src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRFCut_VdwNone_GeomW3P1_avx_256_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_double 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_double.h"
  34 #include "kernelutil_x86_avx_256_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW3P1_VF_avx_256_double
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            None
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwNone_GeomW3P1_VF_avx_256_double
  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 refer to j loop unrolling done with AVX, e.g. for the four 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              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  63     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  64     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  65     real             rcutoff_scalar;
  66     real             *shiftvec,*fshift,*x,*f;
  67     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  68     real             scratch[4*DIM];
  69     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  70     real *           vdwioffsetptr0;
  71     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  72     real *           vdwioffsetptr1;
  73     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  74     real *           vdwioffsetptr2;
  75     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  76     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  77     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  78     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  79     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  80     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  81     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  82     real             *charge;
  83     __m256d          dummy_mask,cutoff_mask;
  84     __m128           tmpmask0,tmpmask1;
  85     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
  86     __m256d          one     = _mm256_set1_pd(1.0);
  87     __m256d          two     = _mm256_set1_pd(2.0);
  88     x                = xx[0];
  89     f                = ff[0];
  90
  91     nri              = nlist->nri;
  92     iinr             = nlist->iinr;
  93     jindex           = nlist->jindex;
  94     jjnr             = nlist->jjnr;
  95     shiftidx         = nlist->shift;
  96     gid              = nlist->gid;
  97     shiftvec         = fr->shift_vec[0];
  98     fshift           = fr->fshift[0];
  99     facel            = _mm256_set1_pd(fr->epsfac);
 100     charge           = mdatoms->chargeA;
 101     krf              = _mm256_set1_pd(fr->ic->k_rf);
 102     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 103     crf              = _mm256_set1_pd(fr->ic->c_rf);
 104
 105     /* Setup water-specific parameters */
 106     inr              = nlist->iinr[0];
 107     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 108     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 109     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 110
 111     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 112     rcutoff_scalar   = fr->rcoulomb;
 113     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 114     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 115
 116     /* Avoid stupid compiler warnings */
 117     jnrA = jnrB = jnrC = jnrD = 0;
 118     j_coord_offsetA = 0;
 119     j_coord_offsetB = 0;
 120     j_coord_offsetC = 0;
 121     j_coord_offsetD = 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_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 147                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 148
 149         fix0             = _mm256_setzero_pd();
 150         fiy0             = _mm256_setzero_pd();
 151         fiz0             = _mm256_setzero_pd();
 152         fix1             = _mm256_setzero_pd();
 153         fiy1             = _mm256_setzero_pd();
 154         fiz1             = _mm256_setzero_pd();
 155         fix2             = _mm256_setzero_pd();
 156         fiy2             = _mm256_setzero_pd();
 157         fiz2             = _mm256_setzero_pd();
 158
 159         /* Reset potential sums */
 160         velecsum         = _mm256_setzero_pd();
 161
 162         /* Start inner kernel loop */
 163         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 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             j_coord_offsetA  = DIM*jnrA;
 172             j_coord_offsetB  = DIM*jnrB;
 173             j_coord_offsetC  = DIM*jnrC;
 174             j_coord_offsetD  = DIM*jnrD;
 175
 176             /* load j atom coordinates */
 177             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 178                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 179                                                  &jx0,&jy0,&jz0);
 180
 181             /* Calculate displacement vector */
 182             dx00             = _mm256_sub_pd(ix0,jx0);
 183             dy00             = _mm256_sub_pd(iy0,jy0);
 184             dz00             = _mm256_sub_pd(iz0,jz0);
 185             dx10             = _mm256_sub_pd(ix1,jx0);
 186             dy10             = _mm256_sub_pd(iy1,jy0);
 187             dz10             = _mm256_sub_pd(iz1,jz0);
 188             dx20             = _mm256_sub_pd(ix2,jx0);
 189             dy20             = _mm256_sub_pd(iy2,jy0);
 190             dz20             = _mm256_sub_pd(iz2,jz0);
 191
 192             /* Calculate squared distance and things based on it */
 193             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 194             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 195             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 196
 197             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 198             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 199             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 200
 201             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 202             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 203             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 204
 205             /* Load parameters for j particles */
 206             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 207                                                                  charge+jnrC+0,charge+jnrD+0);
 208
 209             fjx0             = _mm256_setzero_pd();
 210             fjy0             = _mm256_setzero_pd();
 211             fjz0             = _mm256_setzero_pd();
 212
 213             /**************************
 214              * CALCULATE INTERACTIONS *
 215              **************************/
 216
 217             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 218             {
 219
 220             /* Compute parameters for interactions between i and j atoms */
 221             qq00             = _mm256_mul_pd(iq0,jq0);
 222
 223             /* REACTION-FIELD ELECTROSTATICS */
 224             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 225             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 226
 227             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 228
 229             /* Update potential sum for this i atom from the interaction with this j atom. */
 230             velec            = _mm256_and_pd(velec,cutoff_mask);
 231             velecsum         = _mm256_add_pd(velecsum,velec);
 232
 233             fscal            = felec;
 234
 235             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 236
 237             /* Calculate temporary vectorial force */
 238             tx               = _mm256_mul_pd(fscal,dx00);
 239             ty               = _mm256_mul_pd(fscal,dy00);
 240             tz               = _mm256_mul_pd(fscal,dz00);
 241
 242             /* Update vectorial force */
 243             fix0             = _mm256_add_pd(fix0,tx);
 244             fiy0             = _mm256_add_pd(fiy0,ty);
 245             fiz0             = _mm256_add_pd(fiz0,tz);
 246
 247             fjx0             = _mm256_add_pd(fjx0,tx);
 248             fjy0             = _mm256_add_pd(fjy0,ty);
 249             fjz0             = _mm256_add_pd(fjz0,tz);
 250
 251             }
 252
 253             /**************************
 254              * CALCULATE INTERACTIONS *
 255              **************************/
 256
 257             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 258             {
 259
 260             /* Compute parameters for interactions between i and j atoms */
 261             qq10             = _mm256_mul_pd(iq1,jq0);
 262
 263             /* REACTION-FIELD ELECTROSTATICS */
 264             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 265             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 266
 267             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 268
 269             /* Update potential sum for this i atom from the interaction with this j atom. */
 270             velec            = _mm256_and_pd(velec,cutoff_mask);
 271             velecsum         = _mm256_add_pd(velecsum,velec);
 272
 273             fscal            = felec;
 274
 275             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 276
 277             /* Calculate temporary vectorial force */
 278             tx               = _mm256_mul_pd(fscal,dx10);
 279             ty               = _mm256_mul_pd(fscal,dy10);
 280             tz               = _mm256_mul_pd(fscal,dz10);
 281
 282             /* Update vectorial force */
 283             fix1             = _mm256_add_pd(fix1,tx);
 284             fiy1             = _mm256_add_pd(fiy1,ty);
 285             fiz1             = _mm256_add_pd(fiz1,tz);
 286
 287             fjx0             = _mm256_add_pd(fjx0,tx);
 288             fjy0             = _mm256_add_pd(fjy0,ty);
 289             fjz0             = _mm256_add_pd(fjz0,tz);
 290
 291             }
 292
 293             /**************************
 294              * CALCULATE INTERACTIONS *
 295              **************************/
 296
 297             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 298             {
 299
 300             /* Compute parameters for interactions between i and j atoms */
 301             qq20             = _mm256_mul_pd(iq2,jq0);
 302
 303             /* REACTION-FIELD ELECTROSTATICS */
 304             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 305             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 306
 307             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 308
 309             /* Update potential sum for this i atom from the interaction with this j atom. */
 310             velec            = _mm256_and_pd(velec,cutoff_mask);
 311             velecsum         = _mm256_add_pd(velecsum,velec);
 312
 313             fscal            = felec;
 314
 315             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 316
 317             /* Calculate temporary vectorial force */
 318             tx               = _mm256_mul_pd(fscal,dx20);
 319             ty               = _mm256_mul_pd(fscal,dy20);
 320             tz               = _mm256_mul_pd(fscal,dz20);
 321
 322             /* Update vectorial force */
 323             fix2             = _mm256_add_pd(fix2,tx);
 324             fiy2             = _mm256_add_pd(fiy2,ty);
 325             fiz2             = _mm256_add_pd(fiz2,tz);
 326
 327             fjx0             = _mm256_add_pd(fjx0,tx);
 328             fjy0             = _mm256_add_pd(fjy0,ty);
 329             fjz0             = _mm256_add_pd(fjz0,tz);
 330
 331             }
 332
 333             fjptrA             = f+j_coord_offsetA;
 334             fjptrB             = f+j_coord_offsetB;
 335             fjptrC             = f+j_coord_offsetC;
 336             fjptrD             = f+j_coord_offsetD;
 337
 338             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 339
 340             /* Inner loop uses 111 flops */
 341         }
 342
 343         if(jidx<j_index_end)
 344         {
 345
 346             /* Get j neighbor index, and coordinate index */
 347             jnrlistA         = jjnr[jidx];
 348             jnrlistB         = jjnr[jidx+1];
 349             jnrlistC         = jjnr[jidx+2];
 350             jnrlistD         = jjnr[jidx+3];
 351             /* Sign of each element will be negative for non-real atoms.
 352              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 353              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 354              */
 355             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 356
 357             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 358             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 359             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 360
 361             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 362             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 363             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 364             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 365             j_coord_offsetA  = DIM*jnrA;
 366             j_coord_offsetB  = DIM*jnrB;
 367             j_coord_offsetC  = DIM*jnrC;
 368             j_coord_offsetD  = DIM*jnrD;
 369
 370             /* load j atom coordinates */
 371             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 372                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 373                                                  &jx0,&jy0,&jz0);
 374
 375             /* Calculate displacement vector */
 376             dx00             = _mm256_sub_pd(ix0,jx0);
 377             dy00             = _mm256_sub_pd(iy0,jy0);
 378             dz00             = _mm256_sub_pd(iz0,jz0);
 379             dx10             = _mm256_sub_pd(ix1,jx0);
 380             dy10             = _mm256_sub_pd(iy1,jy0);
 381             dz10             = _mm256_sub_pd(iz1,jz0);
 382             dx20             = _mm256_sub_pd(ix2,jx0);
 383             dy20             = _mm256_sub_pd(iy2,jy0);
 384             dz20             = _mm256_sub_pd(iz2,jz0);
 385
 386             /* Calculate squared distance and things based on it */
 387             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 388             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 389             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 390
 391             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 392             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 393             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 394
 395             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 396             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 397             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 398
 399             /* Load parameters for j particles */
 400             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 401                                                                  charge+jnrC+0,charge+jnrD+0);
 402
 403             fjx0             = _mm256_setzero_pd();
 404             fjy0             = _mm256_setzero_pd();
 405             fjz0             = _mm256_setzero_pd();
 406
 407             /**************************
 408              * CALCULATE INTERACTIONS *
 409              **************************/
 410
 411             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 412             {
 413
 414             /* Compute parameters for interactions between i and j atoms */
 415             qq00             = _mm256_mul_pd(iq0,jq0);
 416
 417             /* REACTION-FIELD ELECTROSTATICS */
 418             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 419             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 420
 421             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 422
 423             /* Update potential sum for this i atom from the interaction with this j atom. */
 424             velec            = _mm256_and_pd(velec,cutoff_mask);
 425             velec            = _mm256_andnot_pd(dummy_mask,velec);
 426             velecsum         = _mm256_add_pd(velecsum,velec);
 427
 428             fscal            = felec;
 429
 430             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 431
 432             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 433
 434             /* Calculate temporary vectorial force */
 435             tx               = _mm256_mul_pd(fscal,dx00);
 436             ty               = _mm256_mul_pd(fscal,dy00);
 437             tz               = _mm256_mul_pd(fscal,dz00);
 438
 439             /* Update vectorial force */
 440             fix0             = _mm256_add_pd(fix0,tx);
 441             fiy0             = _mm256_add_pd(fiy0,ty);
 442             fiz0             = _mm256_add_pd(fiz0,tz);
 443
 444             fjx0             = _mm256_add_pd(fjx0,tx);
 445             fjy0             = _mm256_add_pd(fjy0,ty);
 446             fjz0             = _mm256_add_pd(fjz0,tz);
 447
 448             }
 449
 450             /**************************
 451              * CALCULATE INTERACTIONS *
 452              **************************/
 453
 454             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 455             {
 456
 457             /* Compute parameters for interactions between i and j atoms */
 458             qq10             = _mm256_mul_pd(iq1,jq0);
 459
 460             /* REACTION-FIELD ELECTROSTATICS */
 461             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 462             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 463
 464             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 465
 466             /* Update potential sum for this i atom from the interaction with this j atom. */
 467             velec            = _mm256_and_pd(velec,cutoff_mask);
 468             velec            = _mm256_andnot_pd(dummy_mask,velec);
 469             velecsum         = _mm256_add_pd(velecsum,velec);
 470
 471             fscal            = felec;
 472
 473             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 474
 475             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 476
 477             /* Calculate temporary vectorial force */
 478             tx               = _mm256_mul_pd(fscal,dx10);
 479             ty               = _mm256_mul_pd(fscal,dy10);
 480             tz               = _mm256_mul_pd(fscal,dz10);
 481
 482             /* Update vectorial force */
 483             fix1             = _mm256_add_pd(fix1,tx);
 484             fiy1             = _mm256_add_pd(fiy1,ty);
 485             fiz1             = _mm256_add_pd(fiz1,tz);
 486
 487             fjx0             = _mm256_add_pd(fjx0,tx);
 488             fjy0             = _mm256_add_pd(fjy0,ty);
 489             fjz0             = _mm256_add_pd(fjz0,tz);
 490
 491             }
 492
 493             /**************************
 494              * CALCULATE INTERACTIONS *
 495              **************************/
 496
 497             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 498             {
 499
 500             /* Compute parameters for interactions between i and j atoms */
 501             qq20             = _mm256_mul_pd(iq2,jq0);
 502
 503             /* REACTION-FIELD ELECTROSTATICS */
 504             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 505             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 506
 507             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 508
 509             /* Update potential sum for this i atom from the interaction with this j atom. */
 510             velec            = _mm256_and_pd(velec,cutoff_mask);
 511             velec            = _mm256_andnot_pd(dummy_mask,velec);
 512             velecsum         = _mm256_add_pd(velecsum,velec);
 513
 514             fscal            = felec;
 515
 516             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 517
 518             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 519
 520             /* Calculate temporary vectorial force */
 521             tx               = _mm256_mul_pd(fscal,dx20);
 522             ty               = _mm256_mul_pd(fscal,dy20);
 523             tz               = _mm256_mul_pd(fscal,dz20);
 524
 525             /* Update vectorial force */
 526             fix2             = _mm256_add_pd(fix2,tx);
 527             fiy2             = _mm256_add_pd(fiy2,ty);
 528             fiz2             = _mm256_add_pd(fiz2,tz);
 529
 530             fjx0             = _mm256_add_pd(fjx0,tx);
 531             fjy0             = _mm256_add_pd(fjy0,ty);
 532             fjz0             = _mm256_add_pd(fjz0,tz);
 533
 534             }
 535
 536             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 537             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 538             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 539             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 540
 541             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 542
 543             /* Inner loop uses 111 flops */
 544         }
 545
 546         /* End of innermost loop */
 547
 548         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 549                                                  f+i_coord_offset,fshift+i_shift_offset);
 550
 551         ggid                        = gid[iidx];
 552         /* Update potential energies */
 553         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 554
 555         /* Increment number of inner iterations */
 556         inneriter                  += j_index_end - j_index_start;
 557
 558         /* Outer loop uses 19 flops */
 559     }
 560
 561     /* Increment number of outer iterations */
 562     outeriter        += nri;
 563
 564     /* Update outer/inner flops */
 565
 566     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*111);
 567 }
 568 /*
 569  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW3P1_F_avx_256_double
 570  * Electrostatics interaction: ReactionField
 571  * VdW interaction:            None
 572  * Geometry:                   Water3-Particle
 573  * Calculate force/pot:        Force
 574  */
 575 void
 576 nb_kernel_ElecRFCut_VdwNone_GeomW3P1_F_avx_256_double
 577                     (t_nblist * gmx_restrict                nlist,
 578                      rvec * gmx_restrict                    xx,
 579                      rvec * gmx_restrict                    ff,
 580                      t_forcerec * gmx_restrict              fr,
 581                      t_mdatoms * gmx_restrict               mdatoms,
 582                      nb_kernel_data_t * gmx_restrict        kernel_data,
 583                      t_nrnb * gmx_restrict                  nrnb)
 584 {
 585     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 586      * just 0 for non-waters.
 587      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 588      * jnr indices corresponding to data put in the four positions in the SIMD register.
 589      */
 590     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 591     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 592     int              jnrA,jnrB,jnrC,jnrD;
 593     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 594     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 595     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 596     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 597     real             rcutoff_scalar;
 598     real             *shiftvec,*fshift,*x,*f;
 599     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 600     real             scratch[4*DIM];
 601     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 602     real *           vdwioffsetptr0;
 603     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 604     real *           vdwioffsetptr1;
 605     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 606     real *           vdwioffsetptr2;
 607     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 608     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 609     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 610     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 611     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 612     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 613     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 614     real             *charge;
 615     __m256d          dummy_mask,cutoff_mask;
 616     __m128           tmpmask0,tmpmask1;
 617     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 618     __m256d          one     = _mm256_set1_pd(1.0);
 619     __m256d          two     = _mm256_set1_pd(2.0);
 620     x                = xx[0];
 621     f                = ff[0];
 622
 623     nri              = nlist->nri;
 624     iinr             = nlist->iinr;
 625     jindex           = nlist->jindex;
 626     jjnr             = nlist->jjnr;
 627     shiftidx         = nlist->shift;
 628     gid              = nlist->gid;
 629     shiftvec         = fr->shift_vec[0];
 630     fshift           = fr->fshift[0];
 631     facel            = _mm256_set1_pd(fr->epsfac);
 632     charge           = mdatoms->chargeA;
 633     krf              = _mm256_set1_pd(fr->ic->k_rf);
 634     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 635     crf              = _mm256_set1_pd(fr->ic->c_rf);
 636
 637     /* Setup water-specific parameters */
 638     inr              = nlist->iinr[0];
 639     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 640     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 641     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 642
 643     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 644     rcutoff_scalar   = fr->rcoulomb;
 645     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 646     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 647
 648     /* Avoid stupid compiler warnings */
 649     jnrA = jnrB = jnrC = jnrD = 0;
 650     j_coord_offsetA = 0;
 651     j_coord_offsetB = 0;
 652     j_coord_offsetC = 0;
 653     j_coord_offsetD = 0;
 654
 655     outeriter        = 0;
 656     inneriter        = 0;
 657
 658     for(iidx=0;iidx<4*DIM;iidx++)
 659     {
 660         scratch[iidx] = 0.0;
 661     }
 662
 663     /* Start outer loop over neighborlists */
 664     for(iidx=0; iidx<nri; iidx++)
 665     {
 666         /* Load shift vector for this list */
 667         i_shift_offset   = DIM*shiftidx[iidx];
 668
 669         /* Load limits for loop over neighbors */
 670         j_index_start    = jindex[iidx];
 671         j_index_end      = jindex[iidx+1];
 672
 673         /* Get outer coordinate index */
 674         inr              = iinr[iidx];
 675         i_coord_offset   = DIM*inr;
 676
 677         /* Load i particle coords and add shift vector */
 678         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 679                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 680
 681         fix0             = _mm256_setzero_pd();
 682         fiy0             = _mm256_setzero_pd();
 683         fiz0             = _mm256_setzero_pd();
 684         fix1             = _mm256_setzero_pd();
 685         fiy1             = _mm256_setzero_pd();
 686         fiz1             = _mm256_setzero_pd();
 687         fix2             = _mm256_setzero_pd();
 688         fiy2             = _mm256_setzero_pd();
 689         fiz2             = _mm256_setzero_pd();
 690
 691         /* Start inner kernel loop */
 692         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 693         {
 694
 695             /* Get j neighbor index, and coordinate index */
 696             jnrA             = jjnr[jidx];
 697             jnrB             = jjnr[jidx+1];
 698             jnrC             = jjnr[jidx+2];
 699             jnrD             = jjnr[jidx+3];
 700             j_coord_offsetA  = DIM*jnrA;
 701             j_coord_offsetB  = DIM*jnrB;
 702             j_coord_offsetC  = DIM*jnrC;
 703             j_coord_offsetD  = DIM*jnrD;
 704
 705             /* load j atom coordinates */
 706             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 707                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 708                                                  &jx0,&jy0,&jz0);
 709
 710             /* Calculate displacement vector */
 711             dx00             = _mm256_sub_pd(ix0,jx0);
 712             dy00             = _mm256_sub_pd(iy0,jy0);
 713             dz00             = _mm256_sub_pd(iz0,jz0);
 714             dx10             = _mm256_sub_pd(ix1,jx0);
 715             dy10             = _mm256_sub_pd(iy1,jy0);
 716             dz10             = _mm256_sub_pd(iz1,jz0);
 717             dx20             = _mm256_sub_pd(ix2,jx0);
 718             dy20             = _mm256_sub_pd(iy2,jy0);
 719             dz20             = _mm256_sub_pd(iz2,jz0);
 720
 721             /* Calculate squared distance and things based on it */
 722             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 723             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 724             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 725
 726             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 727             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 728             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 729
 730             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 731             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 732             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 733
 734             /* Load parameters for j particles */
 735             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 736                                                                  charge+jnrC+0,charge+jnrD+0);
 737
 738             fjx0             = _mm256_setzero_pd();
 739             fjy0             = _mm256_setzero_pd();
 740             fjz0             = _mm256_setzero_pd();
 741
 742             /**************************
 743              * CALCULATE INTERACTIONS *
 744              **************************/
 745
 746             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 747             {
 748
 749             /* Compute parameters for interactions between i and j atoms */
 750             qq00             = _mm256_mul_pd(iq0,jq0);
 751
 752             /* REACTION-FIELD ELECTROSTATICS */
 753             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 754
 755             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 756
 757             fscal            = felec;
 758
 759             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 760
 761             /* Calculate temporary vectorial force */
 762             tx               = _mm256_mul_pd(fscal,dx00);
 763             ty               = _mm256_mul_pd(fscal,dy00);
 764             tz               = _mm256_mul_pd(fscal,dz00);
 765
 766             /* Update vectorial force */
 767             fix0             = _mm256_add_pd(fix0,tx);
 768             fiy0             = _mm256_add_pd(fiy0,ty);
 769             fiz0             = _mm256_add_pd(fiz0,tz);
 770
 771             fjx0             = _mm256_add_pd(fjx0,tx);
 772             fjy0             = _mm256_add_pd(fjy0,ty);
 773             fjz0             = _mm256_add_pd(fjz0,tz);
 774
 775             }
 776
 777             /**************************
 778              * CALCULATE INTERACTIONS *
 779              **************************/
 780
 781             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 782             {
 783
 784             /* Compute parameters for interactions between i and j atoms */
 785             qq10             = _mm256_mul_pd(iq1,jq0);
 786
 787             /* REACTION-FIELD ELECTROSTATICS */
 788             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 789
 790             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 791
 792             fscal            = felec;
 793
 794             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 795
 796             /* Calculate temporary vectorial force */
 797             tx               = _mm256_mul_pd(fscal,dx10);
 798             ty               = _mm256_mul_pd(fscal,dy10);
 799             tz               = _mm256_mul_pd(fscal,dz10);
 800
 801             /* Update vectorial force */
 802             fix1             = _mm256_add_pd(fix1,tx);
 803             fiy1             = _mm256_add_pd(fiy1,ty);
 804             fiz1             = _mm256_add_pd(fiz1,tz);
 805
 806             fjx0             = _mm256_add_pd(fjx0,tx);
 807             fjy0             = _mm256_add_pd(fjy0,ty);
 808             fjz0             = _mm256_add_pd(fjz0,tz);
 809
 810             }
 811
 812             /**************************
 813              * CALCULATE INTERACTIONS *
 814              **************************/
 815
 816             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 817             {
 818
 819             /* Compute parameters for interactions between i and j atoms */
 820             qq20             = _mm256_mul_pd(iq2,jq0);
 821
 822             /* REACTION-FIELD ELECTROSTATICS */
 823             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 824
 825             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 826
 827             fscal            = felec;
 828
 829             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 830
 831             /* Calculate temporary vectorial force */
 832             tx               = _mm256_mul_pd(fscal,dx20);
 833             ty               = _mm256_mul_pd(fscal,dy20);
 834             tz               = _mm256_mul_pd(fscal,dz20);
 835
 836             /* Update vectorial force */
 837             fix2             = _mm256_add_pd(fix2,tx);
 838             fiy2             = _mm256_add_pd(fiy2,ty);
 839             fiz2             = _mm256_add_pd(fiz2,tz);
 840
 841             fjx0             = _mm256_add_pd(fjx0,tx);
 842             fjy0             = _mm256_add_pd(fjy0,ty);
 843             fjz0             = _mm256_add_pd(fjz0,tz);
 844
 845             }
 846
 847             fjptrA             = f+j_coord_offsetA;
 848             fjptrB             = f+j_coord_offsetB;
 849             fjptrC             = f+j_coord_offsetC;
 850             fjptrD             = f+j_coord_offsetD;
 851
 852             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 853
 854             /* Inner loop uses 93 flops */
 855         }
 856
 857         if(jidx<j_index_end)
 858         {
 859
 860             /* Get j neighbor index, and coordinate index */
 861             jnrlistA         = jjnr[jidx];
 862             jnrlistB         = jjnr[jidx+1];
 863             jnrlistC         = jjnr[jidx+2];
 864             jnrlistD         = jjnr[jidx+3];
 865             /* Sign of each element will be negative for non-real atoms.
 866              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 867              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 868              */
 869             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 870
 871             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 872             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 873             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 874
 875             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 876             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 877             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 878             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 879             j_coord_offsetA  = DIM*jnrA;
 880             j_coord_offsetB  = DIM*jnrB;
 881             j_coord_offsetC  = DIM*jnrC;
 882             j_coord_offsetD  = DIM*jnrD;
 883
 884             /* load j atom coordinates */
 885             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 886                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 887                                                  &jx0,&jy0,&jz0);
 888
 889             /* Calculate displacement vector */
 890             dx00             = _mm256_sub_pd(ix0,jx0);
 891             dy00             = _mm256_sub_pd(iy0,jy0);
 892             dz00             = _mm256_sub_pd(iz0,jz0);
 893             dx10             = _mm256_sub_pd(ix1,jx0);
 894             dy10             = _mm256_sub_pd(iy1,jy0);
 895             dz10             = _mm256_sub_pd(iz1,jz0);
 896             dx20             = _mm256_sub_pd(ix2,jx0);
 897             dy20             = _mm256_sub_pd(iy2,jy0);
 898             dz20             = _mm256_sub_pd(iz2,jz0);
 899
 900             /* Calculate squared distance and things based on it */
 901             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 902             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 903             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 904
 905             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 906             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 907             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 908
 909             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 910             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 911             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 912
 913             /* Load parameters for j particles */
 914             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 915                                                                  charge+jnrC+0,charge+jnrD+0);
 916
 917             fjx0             = _mm256_setzero_pd();
 918             fjy0             = _mm256_setzero_pd();
 919             fjz0             = _mm256_setzero_pd();
 920
 921             /**************************
 922              * CALCULATE INTERACTIONS *
 923              **************************/
 924
 925             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 926             {
 927
 928             /* Compute parameters for interactions between i and j atoms */
 929             qq00             = _mm256_mul_pd(iq0,jq0);
 930
 931             /* REACTION-FIELD ELECTROSTATICS */
 932             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 933
 934             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 935
 936             fscal            = felec;
 937
 938             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 939
 940             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 941
 942             /* Calculate temporary vectorial force */
 943             tx               = _mm256_mul_pd(fscal,dx00);
 944             ty               = _mm256_mul_pd(fscal,dy00);
 945             tz               = _mm256_mul_pd(fscal,dz00);
 946
 947             /* Update vectorial force */
 948             fix0             = _mm256_add_pd(fix0,tx);
 949             fiy0             = _mm256_add_pd(fiy0,ty);
 950             fiz0             = _mm256_add_pd(fiz0,tz);
 951
 952             fjx0             = _mm256_add_pd(fjx0,tx);
 953             fjy0             = _mm256_add_pd(fjy0,ty);
 954             fjz0             = _mm256_add_pd(fjz0,tz);
 955
 956             }
 957
 958             /**************************
 959              * CALCULATE INTERACTIONS *
 960              **************************/
 961
 962             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 963             {
 964
 965             /* Compute parameters for interactions between i and j atoms */
 966             qq10             = _mm256_mul_pd(iq1,jq0);
 967
 968             /* REACTION-FIELD ELECTROSTATICS */
 969             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 970
 971             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 972
 973             fscal            = felec;
 974
 975             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 976
 977             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 978
 979             /* Calculate temporary vectorial force */
 980             tx               = _mm256_mul_pd(fscal,dx10);
 981             ty               = _mm256_mul_pd(fscal,dy10);
 982             tz               = _mm256_mul_pd(fscal,dz10);
 983
 984             /* Update vectorial force */
 985             fix1             = _mm256_add_pd(fix1,tx);
 986             fiy1             = _mm256_add_pd(fiy1,ty);
 987             fiz1             = _mm256_add_pd(fiz1,tz);
 988
 989             fjx0             = _mm256_add_pd(fjx0,tx);
 990             fjy0             = _mm256_add_pd(fjy0,ty);
 991             fjz0             = _mm256_add_pd(fjz0,tz);
 992
 993             }
 994
 995             /**************************
 996              * CALCULATE INTERACTIONS *
 997              **************************/
 998
 999             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1000             {
1001
1002             /* Compute parameters for interactions between i and j atoms */
1003             qq20             = _mm256_mul_pd(iq2,jq0);
1004
1005             /* REACTION-FIELD ELECTROSTATICS */
1006             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1007
1008             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1009
1010             fscal            = felec;
1011
1012             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1013
1014             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1015
1016             /* Calculate temporary vectorial force */
1017             tx               = _mm256_mul_pd(fscal,dx20);
1018             ty               = _mm256_mul_pd(fscal,dy20);
1019             tz               = _mm256_mul_pd(fscal,dz20);
1020
1021             /* Update vectorial force */
1022             fix2             = _mm256_add_pd(fix2,tx);
1023             fiy2             = _mm256_add_pd(fiy2,ty);
1024             fiz2             = _mm256_add_pd(fiz2,tz);
1025
1026             fjx0             = _mm256_add_pd(fjx0,tx);
1027             fjy0             = _mm256_add_pd(fjy0,ty);
1028             fjz0             = _mm256_add_pd(fjz0,tz);
1029
1030             }
1031
1032             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1033             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1034             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1035             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1036
1037             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1038
1039             /* Inner loop uses 93 flops */
1040         }
1041
1042         /* End of innermost loop */
1043
1044         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1045                                                  f+i_coord_offset,fshift+i_shift_offset);
1046
1047         /* Increment number of inner iterations */
1048         inneriter                  += j_index_end - j_index_start;
1049
1050         /* Outer loop uses 18 flops */
1051     }
1052
1053     /* Increment number of outer iterations */
1054     outeriter        += nri;
1055
1056     /* Update outer/inner flops */
1057
1058     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*93);
1059 }