src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRFCut_VdwNone_GeomW3P1_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW3P1_VF_avx_128_fma_single
  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_128_fma_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, 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              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     __m128           dummy_mask,cutoff_mask;
  97     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  98     __m128           one     = _mm_set1_ps(1.0);
  99     __m128           two     = _mm_set1_ps(2.0);
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     facel            = _mm_set1_ps(fr->epsfac);
 112     charge           = mdatoms->chargeA;
 113     krf              = _mm_set1_ps(fr->ic->k_rf);
 114     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 115     crf              = _mm_set1_ps(fr->ic->c_rf);
 116
 117     /* Setup water-specific parameters */
 118     inr              = nlist->iinr[0];
 119     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 120     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 121     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 122
 123     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 124     rcutoff_scalar   = fr->rcoulomb;
 125     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 126     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 127
 128     /* Avoid stupid compiler warnings */
 129     jnrA = jnrB = jnrC = jnrD = 0;
 130     j_coord_offsetA = 0;
 131     j_coord_offsetB = 0;
 132     j_coord_offsetC = 0;
 133     j_coord_offsetD = 0;
 134
 135     outeriter        = 0;
 136     inneriter        = 0;
 137
 138     for(iidx=0;iidx<4*DIM;iidx++)
 139     {
 140         scratch[iidx] = 0.0;
 141     }
 142
 143     /* Start outer loop over neighborlists */
 144     for(iidx=0; iidx<nri; iidx++)
 145     {
 146         /* Load shift vector for this list */
 147         i_shift_offset   = DIM*shiftidx[iidx];
 148
 149         /* Load limits for loop over neighbors */
 150         j_index_start    = jindex[iidx];
 151         j_index_end      = jindex[iidx+1];
 152
 153         /* Get outer coordinate index */
 154         inr              = iinr[iidx];
 155         i_coord_offset   = DIM*inr;
 156
 157         /* Load i particle coords and add shift vector */
 158         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 159                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 160
 161         fix0             = _mm_setzero_ps();
 162         fiy0             = _mm_setzero_ps();
 163         fiz0             = _mm_setzero_ps();
 164         fix1             = _mm_setzero_ps();
 165         fiy1             = _mm_setzero_ps();
 166         fiz1             = _mm_setzero_ps();
 167         fix2             = _mm_setzero_ps();
 168         fiy2             = _mm_setzero_ps();
 169         fiz2             = _mm_setzero_ps();
 170
 171         /* Reset potential sums */
 172         velecsum         = _mm_setzero_ps();
 173
 174         /* Start inner kernel loop */
 175         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 176         {
 177
 178             /* Get j neighbor index, and coordinate index */
 179             jnrA             = jjnr[jidx];
 180             jnrB             = jjnr[jidx+1];
 181             jnrC             = jjnr[jidx+2];
 182             jnrD             = jjnr[jidx+3];
 183             j_coord_offsetA  = DIM*jnrA;
 184             j_coord_offsetB  = DIM*jnrB;
 185             j_coord_offsetC  = DIM*jnrC;
 186             j_coord_offsetD  = DIM*jnrD;
 187
 188             /* load j atom coordinates */
 189             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 190                                               x+j_coord_offsetC,x+j_coord_offsetD,
 191                                               &jx0,&jy0,&jz0);
 192
 193             /* Calculate displacement vector */
 194             dx00             = _mm_sub_ps(ix0,jx0);
 195             dy00             = _mm_sub_ps(iy0,jy0);
 196             dz00             = _mm_sub_ps(iz0,jz0);
 197             dx10             = _mm_sub_ps(ix1,jx0);
 198             dy10             = _mm_sub_ps(iy1,jy0);
 199             dz10             = _mm_sub_ps(iz1,jz0);
 200             dx20             = _mm_sub_ps(ix2,jx0);
 201             dy20             = _mm_sub_ps(iy2,jy0);
 202             dz20             = _mm_sub_ps(iz2,jz0);
 203
 204             /* Calculate squared distance and things based on it */
 205             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 206             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 207             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 208
 209             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 210             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 211             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 212
 213             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 214             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 215             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 216
 217             /* Load parameters for j particles */
 218             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 219                                                               charge+jnrC+0,charge+jnrD+0);
 220
 221             fjx0             = _mm_setzero_ps();
 222             fjy0             = _mm_setzero_ps();
 223             fjz0             = _mm_setzero_ps();
 224
 225             /**************************
 226              * CALCULATE INTERACTIONS *
 227              **************************/
 228
 229             if (gmx_mm_any_lt(rsq00,rcutoff2))
 230             {
 231
 232             /* Compute parameters for interactions between i and j atoms */
 233             qq00             = _mm_mul_ps(iq0,jq0);
 234
 235             /* REACTION-FIELD ELECTROSTATICS */
 236             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 237             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 238
 239             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 240
 241             /* Update potential sum for this i atom from the interaction with this j atom. */
 242             velec            = _mm_and_ps(velec,cutoff_mask);
 243             velecsum         = _mm_add_ps(velecsum,velec);
 244
 245             fscal            = felec;
 246
 247             fscal            = _mm_and_ps(fscal,cutoff_mask);
 248
 249              /* Update vectorial force */
 250             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 251             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 252             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 253
 254             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 255             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 256             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 257
 258             }
 259
 260             /**************************
 261              * CALCULATE INTERACTIONS *
 262              **************************/
 263
 264             if (gmx_mm_any_lt(rsq10,rcutoff2))
 265             {
 266
 267             /* Compute parameters for interactions between i and j atoms */
 268             qq10             = _mm_mul_ps(iq1,jq0);
 269
 270             /* REACTION-FIELD ELECTROSTATICS */
 271             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 272             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 273
 274             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 275
 276             /* Update potential sum for this i atom from the interaction with this j atom. */
 277             velec            = _mm_and_ps(velec,cutoff_mask);
 278             velecsum         = _mm_add_ps(velecsum,velec);
 279
 280             fscal            = felec;
 281
 282             fscal            = _mm_and_ps(fscal,cutoff_mask);
 283
 284              /* Update vectorial force */
 285             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 286             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 287             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 288
 289             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 290             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 291             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 292
 293             }
 294
 295             /**************************
 296              * CALCULATE INTERACTIONS *
 297              **************************/
 298
 299             if (gmx_mm_any_lt(rsq20,rcutoff2))
 300             {
 301
 302             /* Compute parameters for interactions between i and j atoms */
 303             qq20             = _mm_mul_ps(iq2,jq0);
 304
 305             /* REACTION-FIELD ELECTROSTATICS */
 306             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 307             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 308
 309             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 310
 311             /* Update potential sum for this i atom from the interaction with this j atom. */
 312             velec            = _mm_and_ps(velec,cutoff_mask);
 313             velecsum         = _mm_add_ps(velecsum,velec);
 314
 315             fscal            = felec;
 316
 317             fscal            = _mm_and_ps(fscal,cutoff_mask);
 318
 319              /* Update vectorial force */
 320             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 321             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 322             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 323
 324             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 325             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 326             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 327
 328             }
 329
 330             fjptrA             = f+j_coord_offsetA;
 331             fjptrB             = f+j_coord_offsetB;
 332             fjptrC             = f+j_coord_offsetC;
 333             fjptrD             = f+j_coord_offsetD;
 334
 335             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 336
 337             /* Inner loop uses 117 flops */
 338         }
 339
 340         if(jidx<j_index_end)
 341         {
 342
 343             /* Get j neighbor index, and coordinate index */
 344             jnrlistA         = jjnr[jidx];
 345             jnrlistB         = jjnr[jidx+1];
 346             jnrlistC         = jjnr[jidx+2];
 347             jnrlistD         = jjnr[jidx+3];
 348             /* Sign of each element will be negative for non-real atoms.
 349              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 350              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 351              */
 352             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 353             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 354             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 355             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 356             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 357             j_coord_offsetA  = DIM*jnrA;
 358             j_coord_offsetB  = DIM*jnrB;
 359             j_coord_offsetC  = DIM*jnrC;
 360             j_coord_offsetD  = DIM*jnrD;
 361
 362             /* load j atom coordinates */
 363             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 364                                               x+j_coord_offsetC,x+j_coord_offsetD,
 365                                               &jx0,&jy0,&jz0);
 366
 367             /* Calculate displacement vector */
 368             dx00             = _mm_sub_ps(ix0,jx0);
 369             dy00             = _mm_sub_ps(iy0,jy0);
 370             dz00             = _mm_sub_ps(iz0,jz0);
 371             dx10             = _mm_sub_ps(ix1,jx0);
 372             dy10             = _mm_sub_ps(iy1,jy0);
 373             dz10             = _mm_sub_ps(iz1,jz0);
 374             dx20             = _mm_sub_ps(ix2,jx0);
 375             dy20             = _mm_sub_ps(iy2,jy0);
 376             dz20             = _mm_sub_ps(iz2,jz0);
 377
 378             /* Calculate squared distance and things based on it */
 379             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 380             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 381             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 382
 383             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 384             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 385             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 386
 387             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 388             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 389             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 390
 391             /* Load parameters for j particles */
 392             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 393                                                               charge+jnrC+0,charge+jnrD+0);
 394
 395             fjx0             = _mm_setzero_ps();
 396             fjy0             = _mm_setzero_ps();
 397             fjz0             = _mm_setzero_ps();
 398
 399             /**************************
 400              * CALCULATE INTERACTIONS *
 401              **************************/
 402
 403             if (gmx_mm_any_lt(rsq00,rcutoff2))
 404             {
 405
 406             /* Compute parameters for interactions between i and j atoms */
 407             qq00             = _mm_mul_ps(iq0,jq0);
 408
 409             /* REACTION-FIELD ELECTROSTATICS */
 410             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 411             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 412
 413             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 414
 415             /* Update potential sum for this i atom from the interaction with this j atom. */
 416             velec            = _mm_and_ps(velec,cutoff_mask);
 417             velec            = _mm_andnot_ps(dummy_mask,velec);
 418             velecsum         = _mm_add_ps(velecsum,velec);
 419
 420             fscal            = felec;
 421
 422             fscal            = _mm_and_ps(fscal,cutoff_mask);
 423
 424             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 425
 426              /* Update vectorial force */
 427             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 428             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 429             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 430
 431             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 432             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 433             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 434
 435             }
 436
 437             /**************************
 438              * CALCULATE INTERACTIONS *
 439              **************************/
 440
 441             if (gmx_mm_any_lt(rsq10,rcutoff2))
 442             {
 443
 444             /* Compute parameters for interactions between i and j atoms */
 445             qq10             = _mm_mul_ps(iq1,jq0);
 446
 447             /* REACTION-FIELD ELECTROSTATICS */
 448             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 449             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 450
 451             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 452
 453             /* Update potential sum for this i atom from the interaction with this j atom. */
 454             velec            = _mm_and_ps(velec,cutoff_mask);
 455             velec            = _mm_andnot_ps(dummy_mask,velec);
 456             velecsum         = _mm_add_ps(velecsum,velec);
 457
 458             fscal            = felec;
 459
 460             fscal            = _mm_and_ps(fscal,cutoff_mask);
 461
 462             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 463
 464              /* Update vectorial force */
 465             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 466             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 467             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 468
 469             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 470             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 471             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 472
 473             }
 474
 475             /**************************
 476              * CALCULATE INTERACTIONS *
 477              **************************/
 478
 479             if (gmx_mm_any_lt(rsq20,rcutoff2))
 480             {
 481
 482             /* Compute parameters for interactions between i and j atoms */
 483             qq20             = _mm_mul_ps(iq2,jq0);
 484
 485             /* REACTION-FIELD ELECTROSTATICS */
 486             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 487             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 488
 489             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 490
 491             /* Update potential sum for this i atom from the interaction with this j atom. */
 492             velec            = _mm_and_ps(velec,cutoff_mask);
 493             velec            = _mm_andnot_ps(dummy_mask,velec);
 494             velecsum         = _mm_add_ps(velecsum,velec);
 495
 496             fscal            = felec;
 497
 498             fscal            = _mm_and_ps(fscal,cutoff_mask);
 499
 500             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 501
 502              /* Update vectorial force */
 503             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 504             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 505             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 506
 507             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 508             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 509             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 510
 511             }
 512
 513             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 514             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 515             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 516             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 517
 518             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 519
 520             /* Inner loop uses 117 flops */
 521         }
 522
 523         /* End of innermost loop */
 524
 525         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 526                                               f+i_coord_offset,fshift+i_shift_offset);
 527
 528         ggid                        = gid[iidx];
 529         /* Update potential energies */
 530         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 531
 532         /* Increment number of inner iterations */
 533         inneriter                  += j_index_end - j_index_start;
 534
 535         /* Outer loop uses 19 flops */
 536     }
 537
 538     /* Increment number of outer iterations */
 539     outeriter        += nri;
 540
 541     /* Update outer/inner flops */
 542
 543     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*117);
 544 }
 545 /*
 546  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW3P1_F_avx_128_fma_single
 547  * Electrostatics interaction: ReactionField
 548  * VdW interaction:            None
 549  * Geometry:                   Water3-Particle
 550  * Calculate force/pot:        Force
 551  */
 552 void
 553 nb_kernel_ElecRFCut_VdwNone_GeomW3P1_F_avx_128_fma_single
 554                     (t_nblist                    * gmx_restrict       nlist,
 555                      rvec                        * gmx_restrict          xx,
 556                      rvec                        * gmx_restrict          ff,
 557                      t_forcerec                  * gmx_restrict          fr,
 558                      t_mdatoms                   * gmx_restrict     mdatoms,
 559                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 560                      t_nrnb                      * gmx_restrict        nrnb)
 561 {
 562     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 563      * just 0 for non-waters.
 564      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 565      * jnr indices corresponding to data put in the four positions in the SIMD register.
 566      */
 567     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 568     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 569     int              jnrA,jnrB,jnrC,jnrD;
 570     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 571     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 572     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 573     real             rcutoff_scalar;
 574     real             *shiftvec,*fshift,*x,*f;
 575     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 576     real             scratch[4*DIM];
 577     __m128           fscal,rcutoff,rcutoff2,jidxall;
 578     int              vdwioffset0;
 579     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 580     int              vdwioffset1;
 581     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 582     int              vdwioffset2;
 583     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 584     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 585     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 586     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 587     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 588     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 589     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 590     real             *charge;
 591     __m128           dummy_mask,cutoff_mask;
 592     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 593     __m128           one     = _mm_set1_ps(1.0);
 594     __m128           two     = _mm_set1_ps(2.0);
 595     x                = xx[0];
 596     f                = ff[0];
 597
 598     nri              = nlist->nri;
 599     iinr             = nlist->iinr;
 600     jindex           = nlist->jindex;
 601     jjnr             = nlist->jjnr;
 602     shiftidx         = nlist->shift;
 603     gid              = nlist->gid;
 604     shiftvec         = fr->shift_vec[0];
 605     fshift           = fr->fshift[0];
 606     facel            = _mm_set1_ps(fr->epsfac);
 607     charge           = mdatoms->chargeA;
 608     krf              = _mm_set1_ps(fr->ic->k_rf);
 609     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 610     crf              = _mm_set1_ps(fr->ic->c_rf);
 611
 612     /* Setup water-specific parameters */
 613     inr              = nlist->iinr[0];
 614     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 615     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 616     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 617
 618     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 619     rcutoff_scalar   = fr->rcoulomb;
 620     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 621     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 622
 623     /* Avoid stupid compiler warnings */
 624     jnrA = jnrB = jnrC = jnrD = 0;
 625     j_coord_offsetA = 0;
 626     j_coord_offsetB = 0;
 627     j_coord_offsetC = 0;
 628     j_coord_offsetD = 0;
 629
 630     outeriter        = 0;
 631     inneriter        = 0;
 632
 633     for(iidx=0;iidx<4*DIM;iidx++)
 634     {
 635         scratch[iidx] = 0.0;
 636     }
 637
 638     /* Start outer loop over neighborlists */
 639     for(iidx=0; iidx<nri; iidx++)
 640     {
 641         /* Load shift vector for this list */
 642         i_shift_offset   = DIM*shiftidx[iidx];
 643
 644         /* Load limits for loop over neighbors */
 645         j_index_start    = jindex[iidx];
 646         j_index_end      = jindex[iidx+1];
 647
 648         /* Get outer coordinate index */
 649         inr              = iinr[iidx];
 650         i_coord_offset   = DIM*inr;
 651
 652         /* Load i particle coords and add shift vector */
 653         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 654                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 655
 656         fix0             = _mm_setzero_ps();
 657         fiy0             = _mm_setzero_ps();
 658         fiz0             = _mm_setzero_ps();
 659         fix1             = _mm_setzero_ps();
 660         fiy1             = _mm_setzero_ps();
 661         fiz1             = _mm_setzero_ps();
 662         fix2             = _mm_setzero_ps();
 663         fiy2             = _mm_setzero_ps();
 664         fiz2             = _mm_setzero_ps();
 665
 666         /* Start inner kernel loop */
 667         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 668         {
 669
 670             /* Get j neighbor index, and coordinate index */
 671             jnrA             = jjnr[jidx];
 672             jnrB             = jjnr[jidx+1];
 673             jnrC             = jjnr[jidx+2];
 674             jnrD             = jjnr[jidx+3];
 675             j_coord_offsetA  = DIM*jnrA;
 676             j_coord_offsetB  = DIM*jnrB;
 677             j_coord_offsetC  = DIM*jnrC;
 678             j_coord_offsetD  = DIM*jnrD;
 679
 680             /* load j atom coordinates */
 681             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 682                                               x+j_coord_offsetC,x+j_coord_offsetD,
 683                                               &jx0,&jy0,&jz0);
 684
 685             /* Calculate displacement vector */
 686             dx00             = _mm_sub_ps(ix0,jx0);
 687             dy00             = _mm_sub_ps(iy0,jy0);
 688             dz00             = _mm_sub_ps(iz0,jz0);
 689             dx10             = _mm_sub_ps(ix1,jx0);
 690             dy10             = _mm_sub_ps(iy1,jy0);
 691             dz10             = _mm_sub_ps(iz1,jz0);
 692             dx20             = _mm_sub_ps(ix2,jx0);
 693             dy20             = _mm_sub_ps(iy2,jy0);
 694             dz20             = _mm_sub_ps(iz2,jz0);
 695
 696             /* Calculate squared distance and things based on it */
 697             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 698             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 699             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 700
 701             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 702             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 703             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 704
 705             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 706             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 707             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 708
 709             /* Load parameters for j particles */
 710             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 711                                                               charge+jnrC+0,charge+jnrD+0);
 712
 713             fjx0             = _mm_setzero_ps();
 714             fjy0             = _mm_setzero_ps();
 715             fjz0             = _mm_setzero_ps();
 716
 717             /**************************
 718              * CALCULATE INTERACTIONS *
 719              **************************/
 720
 721             if (gmx_mm_any_lt(rsq00,rcutoff2))
 722             {
 723
 724             /* Compute parameters for interactions between i and j atoms */
 725             qq00             = _mm_mul_ps(iq0,jq0);
 726
 727             /* REACTION-FIELD ELECTROSTATICS */
 728             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 729
 730             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 731
 732             fscal            = felec;
 733
 734             fscal            = _mm_and_ps(fscal,cutoff_mask);
 735
 736              /* Update vectorial force */
 737             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 738             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 739             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 740
 741             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 742             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 743             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 744
 745             }
 746
 747             /**************************
 748              * CALCULATE INTERACTIONS *
 749              **************************/
 750
 751             if (gmx_mm_any_lt(rsq10,rcutoff2))
 752             {
 753
 754             /* Compute parameters for interactions between i and j atoms */
 755             qq10             = _mm_mul_ps(iq1,jq0);
 756
 757             /* REACTION-FIELD ELECTROSTATICS */
 758             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 759
 760             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 761
 762             fscal            = felec;
 763
 764             fscal            = _mm_and_ps(fscal,cutoff_mask);
 765
 766              /* Update vectorial force */
 767             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 768             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 769             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 770
 771             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 772             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 773             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 774
 775             }
 776
 777             /**************************
 778              * CALCULATE INTERACTIONS *
 779              **************************/
 780
 781             if (gmx_mm_any_lt(rsq20,rcutoff2))
 782             {
 783
 784             /* Compute parameters for interactions between i and j atoms */
 785             qq20             = _mm_mul_ps(iq2,jq0);
 786
 787             /* REACTION-FIELD ELECTROSTATICS */
 788             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 789
 790             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 791
 792             fscal            = felec;
 793
 794             fscal            = _mm_and_ps(fscal,cutoff_mask);
 795
 796              /* Update vectorial force */
 797             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 798             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 799             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 800
 801             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 802             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 803             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 804
 805             }
 806
 807             fjptrA             = f+j_coord_offsetA;
 808             fjptrB             = f+j_coord_offsetB;
 809             fjptrC             = f+j_coord_offsetC;
 810             fjptrD             = f+j_coord_offsetD;
 811
 812             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 813
 814             /* Inner loop uses 99 flops */
 815         }
 816
 817         if(jidx<j_index_end)
 818         {
 819
 820             /* Get j neighbor index, and coordinate index */
 821             jnrlistA         = jjnr[jidx];
 822             jnrlistB         = jjnr[jidx+1];
 823             jnrlistC         = jjnr[jidx+2];
 824             jnrlistD         = jjnr[jidx+3];
 825             /* Sign of each element will be negative for non-real atoms.
 826              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 827              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 828              */
 829             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 830             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 831             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 832             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 833             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 834             j_coord_offsetA  = DIM*jnrA;
 835             j_coord_offsetB  = DIM*jnrB;
 836             j_coord_offsetC  = DIM*jnrC;
 837             j_coord_offsetD  = DIM*jnrD;
 838
 839             /* load j atom coordinates */
 840             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 841                                               x+j_coord_offsetC,x+j_coord_offsetD,
 842                                               &jx0,&jy0,&jz0);
 843
 844             /* Calculate displacement vector */
 845             dx00             = _mm_sub_ps(ix0,jx0);
 846             dy00             = _mm_sub_ps(iy0,jy0);
 847             dz00             = _mm_sub_ps(iz0,jz0);
 848             dx10             = _mm_sub_ps(ix1,jx0);
 849             dy10             = _mm_sub_ps(iy1,jy0);
 850             dz10             = _mm_sub_ps(iz1,jz0);
 851             dx20             = _mm_sub_ps(ix2,jx0);
 852             dy20             = _mm_sub_ps(iy2,jy0);
 853             dz20             = _mm_sub_ps(iz2,jz0);
 854
 855             /* Calculate squared distance and things based on it */
 856             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 857             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 858             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 859
 860             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 861             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 862             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 863
 864             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 865             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 866             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 867
 868             /* Load parameters for j particles */
 869             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 870                                                               charge+jnrC+0,charge+jnrD+0);
 871
 872             fjx0             = _mm_setzero_ps();
 873             fjy0             = _mm_setzero_ps();
 874             fjz0             = _mm_setzero_ps();
 875
 876             /**************************
 877              * CALCULATE INTERACTIONS *
 878              **************************/
 879
 880             if (gmx_mm_any_lt(rsq00,rcutoff2))
 881             {
 882
 883             /* Compute parameters for interactions between i and j atoms */
 884             qq00             = _mm_mul_ps(iq0,jq0);
 885
 886             /* REACTION-FIELD ELECTROSTATICS */
 887             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 888
 889             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 890
 891             fscal            = felec;
 892
 893             fscal            = _mm_and_ps(fscal,cutoff_mask);
 894
 895             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 896
 897              /* Update vectorial force */
 898             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 899             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 900             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 901
 902             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 903             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 904             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 905
 906             }
 907
 908             /**************************
 909              * CALCULATE INTERACTIONS *
 910              **************************/
 911
 912             if (gmx_mm_any_lt(rsq10,rcutoff2))
 913             {
 914
 915             /* Compute parameters for interactions between i and j atoms */
 916             qq10             = _mm_mul_ps(iq1,jq0);
 917
 918             /* REACTION-FIELD ELECTROSTATICS */
 919             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 920
 921             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 922
 923             fscal            = felec;
 924
 925             fscal            = _mm_and_ps(fscal,cutoff_mask);
 926
 927             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 928
 929              /* Update vectorial force */
 930             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 931             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 932             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 933
 934             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 935             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 936             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 937
 938             }
 939
 940             /**************************
 941              * CALCULATE INTERACTIONS *
 942              **************************/
 943
 944             if (gmx_mm_any_lt(rsq20,rcutoff2))
 945             {
 946
 947             /* Compute parameters for interactions between i and j atoms */
 948             qq20             = _mm_mul_ps(iq2,jq0);
 949
 950             /* REACTION-FIELD ELECTROSTATICS */
 951             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 952
 953             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 954
 955             fscal            = felec;
 956
 957             fscal            = _mm_and_ps(fscal,cutoff_mask);
 958
 959             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 960
 961              /* Update vectorial force */
 962             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 963             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 964             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 965
 966             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 967             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 968             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 969
 970             }
 971
 972             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 973             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 974             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 975             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 976
 977             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 978
 979             /* Inner loop uses 99 flops */
 980         }
 981
 982         /* End of innermost loop */
 983
 984         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 985                                               f+i_coord_offset,fshift+i_shift_offset);
 986
 987         /* Increment number of inner iterations */
 988         inneriter                  += j_index_end - j_index_start;
 989
 990         /* Outer loop uses 18 flops */
 991     }
 992
 993     /* Increment number of outer iterations */
 994     outeriter        += nri;
 995
 996     /* Update outer/inner flops */
 997
 998     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*99);
 999 }