjptrF             = f+j_coord_offsetF;
 811             fjptrG             = f+j_coord_offsetG;
 812             fjptrH             = f+j_coord_offsetH;
 813
 814             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 815
 816             /* Inner loop uses 84 flops */
 817         }
 818
 819         if(jidx<j_index_end)
 820         {
 821
 822             /* Get j neighbor index, and coordinate index */
 823             jnrlistA         = jjnr[jidx];
 824             jnrlistB         = jjnr[jidx+1];
 825             jnrlistC         = jjnr[jidx+2];
 826             jnrlistD         = jjnr[jidx+3];
 827             jnrlistE         = jjnr[jidx+4];
 828             jnrlistF         = jjnr[jidx+5];
 829             jnrlistG         = jjnr[jidx+6];
 830             jnrlistH         = jjnr[jidx+7];
 831             /* Sign of each element will be negative for non-real atoms.
 832              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 833              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 834              */
 835             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 836                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 837
 838             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 839             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 840             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 841             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 842             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 843             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 844             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 845             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 846             j_coord_offsetA  = DIM*jnrA;
 847             j_coord_offsetB  = DIM*jnrB;
 848             j_coord_offsetC  = DIM*jnrC;
 849             j_coord_offsetD  = DIM*jnrD;
 850             j_coord_offsetE  = DIM*jnrE;
 851             j_coord_offsetF  = DIM*jnrF;
 852             j_coord_offsetG  = DIM*jnrG;
 853             j_coord_offsetH  = DIM*jnrH;
 854
 855             /* load j atom coordinates */
 856             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 857                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 858                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 859                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 860                                                  &jx0,&jy0,&jz0);
 861
 862             /* Calculate displacement vector */
 863             dx10             = _mm256_sub_ps(ix1,jx0);
 864             dy10             = _mm256_sub_ps(iy1,jy0);
 865             dz10             = _mm256_sub_ps(iz1,jz0);
 866             dx20             = _mm256_sub_ps(ix2,jx0);
 867             dy20             = _mm256_sub_ps(iy2,jy0);
 868             dz20             = _mm256_sub_ps(iz2,jz0);
 869             dx30             = _mm256_sub_ps(ix3,jx0);
 870             dy30             = _mm256_sub_ps(iy3,jy0);
 871             dz30             = _mm256_sub_ps(iz3,jz0);
 872
 873             /* Calculate squared distance and things based on it */
 874             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 875             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 876             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 877
 878             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 879             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 880             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 881
 882             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 883             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 884             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 885
 886             /* Load parameters for j particles */
 887             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 888                                                                  charge+jnrC+0,charge+jnrD+0,
 889                                                                  charge+jnrE+0,charge+jnrF+0,
 890                                                                  charge+jnrG+0,charge+jnrH+0);
 891
 892             fjx0             = _mm256_setzero_ps();
 893             fjy0             = _mm256_setzero_ps();
 894             fjz0             = _mm256_setzero_ps();
 895
 896             /**************************
 897              * CALCULATE INTERACTIONS *
 898              **************************/
 899
 900             /* Compute parameters for interactions between i and j atoms */
 901             qq10             = _mm256_mul_ps(iq1,jq0);
 902
 903             /* REACTION-FIELD ELECTROSTATICS */
 904             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 905
 906             fscal            = felec;
 907
 908             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 909
 910             /* Calculate temporary vectorial force */
 911             tx               = _mm256_mul_ps(fscal,dx10);
 912             ty               = _mm256_mul_ps(fscal,dy10);
 913             tz               = _mm256_mul_ps(fscal,dz10);
 914
 915             /* Update vectorial force */
 916             fix1             = _mm256_add_ps(fix1,tx);
 917             fiy1             = _mm256_add_ps(fiy1,ty);
 918             fiz1             = _mm256_add_ps(fiz1,tz);
 919
 920             fjx0             = _mm256_add_ps(fjx0,tx);
 921             fjy0             = _mm256_add_ps(fjy0,ty);
 922             fjz0             = _mm256_add_ps(fjz0,tz);
 923
 924             /**************************
 925              * CALCULATE INTERACTIONS *
 926              **************************/
 927
 928             /* Compute parameters for interactions between i and j atoms */
 929             qq20             = _mm256_mul_ps(iq2,jq0);
 930
 931             /* REACTION-FIELD ELECTROSTATICS */
 932             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 933
 934             fscal            = felec;
 935
 936             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 937
 938             /* Calculate temporary vectorial force */
 939             tx               = _mm256_mul_ps(fscal,dx20);
 940             ty               = _mm256_mul_ps(fscal,dy20);
 941             tz               = _mm256_mul_ps(fscal,dz20);
 942
 943             /* Update vectorial force */
 944             fix2             = _mm256_add_ps(fix2,tx);
 945             fiy2             = _mm256_add_ps(fiy2,ty);
 946             fiz2             = _mm256_add_ps(fiz2,tz);
 947
 948             fjx0             = _mm256_add_ps(fjx0,tx);
 949             fjy0             = _mm256_add_ps(fjy0,ty);
 950             fjz0             = _mm256_add_ps(fjz0,tz);
 951
 952             /**************************
 953              * CALCULATE INTERACTIONS *
 954              **************************/
 955
 956             /* Compute parameters for interactions between i and j atoms */
 957             qq30             = _mm256_mul_ps(iq3,jq0);
 958
 959             /* REACTION-FIELD ELECTROSTATICS */
 960             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 961
 962             fscal            = felec;
 963
 964             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 965
 966             /* Calculate temporary vectorial force */
 967             tx               = _mm256_mul_ps(fscal,dx30);
 968             ty               = _mm256_mul_ps(fscal,dy30);
 969             tz               = _mm256_mul_ps(fscal,dz30);
 970
 971             /* Update vectorial force */
 972             fix3             = _mm256_add_ps(fix3,tx);
 973             fiy3             = _mm256_add_ps(fiy3,ty);
 974             fiz3             = _mm256_add_ps(fiz3,tz);
 975
 976             fjx0             = _mm256_add_ps(fjx0,tx);
 977             fjy0             = _mm256_add_ps(fjy0,ty);
 978             fjz0             = _mm256_add_ps(fjz0,tz);
 979
 980             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 981             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 982             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 983             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 984             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 985             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 986             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 987             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 988
 989             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 990
 991             /* Inner loop uses 84 flops */
 992         }
 993
 994         /* End of innermost loop */
 995
 996         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 997                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
 998
 999         /* Increment number of inner iterations */
1000         inneriter                  += j_index_end - j_index_start;
1001
1002         /* Outer loop uses 18 flops */
1003     }
1004
1005     /* Increment number of outer iterations */
1006     outeriter        += nri;
1007
1008     /* Update outer/inner flops */
1009
1010     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*84);
1011 }