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) 2012,2013,2014,2015,2017, 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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_128_fma_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW3P1_VF_avx_128_fma_single
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            None
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRFCut_VdwNone_GeomW3P1_VF_avx_128_fma_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  89     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     __m128           dummy_mask,cutoff_mask;
  96     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  97     __m128           one     = _mm_set1_ps(1.0);
  98     __m128           two     = _mm_set1_ps(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     facel            = _mm_set1_ps(fr->ic->epsfac);
 111     charge           = mdatoms->chargeA;
 112     krf              = _mm_set1_ps(fr->ic->k_rf);
 113     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 114     crf              = _mm_set1_ps(fr->ic->c_rf);
 115
 116     /* Setup water-specific parameters */
 117     inr              = nlist->iinr[0];
 118     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 119     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 120     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 121
 122     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 123     rcutoff_scalar   = fr->ic->rcoulomb;
 124     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 125     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 126
 127     /* Avoid stupid compiler warnings */
 128     jnrA = jnrB = jnrC = jnrD = 0;
 129     j_coord_offsetA = 0;
 130     j_coord_offsetB = 0;
 131     j_coord_offsetC = 0;
 132     j_coord_offsetD = 0;
 133
 134     outeriter        = 0;
 135     inneriter        = 0;
 136
 137     for(iidx=0;iidx<4*DIM;iidx++)
 138     {
 139         scratch[iidx] = 0.0;
 140     }
 141
 142     /* Start outer loop over neighborlists */
 143     for(iidx=0; iidx<nri; iidx++)
 144     {
 145         /* Load shift vector for this list */
 146         i_shift_offset   = DIM*shiftidx[iidx];
 147
 148         /* Load limits for loop over neighbors */
 149         j_index_start    = jindex[iidx];
 150         j_index_end      = jindex[iidx+1];
 151
 152         /* Get outer coordinate index */
 153         inr              = iinr[iidx];
 154         i_coord_offset   = DIM*inr;
 155
 156         /* Load i particle coords and add shift vector */
 157         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 158                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 159
 160         fix0             = _mm_setzero_ps();
 161         fiy0             = _mm_setzero_ps();
 162         fiz0             = _mm_setzero_ps();
 163         fix1             = _mm_setzero_ps();
 164         fiy1             = _mm_setzero_ps();
 165         fiz1             = _mm_setzero_ps();
 166         fix2             = _mm_setzero_ps();
 167         fiy2             = _mm_setzero_ps();
 168         fiz2             = _mm_setzero_ps();
 169
 170         /* Reset potential sums */
 171         velecsum         = _mm_setzero_ps();
 172
 173         /* Start inner kernel loop */
 174         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 175         {
 176
 177             /* Get j neighbor index, and coordinate index */
 178             jnrA             = jjnr[jidx];
 179             jnrB             = jjnr[jidx+1];
 180             jnrC             = jjnr[jidx+2];
 181             jnrD             = jjnr[jidx+3];
 182             j_coord_offsetA  = DIM*jnrA;
 183             j_coord_offsetB  = DIM*jnrB;
 184             j_coord_offsetC  = DIM*jnrC;
 185             j_coord_offsetD  = DIM*jnrD;
 186
 187             /* load j atom coordinates */
 188             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 189                                               x+j_coord_offsetC,x+j_coord_offsetD,
 190                                               &jx0,&jy0,&jz0);
 191
 192             /* Calculate displacement vector */
 193             dx00             = _mm_sub_ps(ix0,jx0);
 194             dy00             = _mm_sub_ps(iy0,jy0);
 195             dz00             = _mm_sub_ps(iz0,jz0);
 196             dx10             = _mm_sub_ps(ix1,jx0);
 197             dy10             = _mm_sub_ps(iy1,jy0);
 198             dz10             = _mm_sub_ps(iz1,jz0);
 199             dx20             = _mm_sub_ps(ix2,jx0);
 200             dy20             = _mm_sub_ps(iy2,jy0);
 201             dz20             = _mm_sub_ps(iz2,jz0);
 202
 203             /* Calculate squared distance and things based on it */
 204             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 205             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 206             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 207
 208             rinv00           = avx128fma_invsqrt_f(rsq00);
 209             rinv10           = avx128fma_invsqrt_f(rsq10);
 210             rinv20           = avx128fma_invsqrt_f(rsq20);
 211
 212             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 213             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 214             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 215
 216             /* Load parameters for j particles */
 217             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 218                                                               charge+jnrC+0,charge+jnrD+0);
 219
 220             fjx0             = _mm_setzero_ps();
 221             fjy0             = _mm_setzero_ps();
 222             fjz0             = _mm_setzero_ps();
 223
 224             /**************************
 225              * CALCULATE INTERACTIONS *
 226              **************************/
 227
 228             if (gmx_mm_any_lt(rsq00,rcutoff2))
 229             {
 230
 231             /* Compute parameters for interactions between i and j atoms */
 232             qq00             = _mm_mul_ps(iq0,jq0);
 233
 234             /* REACTION-FIELD ELECTROSTATICS */
 235             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 236             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 237
 238             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 239
 240             /* Update potential sum for this i atom from the interaction with this j atom. */
 241             velec            = _mm_and_ps(velec,cutoff_mask);
 242             velecsum         = _mm_add_ps(velecsum,velec);
 243
 244             fscal            = felec;
 245
 246             fscal            = _mm_and_ps(fscal,cutoff_mask);
 247
 248              /* Update vectorial force */
 249             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 250             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 251             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 252
 253             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 254             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 255             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 256
 257             }
 258
 259             /**************************
 260              * CALCULATE INTERACTIONS *
 261              **************************/
 262
 263             if (gmx_mm_any_lt(rsq10,rcutoff2))
 264             {
 265
 266             /* Compute parameters for interactions between i and j atoms */
 267             qq10             = _mm_mul_ps(iq1,jq0);
 268
 269             /* REACTION-FIELD ELECTROSTATICS */
 270             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 271             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 272
 273             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 274
 275             /* Update potential sum for this i atom from the interaction with this j atom. */
 276             velec            = _mm_and_ps(velec,cutoff_mask);
 277             velecsum         = _mm_add_ps(velecsum,velec);
 278
 279             fscal            = felec;
 280
 281             fscal            = _mm_and_ps(fscal,cutoff_mask);
 282
 283              /* Update vectorial force */
 284             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 285             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 286             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 287
 288             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 289             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 290             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 291
 292             }
 293
 294             /**************************
 295              * CALCULATE INTERACTIONS *
 296              **************************/
 297
 298             if (gmx_mm_any_lt(rsq20,rcutoff2))
 299             {
 300
 301             /* Compute parameters for interactions between i and j atoms */
 302             qq20             = _mm_mul_ps(iq2,jq0);
 303
 304             /* REACTION-FIELD ELECTROSTATICS */
 305             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 306             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 307
 308             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 309
 310             /* Update potential sum for this i atom from the interaction with this j atom. */
 311             velec            = _mm_and_ps(velec,cutoff_mask);
 312             velecsum         = _mm_add_ps(velecsum,velec);
 313
 314             fscal            = felec;
 315
 316             fscal            = _mm_and_ps(fscal,cutoff_mask);
 317
 318              /* Update vectorial force */
 319             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 320             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 321             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 322
 323             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 324             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 325             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 326
 327             }
 328
 329             fjptrA             = f+j_coord_offsetA;
 330             fjptrB             = f+j_coord_offsetB;
 331             fjptrC             = f+j_coord_offsetC;
 332             fjptrD             = f+j_coord_offsetD;
 333
 334             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 335
 336             /* Inner loop uses 117 flops */
 337         }
 338
 339         if(jidx<j_index_end)
 340         {
 341
 342             /* Get j neighbor index, and coordinate index */
 343             jnrlistA         = jjnr[jidx];
 344             jnrlistB         = jjnr[jidx+1];
 345             jnrlistC         = jjnr[jidx+2];
 346             jnrlistD         = jjnr[jidx+3];
 347             /* Sign of each element will be negative for non-real atoms.
 348              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 349              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 350              */
 351             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 352             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 353             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 354             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 355             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 356             j_coord_offsetA  = DIM*jnrA;
 357             j_coord_offsetB  = DIM*jnrB;
 358             j_coord_offsetC  = DIM*jnrC;
 359             j_coord_offsetD  = DIM*jnrD;
 360
 361             /* load j atom coordinates */
 362             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 363                                               x+j_coord_offsetC,x+j_coord_offsetD,
 364                                               &jx0,&jy0,&jz0);
 365
 366             /* Calculate displacement vector */
 367             dx00             = _mm_sub_ps(ix0,jx0);
 368             dy00             = _mm_sub_ps(iy0,jy0);
 369             dz00             = _mm_sub_ps(iz0,jz0);
 370             dx10             = _mm_sub_ps(ix1,jx0);
 371             dy10             = _mm_sub_ps(iy1,jy0);
 372             dz10             = _mm_sub_ps(iz1,jz0);
 373             dx20             = _mm_sub_ps(ix2,jx0);
 374             dy20             = _mm_sub_ps(iy2,jy0);
 375             dz20             = _mm_sub_ps(iz2,jz0);
 376
 377             /* Calculate squared distance and things based on it */
 378             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 379             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 380             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 381
 382             rinv00           = avx128fma_invsqrt_f(rsq00);
 383             rinv10           = avx128fma_invsqrt_f(rsq10);
 384             rinv20           = avx128fma_invsqrt_f(rsq20);
 385
 386             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 387             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 388             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 389
 390             /* Load parameters for j particles */
 391             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 392                                                               charge+jnrC+0,charge+jnrD+0);
 393
 394             fjx0             = _mm_setzero_ps();
 395             fjy0             = _mm_setzero_ps();
 396             fjz0             = _mm_setzero_ps();
 397
 398             /**************************
 399              * CALCULATE INTERACTIONS *
 400              **************************/
 401
 402             if (gmx_mm_any_lt(rsq00,rcutoff2))
 403             {
 404
 405             /* Compute parameters for interactions between i and j atoms */
 406             qq00             = _mm_mul_ps(iq0,jq0);
 407
 408             /* REACTION-FIELD ELECTROSTATICS */
 409             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 410             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 411
 412             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 413
 414             /* Update potential sum for this i atom from the interaction with this j atom. */
 415             velec            = _mm_and_ps(velec,cutoff_mask);
 416             velec            = _mm_andnot_ps(dummy_mask,velec);
 417             velecsum         = _mm_add_ps(velecsum,velec);
 418
 419             fscal            = felec;
 420
 421             fscal            = _mm_and_ps(fscal,cutoff_mask);
 422
 423             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 424
 425              /* Update vectorial force */
 426             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 427             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 428             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 429
 430             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 431             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 432             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 433
 434             }
 435
 436             /**************************
 437              * CALCULATE INTERACTIONS *
 438              **************************/
 439
 440             if (gmx_mm_any_lt(rsq10,rcutoff2))
 441             {
 442
 443             /* Compute parameters for interactions between i and j atoms */
 444             qq10             = _mm_mul_ps(iq1,jq0);
 445
 446             /* REACTION-FIELD ELECTROSTATICS */
 447             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 448             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 449
 450             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 451
 452             /* Update potential sum for this i atom from the interaction with this j atom. */
 453             velec            = _mm_and_ps(velec,cutoff_mask);
 454             velec            = _mm_andnot_ps(dummy_mask,velec);
 455             velecsum         = _mm_add_ps(velecsum,velec);
 456
 457             fscal            = felec;
 458
 459             fscal            = _mm_and_ps(fscal,cutoff_mask);
 460
 461             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 462
 463              /* Update vectorial force */
 464             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 465             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 466             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 467
 468             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 469             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 470             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 471
 472             }
 473
 474             /**************************
 475              * CALCULATE INTERACTIONS *
 476              **************************/
 477
 478             if (gmx_mm_any_lt(rsq20,rcutoff2))
 479             {
 480
 481             /* Compute parameters for interactions between i and j atoms */
 482             qq20             = _mm_mul_ps(iq2,jq0);
 483
 484             /* REACTION-FIELD ELECTROSTATICS */
 485             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 486             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 487
 488             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 489
 490             /* Update potential sum for this i atom from the interaction with this j atom. */
 491             velec            = _mm_and_ps(velec,cutoff_mask);
 492             velec            = _mm_andnot_ps(dummy_mask,velec);
 493             velecsum         = _mm_add_ps(velecsum,velec);
 494
 495             fscal            = felec;
 496
 497             fscal            = _mm_and_ps(fscal,cutoff_mask);
 498
 499             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 500
 501              /* Update vectorial force */
 502             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 503             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 504             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 505
 506             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 507             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 508             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 509
 510             }
 511
 512             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 513             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 514             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 515             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 516
 517             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 518
 519             /* Inner loop uses 117 flops */
 520         }
 521
 522         /* End of innermost loop */
 523
 524         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 525                                               f+i_coord_offset,fshift+i_shift_offset);
 526
 527         ggid                        = gid[iidx];
 528         /* Update potential energies */
 529         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 530
 531         /* Increment number of inner iterations */
 532         inneriter                  += j_index_end - j_index_start;
 533
 534         /* Outer loop uses 19 flops */
 535     }
 536
 537     /* Increment number of outer iterations */
 538     outeriter        += nri;
 539
 540     /* Update outer/inner flops */
 541
 542     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*117);
 543 }
 544 /*
 545  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW3P1_F_avx_128_fma_single
 546  * Electrostatics interaction: ReactionField
 547  * VdW interaction:            None
 548  * Geometry:                   Water3-Particle
 549  * Calculate force/pot:        Force
 550  */
 551 void
 552 nb_kernel_ElecRFCut_VdwNone_GeomW3P1_F_avx_128_fma_single
 553                     (t_nblist                    * gmx_restrict       nlist,
 554                      rvec                        * gmx_restrict          xx,
 555                      rvec                        * gmx_restrict          ff,
 556                      struct t_forcerec           * gmx_restrict          fr,
 557                      t_mdatoms                   * gmx_restrict     mdatoms,
 558                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 559                      t_nrnb                      * gmx_restrict        nrnb)
 560 {
 561     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 562      * just 0 for non-waters.
 563      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 564      * jnr indices corresponding to data put in the four positions in the SIMD register.
 565      */
 566     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 567     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 568     int              jnrA,jnrB,jnrC,jnrD;
 569     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 570     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 571     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 572     real             rcutoff_scalar;
 573     real             *shiftvec,*fshift,*x,*f;
 574     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 575     real             scratch[4*DIM];
 576     __m128           fscal,rcutoff,rcutoff2,jidxall;
 577     int              vdwioffset0;
 578     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 579     int              vdwioffset1;
 580     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 581     int              vdwioffset2;
 582     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 583     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 584     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 585     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 586     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 587     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 588     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 589     real             *charge;
 590     __m128           dummy_mask,cutoff_mask;
 591     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 592     __m128           one     = _mm_set1_ps(1.0);
 593     __m128           two     = _mm_set1_ps(2.0);
 594     x                = xx[0];
 595     f                = ff[0];
 596
 597     nri              = nlist->nri;
 598     iinr             = nlist->iinr;
 599     jindex           = nlist->jindex;
 600     jjnr             = nlist->jjnr;
 601     shiftidx         = nlist->shift;
 602     gid              = nlist->gid;
 603     shiftvec         = fr->shift_vec[0];
 604     fshift           = fr->fshift[0];
 605     facel            = _mm_set1_ps(fr->ic->epsfac);
 606     charge           = mdatoms->chargeA;
 607     krf              = _mm_set1_ps(fr->ic->k_rf);
 608     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 609     crf              = _mm_set1_ps(fr->ic->c_rf);
 610
 611     /* Setup water-specific parameters */
 612     inr              = nlist->iinr[0];
 613     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 614     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 615     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 616
 617     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 618     rcutoff_scalar   = fr->ic->rcoulomb;
 619     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 620     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 621
 622     /* Avoid stupid compiler warnings */
 623     jnrA = jnrB = jnrC = jnrD = 0;
 624     j_coord_offsetA = 0;
 625     j_coord_offsetB = 0;
 626     j_coord_offsetC = 0;
 627     j_coord_offsetD = 0;
 628
 629     outeriter        = 0;
 630     inneriter        = 0;
 631
 632     for(iidx=0;iidx<4*DIM;iidx++)
 633     {
 634         scratch[iidx] = 0.0;
 635     }
 636
 637     /* Start outer loop over neighborlists */
 638     for(iidx=0; iidx<nri; iidx++)
 639     {
 640         /* Load shift vector for this list */
 641         i_shift_offset   = DIM*shiftidx[iidx];
 642
 643         /* Load limits for loop over neighbors */
 644         j_index_start    = jindex[iidx];
 645         j_index_end      = jindex[iidx+1];
 646
 647         /* Get outer coordinate index */
 648         inr              = iinr[iidx];
 649         i_coord_offset   = DIM*inr;
 650
 651         /* Load i particle coords and add shift vector */
 652         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 653                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 654
 655         fix0             = _mm_setzero_ps();
 656         fiy0             = _mm_setzero_ps();
 657         fiz0             = _mm_setzero_ps();
 658         fix1             = _mm_setzero_ps();
 659         fiy1             = _mm_setzero_ps();
 660         fiz1             = _mm_setzero_ps();
 661         fix2             = _mm_setzero_ps();
 662         fiy2             = _mm_setzero_ps();
 663         fiz2             = _mm_setzero_ps();
 664
 665         /* Start inner kernel loop */
 666         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 667         {
 668
 669             /* Get j neighbor index, and coordinate index */
 670             jnrA             = jjnr[jidx];
 671             jnrB             = jjnr[jidx+1];
 672             jnrC             = jjnr[jidx+2];
 673             jnrD             = jjnr[jidx+3];
 674             j_coord_offsetA  = DIM*jnrA;
 675             j_coord_offsetB  = DIM*jnrB;
 676             j_coord_offsetC  = DIM*jnrC;
 677             j_coord_offsetD  = DIM*jnrD;
 678
 679             /* load j atom coordinates */
 680             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 681                                               x+j_coord_offsetC,x+j_coord_offsetD,
 682                                               &jx0,&jy0,&jz0);
 683
 684             /* Calculate displacement vector */
 685             dx00             = _mm_sub_ps(ix0,jx0);
 686             dy00             = _mm_sub_ps(iy0,jy0);
 687             dz00             = _mm_sub_ps(iz0,jz0);
 688             dx10             = _mm_sub_ps(ix1,jx0);
 689             dy10             = _mm_sub_ps(iy1,jy0);
 690             dz10             = _mm_sub_ps(iz1,jz0);
 691             dx20             = _mm_sub_ps(ix2,jx0);
 692             dy20             = _mm_sub_ps(iy2,jy0);
 693             dz20             = _mm_sub_ps(iz2,jz0);
 694
 695             /* Calculate squared distance and things based on it */
 696             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 697             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 698             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 699
 700             rinv00           = avx128fma_invsqrt_f(rsq00);
 701             rinv10           = avx128fma_invsqrt_f(rsq10);
 702             rinv20           = avx128fma_invsqrt_f(rsq20);
 703
 704             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 705             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 706             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 707
 708             /* Load parameters for j particles */
 709             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 710                                                               charge+jnrC+0,charge+jnrD+0);
 711
 712             fjx0             = _mm_setzero_ps();
 713             fjy0             = _mm_setzero_ps();
 714             fjz0             = _mm_setzero_ps();
 715
 716             /**************************
 717              * CALCULATE INTERACTIONS *
 718              **************************/
 719
 720             if (gmx_mm_any_lt(rsq00,rcutoff2))
 721             {
 722
 723             /* Compute parameters for interactions between i and j atoms */
 724             qq00             = _mm_mul_ps(iq0,jq0);
 725
 726             /* REACTION-FIELD ELECTROSTATICS */
 727             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 728
 729             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 730
 731             fscal            = felec;
 732
 733             fscal            = _mm_and_ps(fscal,cutoff_mask);
 734
 735              /* Update vectorial force */
 736             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 737             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 738             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 739
 740             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 741             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 742             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 743
 744             }
 745
 746             /**************************
 747              * CALCULATE INTERACTIONS *
 748              **************************/
 749
 750             if (gmx_mm_any_lt(rsq10,rcutoff2))
 751             {
 752
 753             /* Compute parameters for interactions between i and j atoms */
 754             qq10             = _mm_mul_ps(iq1,jq0);
 755
 756             /* REACTION-FIELD ELECTROSTATICS */
 757             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 758
 759             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 760
 761             fscal            = felec;
 762
 763             fscal            = _mm_and_ps(fscal,cutoff_mask);
 764
 765              /* Update vectorial force */
 766             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 767             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 768             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 769
 770             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 771             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 772             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 773
 774             }
 775
 776             /**************************
 777              * CALCULATE INTERACTIONS *
 778              **************************/
 779
 780             if (gmx_mm_any_lt(rsq20,rcutoff2))
 781             {
 782
 783             /* Compute parameters for interactions between i and j atoms */
 784             qq20             = _mm_mul_ps(iq2,jq0);
 785
 786             /* REACTION-FIELD ELECTROSTATICS */
 787             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 788
 789             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 790
 791             fscal            = felec;
 792
 793             fscal            = _mm_and_ps(fscal,cutoff_mask);
 794
 795              /* Update vectorial force */
 796             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 797             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 798             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 799
 800             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 801             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 802             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 803
 804             }
 805
 806             fjptrA             = f+j_coord_offsetA;
 807             fjptrB             = f+j_coord_offsetB;
 808             fjptrC             = f+j_coord_offsetC;
 809             fjptrD             = f+j_coord_offsetD;
 810
 811             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 812
 813             /* Inner loop uses 99 flops */
 814         }
 815
 816         if(jidx<j_index_end)
 817         {
 818
 819             /* Get j neighbor index, and coordinate index */
 820             jnrlistA         = jjnr[jidx];
 821             jnrlistB         = jjnr[jidx+1];
 822             jnrlistC         = jjnr[jidx+2];
 823             jnrlistD         = jjnr[jidx+3];
 824             /* Sign of each element will be negative for non-real atoms.
 825              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 826              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 827              */
 828             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 829             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 830             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 831             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 832             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 833             j_coord_offsetA  = DIM*jnrA;
 834             j_coord_offsetB  = DIM*jnrB;
 835             j_coord_offsetC  = DIM*jnrC;
 836             j_coord_offsetD  = DIM*jnrD;
 837
 838             /* load j atom coordinates */
 839             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 840                                               x+j_coord_offsetC,x+j_coord_offsetD,
 841                                               &jx0,&jy0,&jz0);
 842
 843             /* Calculate displacement vector */
 844             dx00             = _mm_sub_ps(ix0,jx0);
 845             dy00             = _mm_sub_ps(iy0,jy0);
 846             dz00             = _mm_sub_ps(iz0,jz0);
 847             dx10             = _mm_sub_ps(ix1,jx0);
 848             dy10             = _mm_sub_ps(iy1,jy0);
 849             dz10             = _mm_sub_ps(iz1,jz0);
 850             dx20             = _mm_sub_ps(ix2,jx0);
 851             dy20             = _mm_sub_ps(iy2,jy0);
 852             dz20             = _mm_sub_ps(iz2,jz0);
 853
 854             /* Calculate squared distance and things based on it */
 855             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 856             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 857             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 858
 859             rinv00           = avx128fma_invsqrt_f(rsq00);
 860             rinv10           = avx128fma_invsqrt_f(rsq10);
 861             rinv20           = avx128fma_invsqrt_f(rsq20);
 862
 863             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 864             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 865             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 866
 867             /* Load parameters for j particles */
 868             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 869                                                               charge+jnrC+0,charge+jnrD+0);
 870
 871             fjx0             = _mm_setzero_ps();
 872             fjy0             = _mm_setzero_ps();
 873             fjz0             = _mm_setzero_ps();
 874
 875             /**************************
 876              * CALCULATE INTERACTIONS *
 877              **************************/
 878
 879             if (gmx_mm_any_lt(rsq00,rcutoff2))
 880             {
 881
 882             /* Compute parameters for interactions between i and j atoms */
 883             qq00             = _mm_mul_ps(iq0,jq0);
 884
 885             /* REACTION-FIELD ELECTROSTATICS */
 886             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 887
 888             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 889
 890             fscal            = felec;
 891
 892             fscal            = _mm_and_ps(fscal,cutoff_mask);
 893
 894             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 895
 896              /* Update vectorial force */
 897             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 898             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 899             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 900
 901             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 902             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 903             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 904
 905             }
 906
 907             /**************************
 908              * CALCULATE INTERACTIONS *
 909              **************************/
 910
 911             if (gmx_mm_any_lt(rsq10,rcutoff2))
 912             {
 913
 914             /* Compute parameters for interactions between i and j atoms */
 915             qq10             = _mm_mul_ps(iq1,jq0);
 916
 917             /* REACTION-FIELD ELECTROSTATICS */
 918             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 919
 920             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 921
 922             fscal            = felec;
 923
 924             fscal            = _mm_and_ps(fscal,cutoff_mask);
 925
 926             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 927
 928              /* Update vectorial force */
 929             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 930             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 931             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 932
 933             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 934             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 935             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 936
 937             }
 938
 939             /**************************
 940              * CALCULATE INTERACTIONS *
 941              **************************/
 942
 943             if (gmx_mm_any_lt(rsq20,rcutoff2))
 944             {
 945
 946             /* Compute parameters for interactions between i and j atoms */
 947             qq20             = _mm_mul_ps(iq2,jq0);
 948
 949             /* REACTION-FIELD ELECTROSTATICS */
 950             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 951
 952             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 953
 954             fscal            = felec;
 955
 956             fscal            = _mm_and_ps(fscal,cutoff_mask);
 957
 958             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 959
 960              /* Update vectorial force */
 961             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 962             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 963             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 964
 965             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 966             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 967             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 968
 969             }
 970
 971             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 972             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 973             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 974             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 975
 976             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 977
 978             /* Inner loop uses 99 flops */
 979         }
 980
 981         /* End of innermost loop */
 982
 983         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 984                                               f+i_coord_offset,fshift+i_shift_offset);
 985
 986         /* Increment number of inner iterations */
 987         inneriter                  += j_index_end - j_index_start;
 988
 989         /* Outer loop uses 18 flops */
 990     }
 991
 992     /* Increment number of outer iterations */
 993     outeriter        += nri;
 994
 995     /* Update outer/inner flops */
 996
 997     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*99);
 998 }