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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_double.h"
  48 #include "kernelutil_x86_avx_256_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW3P1_VF_avx_256_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            None
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwNone_GeomW3P1_VF_avx_256_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  84     real *           vdwioffsetptr0;
  85     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  86     real *           vdwioffsetptr1;
  87     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  88     real *           vdwioffsetptr2;
  89     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  91     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  96     real             *charge;
  97     __m256d          dummy_mask,cutoff_mask;
  98     __m128           tmpmask0,tmpmask1;
  99     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 100     __m256d          one     = _mm256_set1_pd(1.0);
 101     __m256d          two     = _mm256_set1_pd(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm256_set1_pd(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115     krf              = _mm256_set1_pd(fr->ic->k_rf);
 116     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 117     crf              = _mm256_set1_pd(fr->ic->c_rf);
 118
 119     /* Setup water-specific parameters */
 120     inr              = nlist->iinr[0];
 121     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 122     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 123     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 124
 125     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 126     rcutoff_scalar   = fr->rcoulomb;
 127     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 128     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 129
 130     /* Avoid stupid compiler warnings */
 131     jnrA = jnrB = jnrC = jnrD = 0;
 132     j_coord_offsetA = 0;
 133     j_coord_offsetB = 0;
 134     j_coord_offsetC = 0;
 135     j_coord_offsetD = 0;
 136
 137     outeriter        = 0;
 138     inneriter        = 0;
 139
 140     for(iidx=0;iidx<4*DIM;iidx++)
 141     {
 142         scratch[iidx] = 0.0;
 143     }
 144
 145     /* Start outer loop over neighborlists */
 146     for(iidx=0; iidx<nri; iidx++)
 147     {
 148         /* Load shift vector for this list */
 149         i_shift_offset   = DIM*shiftidx[iidx];
 150
 151         /* Load limits for loop over neighbors */
 152         j_index_start    = jindex[iidx];
 153         j_index_end      = jindex[iidx+1];
 154
 155         /* Get outer coordinate index */
 156         inr              = iinr[iidx];
 157         i_coord_offset   = DIM*inr;
 158
 159         /* Load i particle coords and add shift vector */
 160         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 161                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 162
 163         fix0             = _mm256_setzero_pd();
 164         fiy0             = _mm256_setzero_pd();
 165         fiz0             = _mm256_setzero_pd();
 166         fix1             = _mm256_setzero_pd();
 167         fiy1             = _mm256_setzero_pd();
 168         fiz1             = _mm256_setzero_pd();
 169         fix2             = _mm256_setzero_pd();
 170         fiy2             = _mm256_setzero_pd();
 171         fiz2             = _mm256_setzero_pd();
 172
 173         /* Reset potential sums */
 174         velecsum         = _mm256_setzero_pd();
 175
 176         /* Start inner kernel loop */
 177         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 178         {
 179
 180             /* Get j neighbor index, and coordinate index */
 181             jnrA             = jjnr[jidx];
 182             jnrB             = jjnr[jidx+1];
 183             jnrC             = jjnr[jidx+2];
 184             jnrD             = jjnr[jidx+3];
 185             j_coord_offsetA  = DIM*jnrA;
 186             j_coord_offsetB  = DIM*jnrB;
 187             j_coord_offsetC  = DIM*jnrC;
 188             j_coord_offsetD  = DIM*jnrD;
 189
 190             /* load j atom coordinates */
 191             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 192                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 193                                                  &jx0,&jy0,&jz0);
 194
 195             /* Calculate displacement vector */
 196             dx00             = _mm256_sub_pd(ix0,jx0);
 197             dy00             = _mm256_sub_pd(iy0,jy0);
 198             dz00             = _mm256_sub_pd(iz0,jz0);
 199             dx10             = _mm256_sub_pd(ix1,jx0);
 200             dy10             = _mm256_sub_pd(iy1,jy0);
 201             dz10             = _mm256_sub_pd(iz1,jz0);
 202             dx20             = _mm256_sub_pd(ix2,jx0);
 203             dy20             = _mm256_sub_pd(iy2,jy0);
 204             dz20             = _mm256_sub_pd(iz2,jz0);
 205
 206             /* Calculate squared distance and things based on it */
 207             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 208             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 209             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 210
 211             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 212             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 213             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 214
 215             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 216             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 217             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 218
 219             /* Load parameters for j particles */
 220             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 221                                                                  charge+jnrC+0,charge+jnrD+0);
 222
 223             fjx0             = _mm256_setzero_pd();
 224             fjy0             = _mm256_setzero_pd();
 225             fjz0             = _mm256_setzero_pd();
 226
 227             /**************************
 228              * CALCULATE INTERACTIONS *
 229              **************************/
 230
 231             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 232             {
 233
 234             /* Compute parameters for interactions between i and j atoms */
 235             qq00             = _mm256_mul_pd(iq0,jq0);
 236
 237             /* REACTION-FIELD ELECTROSTATICS */
 238             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 239             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 240
 241             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 242
 243             /* Update potential sum for this i atom from the interaction with this j atom. */
 244             velec            = _mm256_and_pd(velec,cutoff_mask);
 245             velecsum         = _mm256_add_pd(velecsum,velec);
 246
 247             fscal            = felec;
 248
 249             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 250
 251             /* Calculate temporary vectorial force */
 252             tx               = _mm256_mul_pd(fscal,dx00);
 253             ty               = _mm256_mul_pd(fscal,dy00);
 254             tz               = _mm256_mul_pd(fscal,dz00);
 255
 256             /* Update vectorial force */
 257             fix0             = _mm256_add_pd(fix0,tx);
 258             fiy0             = _mm256_add_pd(fiy0,ty);
 259             fiz0             = _mm256_add_pd(fiz0,tz);
 260
 261             fjx0             = _mm256_add_pd(fjx0,tx);
 262             fjy0             = _mm256_add_pd(fjy0,ty);
 263             fjz0             = _mm256_add_pd(fjz0,tz);
 264
 265             }
 266
 267             /**************************
 268              * CALCULATE INTERACTIONS *
 269              **************************/
 270
 271             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 272             {
 273
 274             /* Compute parameters for interactions between i and j atoms */
 275             qq10             = _mm256_mul_pd(iq1,jq0);
 276
 277             /* REACTION-FIELD ELECTROSTATICS */
 278             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 279             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 280
 281             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 282
 283             /* Update potential sum for this i atom from the interaction with this j atom. */
 284             velec            = _mm256_and_pd(velec,cutoff_mask);
 285             velecsum         = _mm256_add_pd(velecsum,velec);
 286
 287             fscal            = felec;
 288
 289             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 290
 291             /* Calculate temporary vectorial force */
 292             tx               = _mm256_mul_pd(fscal,dx10);
 293             ty               = _mm256_mul_pd(fscal,dy10);
 294             tz               = _mm256_mul_pd(fscal,dz10);
 295
 296             /* Update vectorial force */
 297             fix1             = _mm256_add_pd(fix1,tx);
 298             fiy1             = _mm256_add_pd(fiy1,ty);
 299             fiz1             = _mm256_add_pd(fiz1,tz);
 300
 301             fjx0             = _mm256_add_pd(fjx0,tx);
 302             fjy0             = _mm256_add_pd(fjy0,ty);
 303             fjz0             = _mm256_add_pd(fjz0,tz);
 304
 305             }
 306
 307             /**************************
 308              * CALCULATE INTERACTIONS *
 309              **************************/
 310
 311             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 312             {
 313
 314             /* Compute parameters for interactions between i and j atoms */
 315             qq20             = _mm256_mul_pd(iq2,jq0);
 316
 317             /* REACTION-FIELD ELECTROSTATICS */
 318             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 319             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 320
 321             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 322
 323             /* Update potential sum for this i atom from the interaction with this j atom. */
 324             velec            = _mm256_and_pd(velec,cutoff_mask);
 325             velecsum         = _mm256_add_pd(velecsum,velec);
 326
 327             fscal            = felec;
 328
 329             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 330
 331             /* Calculate temporary vectorial force */
 332             tx               = _mm256_mul_pd(fscal,dx20);
 333             ty               = _mm256_mul_pd(fscal,dy20);
 334             tz               = _mm256_mul_pd(fscal,dz20);
 335
 336             /* Update vectorial force */
 337             fix2             = _mm256_add_pd(fix2,tx);
 338             fiy2             = _mm256_add_pd(fiy2,ty);
 339             fiz2             = _mm256_add_pd(fiz2,tz);
 340
 341             fjx0             = _mm256_add_pd(fjx0,tx);
 342             fjy0             = _mm256_add_pd(fjy0,ty);
 343             fjz0             = _mm256_add_pd(fjz0,tz);
 344
 345             }
 346
 347             fjptrA             = f+j_coord_offsetA;
 348             fjptrB             = f+j_coord_offsetB;
 349             fjptrC             = f+j_coord_offsetC;
 350             fjptrD             = f+j_coord_offsetD;
 351
 352             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 353
 354             /* Inner loop uses 111 flops */
 355         }
 356
 357         if(jidx<j_index_end)
 358         {
 359
 360             /* Get j neighbor index, and coordinate index */
 361             jnrlistA         = jjnr[jidx];
 362             jnrlistB         = jjnr[jidx+1];
 363             jnrlistC         = jjnr[jidx+2];
 364             jnrlistD         = jjnr[jidx+3];
 365             /* Sign of each element will be negative for non-real atoms.
 366              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 367              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 368              */
 369             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 370
 371             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 372             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 373             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 374
 375             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 376             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 377             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 378             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 379             j_coord_offsetA  = DIM*jnrA;
 380             j_coord_offsetB  = DIM*jnrB;
 381             j_coord_offsetC  = DIM*jnrC;
 382             j_coord_offsetD  = DIM*jnrD;
 383
 384             /* load j atom coordinates */
 385             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 386                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 387                                                  &jx0,&jy0,&jz0);
 388
 389             /* Calculate displacement vector */
 390             dx00             = _mm256_sub_pd(ix0,jx0);
 391             dy00             = _mm256_sub_pd(iy0,jy0);
 392             dz00             = _mm256_sub_pd(iz0,jz0);
 393             dx10             = _mm256_sub_pd(ix1,jx0);
 394             dy10             = _mm256_sub_pd(iy1,jy0);
 395             dz10             = _mm256_sub_pd(iz1,jz0);
 396             dx20             = _mm256_sub_pd(ix2,jx0);
 397             dy20             = _mm256_sub_pd(iy2,jy0);
 398             dz20             = _mm256_sub_pd(iz2,jz0);
 399
 400             /* Calculate squared distance and things based on it */
 401             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 402             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 403             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 404
 405             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 406             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 407             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 408
 409             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 410             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 411             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 412
 413             /* Load parameters for j particles */
 414             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 415                                                                  charge+jnrC+0,charge+jnrD+0);
 416
 417             fjx0             = _mm256_setzero_pd();
 418             fjy0             = _mm256_setzero_pd();
 419             fjz0             = _mm256_setzero_pd();
 420
 421             /**************************
 422              * CALCULATE INTERACTIONS *
 423              **************************/
 424
 425             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 426             {
 427
 428             /* Compute parameters for interactions between i and j atoms */
 429             qq00             = _mm256_mul_pd(iq0,jq0);
 430
 431             /* REACTION-FIELD ELECTROSTATICS */
 432             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 433             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 434
 435             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 436
 437             /* Update potential sum for this i atom from the interaction with this j atom. */
 438             velec            = _mm256_and_pd(velec,cutoff_mask);
 439             velec            = _mm256_andnot_pd(dummy_mask,velec);
 440             velecsum         = _mm256_add_pd(velecsum,velec);
 441
 442             fscal            = felec;
 443
 444             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 445
 446             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 447
 448             /* Calculate temporary vectorial force */
 449             tx               = _mm256_mul_pd(fscal,dx00);
 450             ty               = _mm256_mul_pd(fscal,dy00);
 451             tz               = _mm256_mul_pd(fscal,dz00);
 452
 453             /* Update vectorial force */
 454             fix0             = _mm256_add_pd(fix0,tx);
 455             fiy0             = _mm256_add_pd(fiy0,ty);
 456             fiz0             = _mm256_add_pd(fiz0,tz);
 457
 458             fjx0             = _mm256_add_pd(fjx0,tx);
 459             fjy0             = _mm256_add_pd(fjy0,ty);
 460             fjz0             = _mm256_add_pd(fjz0,tz);
 461
 462             }
 463
 464             /**************************
 465              * CALCULATE INTERACTIONS *
 466              **************************/
 467
 468             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 469             {
 470
 471             /* Compute parameters for interactions between i and j atoms */
 472             qq10             = _mm256_mul_pd(iq1,jq0);
 473
 474             /* REACTION-FIELD ELECTROSTATICS */
 475             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 476             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 477
 478             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 479
 480             /* Update potential sum for this i atom from the interaction with this j atom. */
 481             velec            = _mm256_and_pd(velec,cutoff_mask);
 482             velec            = _mm256_andnot_pd(dummy_mask,velec);
 483             velecsum         = _mm256_add_pd(velecsum,velec);
 484
 485             fscal            = felec;
 486
 487             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 488
 489             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 490
 491             /* Calculate temporary vectorial force */
 492             tx               = _mm256_mul_pd(fscal,dx10);
 493             ty               = _mm256_mul_pd(fscal,dy10);
 494             tz               = _mm256_mul_pd(fscal,dz10);
 495
 496             /* Update vectorial force */
 497             fix1             = _mm256_add_pd(fix1,tx);
 498             fiy1             = _mm256_add_pd(fiy1,ty);
 499             fiz1             = _mm256_add_pd(fiz1,tz);
 500
 501             fjx0             = _mm256_add_pd(fjx0,tx);
 502             fjy0             = _mm256_add_pd(fjy0,ty);
 503             fjz0             = _mm256_add_pd(fjz0,tz);
 504
 505             }
 506
 507             /**************************
 508              * CALCULATE INTERACTIONS *
 509              **************************/
 510
 511             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 512             {
 513
 514             /* Compute parameters for interactions between i and j atoms */
 515             qq20             = _mm256_mul_pd(iq2,jq0);
 516
 517             /* REACTION-FIELD ELECTROSTATICS */
 518             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 519             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 520
 521             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 522
 523             /* Update potential sum for this i atom from the interaction with this j atom. */
 524             velec            = _mm256_and_pd(velec,cutoff_mask);
 525             velec            = _mm256_andnot_pd(dummy_mask,velec);
 526             velecsum         = _mm256_add_pd(velecsum,velec);
 527
 528             fscal            = felec;
 529
 530             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 531
 532             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 533
 534             /* Calculate temporary vectorial force */
 535             tx               = _mm256_mul_pd(fscal,dx20);
 536             ty               = _mm256_mul_pd(fscal,dy20);
 537             tz               = _mm256_mul_pd(fscal,dz20);
 538
 539             /* Update vectorial force */
 540             fix2             = _mm256_add_pd(fix2,tx);
 541             fiy2             = _mm256_add_pd(fiy2,ty);
 542             fiz2             = _mm256_add_pd(fiz2,tz);
 543
 544             fjx0             = _mm256_add_pd(fjx0,tx);
 545             fjy0             = _mm256_add_pd(fjy0,ty);
 546             fjz0             = _mm256_add_pd(fjz0,tz);
 547
 548             }
 549
 550             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 551             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 552             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 553             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 554
 555             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 556
 557             /* Inner loop uses 111 flops */
 558         }
 559
 560         /* End of innermost loop */
 561
 562         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 563                                                  f+i_coord_offset,fshift+i_shift_offset);
 564
 565         ggid                        = gid[iidx];
 566         /* Update potential energies */
 567         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 568
 569         /* Increment number of inner iterations */
 570         inneriter                  += j_index_end - j_index_start;
 571
 572         /* Outer loop uses 19 flops */
 573     }
 574
 575     /* Increment number of outer iterations */
 576     outeriter        += nri;
 577
 578     /* Update outer/inner flops */
 579
 580     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*111);
 581 }
 582 /*
 583  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW3P1_F_avx_256_double
 584  * Electrostatics interaction: ReactionField
 585  * VdW interaction:            None
 586  * Geometry:                   Water3-Particle
 587  * Calculate force/pot:        Force
 588  */
 589 void
 590 nb_kernel_ElecRFCut_VdwNone_GeomW3P1_F_avx_256_double
 591                     (t_nblist                    * gmx_restrict       nlist,
 592                      rvec                        * gmx_restrict          xx,
 593                      rvec                        * gmx_restrict          ff,
 594                      t_forcerec                  * gmx_restrict          fr,
 595                      t_mdatoms                   * gmx_restrict     mdatoms,
 596                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 597                      t_nrnb                      * gmx_restrict        nrnb)
 598 {
 599     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 600      * just 0 for non-waters.
 601      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 602      * jnr indices corresponding to data put in the four positions in the SIMD register.
 603      */
 604     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 605     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 606     int              jnrA,jnrB,jnrC,jnrD;
 607     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 608     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 609     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 610     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 611     real             rcutoff_scalar;
 612     real             *shiftvec,*fshift,*x,*f;
 613     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 614     real             scratch[4*DIM];
 615     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 616     real *           vdwioffsetptr0;
 617     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 618     real *           vdwioffsetptr1;
 619     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 620     real *           vdwioffsetptr2;
 621     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 622     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 623     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 624     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 625     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 626     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 627     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 628     real             *charge;
 629     __m256d          dummy_mask,cutoff_mask;
 630     __m128           tmpmask0,tmpmask1;
 631     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 632     __m256d          one     = _mm256_set1_pd(1.0);
 633     __m256d          two     = _mm256_set1_pd(2.0);
 634     x                = xx[0];
 635     f                = ff[0];
 636
 637     nri              = nlist->nri;
 638     iinr             = nlist->iinr;
 639     jindex           = nlist->jindex;
 640     jjnr             = nlist->jjnr;
 641     shiftidx         = nlist->shift;
 642     gid              = nlist->gid;
 643     shiftvec         = fr->shift_vec[0];
 644     fshift           = fr->fshift[0];
 645     facel            = _mm256_set1_pd(fr->epsfac);
 646     charge           = mdatoms->chargeA;
 647     krf              = _mm256_set1_pd(fr->ic->k_rf);
 648     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 649     crf              = _mm256_set1_pd(fr->ic->c_rf);
 650
 651     /* Setup water-specific parameters */
 652     inr              = nlist->iinr[0];
 653     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 654     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 655     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 656
 657     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 658     rcutoff_scalar   = fr->rcoulomb;
 659     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 660     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 661
 662     /* Avoid stupid compiler warnings */
 663     jnrA = jnrB = jnrC = jnrD = 0;
 664     j_coord_offsetA = 0;
 665     j_coord_offsetB = 0;
 666     j_coord_offsetC = 0;
 667     j_coord_offsetD = 0;
 668
 669     outeriter        = 0;
 670     inneriter        = 0;
 671
 672     for(iidx=0;iidx<4*DIM;iidx++)
 673     {
 674         scratch[iidx] = 0.0;
 675     }
 676
 677     /* Start outer loop over neighborlists */
 678     for(iidx=0; iidx<nri; iidx++)
 679     {
 680         /* Load shift vector for this list */
 681         i_shift_offset   = DIM*shiftidx[iidx];
 682
 683         /* Load limits for loop over neighbors */
 684         j_index_start    = jindex[iidx];
 685         j_index_end      = jindex[iidx+1];
 686
 687         /* Get outer coordinate index */
 688         inr              = iinr[iidx];
 689         i_coord_offset   = DIM*inr;
 690
 691         /* Load i particle coords and add shift vector */
 692         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 693                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 694
 695         fix0             = _mm256_setzero_pd();
 696         fiy0             = _mm256_setzero_pd();
 697         fiz0             = _mm256_setzero_pd();
 698         fix1             = _mm256_setzero_pd();
 699         fiy1             = _mm256_setzero_pd();
 700         fiz1             = _mm256_setzero_pd();
 701         fix2             = _mm256_setzero_pd();
 702         fiy2             = _mm256_setzero_pd();
 703         fiz2             = _mm256_setzero_pd();
 704
 705         /* Start inner kernel loop */
 706         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 707         {
 708
 709             /* Get j neighbor index, and coordinate index */
 710             jnrA             = jjnr[jidx];
 711             jnrB             = jjnr[jidx+1];
 712             jnrC             = jjnr[jidx+2];
 713             jnrD             = jjnr[jidx+3];
 714             j_coord_offsetA  = DIM*jnrA;
 715             j_coord_offsetB  = DIM*jnrB;
 716             j_coord_offsetC  = DIM*jnrC;
 717             j_coord_offsetD  = DIM*jnrD;
 718
 719             /* load j atom coordinates */
 720             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 721                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 722                                                  &jx0,&jy0,&jz0);
 723
 724             /* Calculate displacement vector */
 725             dx00             = _mm256_sub_pd(ix0,jx0);
 726             dy00             = _mm256_sub_pd(iy0,jy0);
 727             dz00             = _mm256_sub_pd(iz0,jz0);
 728             dx10             = _mm256_sub_pd(ix1,jx0);
 729             dy10             = _mm256_sub_pd(iy1,jy0);
 730             dz10             = _mm256_sub_pd(iz1,jz0);
 731             dx20             = _mm256_sub_pd(ix2,jx0);
 732             dy20             = _mm256_sub_pd(iy2,jy0);
 733             dz20             = _mm256_sub_pd(iz2,jz0);
 734
 735             /* Calculate squared distance and things based on it */
 736             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 737             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 738             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 739
 740             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 741             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 742             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 743
 744             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 745             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 746             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 747
 748             /* Load parameters for j particles */
 749             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 750                                                                  charge+jnrC+0,charge+jnrD+0);
 751
 752             fjx0             = _mm256_setzero_pd();
 753             fjy0             = _mm256_setzero_pd();
 754             fjz0             = _mm256_setzero_pd();
 755
 756             /**************************
 757              * CALCULATE INTERACTIONS *
 758              **************************/
 759
 760             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 761             {
 762
 763             /* Compute parameters for interactions between i and j atoms */
 764             qq00             = _mm256_mul_pd(iq0,jq0);
 765
 766             /* REACTION-FIELD ELECTROSTATICS */
 767             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 768
 769             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 770
 771             fscal            = felec;
 772
 773             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 774
 775             /* Calculate temporary vectorial force */
 776             tx               = _mm256_mul_pd(fscal,dx00);
 777             ty               = _mm256_mul_pd(fscal,dy00);
 778             tz               = _mm256_mul_pd(fscal,dz00);
 779
 780             /* Update vectorial force */
 781             fix0             = _mm256_add_pd(fix0,tx);
 782             fiy0             = _mm256_add_pd(fiy0,ty);
 783             fiz0             = _mm256_add_pd(fiz0,tz);
 784
 785             fjx0             = _mm256_add_pd(fjx0,tx);
 786             fjy0             = _mm256_add_pd(fjy0,ty);
 787             fjz0             = _mm256_add_pd(fjz0,tz);
 788
 789             }
 790
 791             /**************************
 792              * CALCULATE INTERACTIONS *
 793              **************************/
 794
 795             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 796             {
 797
 798             /* Compute parameters for interactions between i and j atoms */
 799             qq10             = _mm256_mul_pd(iq1,jq0);
 800
 801             /* REACTION-FIELD ELECTROSTATICS */
 802             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 803
 804             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 805
 806             fscal            = felec;
 807
 808             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 809
 810             /* Calculate temporary vectorial force */
 811             tx               = _mm256_mul_pd(fscal,dx10);
 812             ty               = _mm256_mul_pd(fscal,dy10);
 813             tz               = _mm256_mul_pd(fscal,dz10);
 814
 815             /* Update vectorial force */
 816             fix1             = _mm256_add_pd(fix1,tx);
 817             fiy1             = _mm256_add_pd(fiy1,ty);
 818             fiz1             = _mm256_add_pd(fiz1,tz);
 819
 820             fjx0             = _mm256_add_pd(fjx0,tx);
 821             fjy0             = _mm256_add_pd(fjy0,ty);
 822             fjz0             = _mm256_add_pd(fjz0,tz);
 823
 824             }
 825
 826             /**************************
 827              * CALCULATE INTERACTIONS *
 828              **************************/
 829
 830             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 831             {
 832
 833             /* Compute parameters for interactions between i and j atoms */
 834             qq20             = _mm256_mul_pd(iq2,jq0);
 835
 836             /* REACTION-FIELD ELECTROSTATICS */
 837             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 838
 839             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 840
 841             fscal            = felec;
 842
 843             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 844
 845             /* Calculate temporary vectorial force */
 846             tx               = _mm256_mul_pd(fscal,dx20);
 847             ty               = _mm256_mul_pd(fscal,dy20);
 848             tz               = _mm256_mul_pd(fscal,dz20);
 849
 850             /* Update vectorial force */
 851             fix2             = _mm256_add_pd(fix2,tx);
 852             fiy2             = _mm256_add_pd(fiy2,ty);
 853             fiz2             = _mm256_add_pd(fiz2,tz);
 854
 855             fjx0             = _mm256_add_pd(fjx0,tx);
 856             fjy0             = _mm256_add_pd(fjy0,ty);
 857             fjz0             = _mm256_add_pd(fjz0,tz);
 858
 859             }
 860
 861             fjptrA             = f+j_coord_offsetA;
 862             fjptrB             = f+j_coord_offsetB;
 863             fjptrC             = f+j_coord_offsetC;
 864             fjptrD             = f+j_coord_offsetD;
 865
 866             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 867
 868             /* Inner loop uses 93 flops */
 869         }
 870
 871         if(jidx<j_index_end)
 872         {
 873
 874             /* Get j neighbor index, and coordinate index */
 875             jnrlistA         = jjnr[jidx];
 876             jnrlistB         = jjnr[jidx+1];
 877             jnrlistC         = jjnr[jidx+2];
 878             jnrlistD         = jjnr[jidx+3];
 879             /* Sign of each element will be negative for non-real atoms.
 880              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 881              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 882              */
 883             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 884
 885             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 886             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 887             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 888
 889             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 890             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 891             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 892             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 893             j_coord_offsetA  = DIM*jnrA;
 894             j_coord_offsetB  = DIM*jnrB;
 895             j_coord_offsetC  = DIM*jnrC;
 896             j_coord_offsetD  = DIM*jnrD;
 897
 898             /* load j atom coordinates */
 899             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 900                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 901                                                  &jx0,&jy0,&jz0);
 902
 903             /* Calculate displacement vector */
 904             dx00             = _mm256_sub_pd(ix0,jx0);
 905             dy00             = _mm256_sub_pd(iy0,jy0);
 906             dz00             = _mm256_sub_pd(iz0,jz0);
 907             dx10             = _mm256_sub_pd(ix1,jx0);
 908             dy10             = _mm256_sub_pd(iy1,jy0);
 909             dz10             = _mm256_sub_pd(iz1,jz0);
 910             dx20             = _mm256_sub_pd(ix2,jx0);
 911             dy20             = _mm256_sub_pd(iy2,jy0);
 912             dz20             = _mm256_sub_pd(iz2,jz0);
 913
 914             /* Calculate squared distance and things based on it */
 915             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 916             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 917             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 918
 919             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 920             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 921             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 922
 923             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 924             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 925             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 926
 927             /* Load parameters for j particles */
 928             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 929                                                                  charge+jnrC+0,charge+jnrD+0);
 930
 931             fjx0             = _mm256_setzero_pd();
 932             fjy0             = _mm256_setzero_pd();
 933             fjz0             = _mm256_setzero_pd();
 934
 935             /**************************
 936              * CALCULATE INTERACTIONS *
 937              **************************/
 938
 939             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 940             {
 941
 942             /* Compute parameters for interactions between i and j atoms */
 943             qq00             = _mm256_mul_pd(iq0,jq0);
 944
 945             /* REACTION-FIELD ELECTROSTATICS */
 946             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 947
 948             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 949
 950             fscal            = felec;
 951
 952             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 953
 954             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 955
 956             /* Calculate temporary vectorial force */
 957             tx               = _mm256_mul_pd(fscal,dx00);
 958             ty               = _mm256_mul_pd(fscal,dy00);
 959             tz               = _mm256_mul_pd(fscal,dz00);
 960
 961             /* Update vectorial force */
 962             fix0             = _mm256_add_pd(fix0,tx);
 963             fiy0             = _mm256_add_pd(fiy0,ty);
 964             fiz0             = _mm256_add_pd(fiz0,tz);
 965
 966             fjx0             = _mm256_add_pd(fjx0,tx);
 967             fjy0             = _mm256_add_pd(fjy0,ty);
 968             fjz0             = _mm256_add_pd(fjz0,tz);
 969
 970             }
 971
 972             /**************************
 973              * CALCULATE INTERACTIONS *
 974              **************************/
 975
 976             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 977             {
 978
 979             /* Compute parameters for interactions between i and j atoms */
 980             qq10             = _mm256_mul_pd(iq1,jq0);
 981
 982             /* REACTION-FIELD ELECTROSTATICS */
 983             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 984
 985             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 986
 987             fscal            = felec;
 988
 989             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 990
 991             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 992
 993             /* Calculate temporary vectorial force */
 994             tx               = _mm256_mul_pd(fscal,dx10);
 995             ty               = _mm256_mul_pd(fscal,dy10);
 996             tz               = _mm256_mul_pd(fscal,dz10);
 997
 998             /* Update vectorial force */
 999             fix1             = _mm256_add_pd(fix1,tx);
1000             fiy1             = _mm256_add_pd(fiy1,ty);
1001             fiz1             = _mm256_add_pd(fiz1,tz);
1002
1003             fjx0             = _mm256_add_pd(fjx0,tx);
1004             fjy0             = _mm256_add_pd(fjy0,ty);
1005             fjz0             = _mm256_add_pd(fjz0,tz);
1006
1007             }
1008
1009             /**************************
1010              * CALCULATE INTERACTIONS *
1011              **************************/
1012
1013             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1014             {
1015
1016             /* Compute parameters for interactions between i and j atoms */
1017             qq20             = _mm256_mul_pd(iq2,jq0);
1018
1019             /* REACTION-FIELD ELECTROSTATICS */
1020             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1021
1022             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1023
1024             fscal            = felec;
1025
1026             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1027
1028             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1029
1030             /* Calculate temporary vectorial force */
1031             tx               = _mm256_mul_pd(fscal,dx20);
1032             ty               = _mm256_mul_pd(fscal,dy20);
1033             tz               = _mm256_mul_pd(fscal,dz20);
1034
1035             /* Update vectorial force */
1036             fix2             = _mm256_add_pd(fix2,tx);
1037             fiy2             = _mm256_add_pd(fiy2,ty);
1038             fiz2             = _mm256_add_pd(fiz2,tz);
1039
1040             fjx0             = _mm256_add_pd(fjx0,tx);
1041             fjy0             = _mm256_add_pd(fjy0,ty);
1042             fjz0             = _mm256_add_pd(fjz0,tz);
1043
1044             }
1045
1046             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1047             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1048             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1049             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1050
1051             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1052
1053             /* Inner loop uses 93 flops */
1054         }
1055
1056         /* End of innermost loop */
1057
1058         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1059                                                  f+i_coord_offset,fshift+i_shift_offset);
1060
1061         /* Increment number of inner iterations */
1062         inneriter                  += j_index_end - j_index_start;
1063
1064         /* Outer loop uses 18 flops */
1065     }
1066
1067     /* Increment number of outer iterations */
1068     outeriter        += nri;
1069
1070     /* Update outer/inner flops */
1071
1072     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*93);
1073 }