src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRF_VdwNone_GeomW4P1_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 "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_256_double.h"
  48 #include "kernelutil_x86_avx_256_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW4P1_VF_avx_256_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            None
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRF_VdwNone_GeomW4P1_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 *           vdwioffsetptr1;
  85     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     real *           vdwioffsetptr2;
  87     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     real *           vdwioffsetptr3;
  89     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  91     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  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     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 122     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 123     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 124
 125     /* Avoid stupid compiler warnings */
 126     jnrA = jnrB = jnrC = jnrD = 0;
 127     j_coord_offsetA = 0;
 128     j_coord_offsetB = 0;
 129     j_coord_offsetC = 0;
 130     j_coord_offsetD = 0;
 131
 132     outeriter        = 0;
 133     inneriter        = 0;
 134
 135     for(iidx=0;iidx<4*DIM;iidx++)
 136     {
 137         scratch[iidx] = 0.0;
 138     }
 139
 140     /* Start outer loop over neighborlists */
 141     for(iidx=0; iidx<nri; iidx++)
 142     {
 143         /* Load shift vector for this list */
 144         i_shift_offset   = DIM*shiftidx[iidx];
 145
 146         /* Load limits for loop over neighbors */
 147         j_index_start    = jindex[iidx];
 148         j_index_end      = jindex[iidx+1];
 149
 150         /* Get outer coordinate index */
 151         inr              = iinr[iidx];
 152         i_coord_offset   = DIM*inr;
 153
 154         /* Load i particle coords and add shift vector */
 155         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 156                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 157
 158         fix1             = _mm256_setzero_pd();
 159         fiy1             = _mm256_setzero_pd();
 160         fiz1             = _mm256_setzero_pd();
 161         fix2             = _mm256_setzero_pd();
 162         fiy2             = _mm256_setzero_pd();
 163         fiz2             = _mm256_setzero_pd();
 164         fix3             = _mm256_setzero_pd();
 165         fiy3             = _mm256_setzero_pd();
 166         fiz3             = _mm256_setzero_pd();
 167
 168         /* Reset potential sums */
 169         velecsum         = _mm256_setzero_pd();
 170
 171         /* Start inner kernel loop */
 172         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 173         {
 174
 175             /* Get j neighbor index, and coordinate index */
 176             jnrA             = jjnr[jidx];
 177             jnrB             = jjnr[jidx+1];
 178             jnrC             = jjnr[jidx+2];
 179             jnrD             = jjnr[jidx+3];
 180             j_coord_offsetA  = DIM*jnrA;
 181             j_coord_offsetB  = DIM*jnrB;
 182             j_coord_offsetC  = DIM*jnrC;
 183             j_coord_offsetD  = DIM*jnrD;
 184
 185             /* load j atom coordinates */
 186             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 187                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 188                                                  &jx0,&jy0,&jz0);
 189
 190             /* Calculate displacement vector */
 191             dx10             = _mm256_sub_pd(ix1,jx0);
 192             dy10             = _mm256_sub_pd(iy1,jy0);
 193             dz10             = _mm256_sub_pd(iz1,jz0);
 194             dx20             = _mm256_sub_pd(ix2,jx0);
 195             dy20             = _mm256_sub_pd(iy2,jy0);
 196             dz20             = _mm256_sub_pd(iz2,jz0);
 197             dx30             = _mm256_sub_pd(ix3,jx0);
 198             dy30             = _mm256_sub_pd(iy3,jy0);
 199             dz30             = _mm256_sub_pd(iz3,jz0);
 200
 201             /* Calculate squared distance and things based on it */
 202             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 203             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 204             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 205
 206             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 207             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 208             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 209
 210             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 211             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 212             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 213
 214             /* Load parameters for j particles */
 215             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 216                                                                  charge+jnrC+0,charge+jnrD+0);
 217
 218             fjx0             = _mm256_setzero_pd();
 219             fjy0             = _mm256_setzero_pd();
 220             fjz0             = _mm256_setzero_pd();
 221
 222             /**************************
 223              * CALCULATE INTERACTIONS *
 224              **************************/
 225
 226             /* Compute parameters for interactions between i and j atoms */
 227             qq10             = _mm256_mul_pd(iq1,jq0);
 228
 229             /* REACTION-FIELD ELECTROSTATICS */
 230             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 231             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 232
 233             /* Update potential sum for this i atom from the interaction with this j atom. */
 234             velecsum         = _mm256_add_pd(velecsum,velec);
 235
 236             fscal            = felec;
 237
 238             /* Calculate temporary vectorial force */
 239             tx               = _mm256_mul_pd(fscal,dx10);
 240             ty               = _mm256_mul_pd(fscal,dy10);
 241             tz               = _mm256_mul_pd(fscal,dz10);
 242
 243             /* Update vectorial force */
 244             fix1             = _mm256_add_pd(fix1,tx);
 245             fiy1             = _mm256_add_pd(fiy1,ty);
 246             fiz1             = _mm256_add_pd(fiz1,tz);
 247
 248             fjx0             = _mm256_add_pd(fjx0,tx);
 249             fjy0             = _mm256_add_pd(fjy0,ty);
 250             fjz0             = _mm256_add_pd(fjz0,tz);
 251
 252             /**************************
 253              * CALCULATE INTERACTIONS *
 254              **************************/
 255
 256             /* Compute parameters for interactions between i and j atoms */
 257             qq20             = _mm256_mul_pd(iq2,jq0);
 258
 259             /* REACTION-FIELD ELECTROSTATICS */
 260             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 261             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 262
 263             /* Update potential sum for this i atom from the interaction with this j atom. */
 264             velecsum         = _mm256_add_pd(velecsum,velec);
 265
 266             fscal            = felec;
 267
 268             /* Calculate temporary vectorial force */
 269             tx               = _mm256_mul_pd(fscal,dx20);
 270             ty               = _mm256_mul_pd(fscal,dy20);
 271             tz               = _mm256_mul_pd(fscal,dz20);
 272
 273             /* Update vectorial force */
 274             fix2             = _mm256_add_pd(fix2,tx);
 275             fiy2             = _mm256_add_pd(fiy2,ty);
 276             fiz2             = _mm256_add_pd(fiz2,tz);
 277
 278             fjx0             = _mm256_add_pd(fjx0,tx);
 279             fjy0             = _mm256_add_pd(fjy0,ty);
 280             fjz0             = _mm256_add_pd(fjz0,tz);
 281
 282             /**************************
 283              * CALCULATE INTERACTIONS *
 284              **************************/
 285
 286             /* Compute parameters for interactions between i and j atoms */
 287             qq30             = _mm256_mul_pd(iq3,jq0);
 288
 289             /* REACTION-FIELD ELECTROSTATICS */
 290             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
 291             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 292
 293             /* Update potential sum for this i atom from the interaction with this j atom. */
 294             velecsum         = _mm256_add_pd(velecsum,velec);
 295
 296             fscal            = felec;
 297
 298             /* Calculate temporary vectorial force */
 299             tx               = _mm256_mul_pd(fscal,dx30);
 300             ty               = _mm256_mul_pd(fscal,dy30);
 301             tz               = _mm256_mul_pd(fscal,dz30);
 302
 303             /* Update vectorial force */
 304             fix3             = _mm256_add_pd(fix3,tx);
 305             fiy3             = _mm256_add_pd(fiy3,ty);
 306             fiz3             = _mm256_add_pd(fiz3,tz);
 307
 308             fjx0             = _mm256_add_pd(fjx0,tx);
 309             fjy0             = _mm256_add_pd(fjy0,ty);
 310             fjz0             = _mm256_add_pd(fjz0,tz);
 311
 312             fjptrA             = f+j_coord_offsetA;
 313             fjptrB             = f+j_coord_offsetB;
 314             fjptrC             = f+j_coord_offsetC;
 315             fjptrD             = f+j_coord_offsetD;
 316
 317             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 318
 319             /* Inner loop uses 99 flops */
 320         }
 321
 322         if(jidx<j_index_end)
 323         {
 324
 325             /* Get j neighbor index, and coordinate index */
 326             jnrlistA         = jjnr[jidx];
 327             jnrlistB         = jjnr[jidx+1];
 328             jnrlistC         = jjnr[jidx+2];
 329             jnrlistD         = jjnr[jidx+3];
 330             /* Sign of each element will be negative for non-real atoms.
 331              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 332              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 333              */
 334             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 335
 336             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 337             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 338             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 339
 340             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 341             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 342             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 343             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 344             j_coord_offsetA  = DIM*jnrA;
 345             j_coord_offsetB  = DIM*jnrB;
 346             j_coord_offsetC  = DIM*jnrC;
 347             j_coord_offsetD  = DIM*jnrD;
 348
 349             /* load j atom coordinates */
 350             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 351                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 352                                                  &jx0,&jy0,&jz0);
 353
 354             /* Calculate displacement vector */
 355             dx10             = _mm256_sub_pd(ix1,jx0);
 356             dy10             = _mm256_sub_pd(iy1,jy0);
 357             dz10             = _mm256_sub_pd(iz1,jz0);
 358             dx20             = _mm256_sub_pd(ix2,jx0);
 359             dy20             = _mm256_sub_pd(iy2,jy0);
 360             dz20             = _mm256_sub_pd(iz2,jz0);
 361             dx30             = _mm256_sub_pd(ix3,jx0);
 362             dy30             = _mm256_sub_pd(iy3,jy0);
 363             dz30             = _mm256_sub_pd(iz3,jz0);
 364
 365             /* Calculate squared distance and things based on it */
 366             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 367             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 368             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 369
 370             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 371             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 372             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 373
 374             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 375             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 376             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 377
 378             /* Load parameters for j particles */
 379             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 380                                                                  charge+jnrC+0,charge+jnrD+0);
 381
 382             fjx0             = _mm256_setzero_pd();
 383             fjy0             = _mm256_setzero_pd();
 384             fjz0             = _mm256_setzero_pd();
 385
 386             /**************************
 387              * CALCULATE INTERACTIONS *
 388              **************************/
 389
 390             /* Compute parameters for interactions between i and j atoms */
 391             qq10             = _mm256_mul_pd(iq1,jq0);
 392
 393             /* REACTION-FIELD ELECTROSTATICS */
 394             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 395             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 396
 397             /* Update potential sum for this i atom from the interaction with this j atom. */
 398             velec            = _mm256_andnot_pd(dummy_mask,velec);
 399             velecsum         = _mm256_add_pd(velecsum,velec);
 400
 401             fscal            = felec;
 402
 403             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 404
 405             /* Calculate temporary vectorial force */
 406             tx               = _mm256_mul_pd(fscal,dx10);
 407             ty               = _mm256_mul_pd(fscal,dy10);
 408             tz               = _mm256_mul_pd(fscal,dz10);
 409
 410             /* Update vectorial force */
 411             fix1             = _mm256_add_pd(fix1,tx);
 412             fiy1             = _mm256_add_pd(fiy1,ty);
 413             fiz1             = _mm256_add_pd(fiz1,tz);
 414
 415             fjx0             = _mm256_add_pd(fjx0,tx);
 416             fjy0             = _mm256_add_pd(fjy0,ty);
 417             fjz0             = _mm256_add_pd(fjz0,tz);
 418
 419             /**************************
 420              * CALCULATE INTERACTIONS *
 421              **************************/
 422
 423             /* Compute parameters for interactions between i and j atoms */
 424             qq20             = _mm256_mul_pd(iq2,jq0);
 425
 426             /* REACTION-FIELD ELECTROSTATICS */
 427             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 428             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 429
 430             /* Update potential sum for this i atom from the interaction with this j atom. */
 431             velec            = _mm256_andnot_pd(dummy_mask,velec);
 432             velecsum         = _mm256_add_pd(velecsum,velec);
 433
 434             fscal            = felec;
 435
 436             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 437
 438             /* Calculate temporary vectorial force */
 439             tx               = _mm256_mul_pd(fscal,dx20);
 440             ty               = _mm256_mul_pd(fscal,dy20);
 441             tz               = _mm256_mul_pd(fscal,dz20);
 442
 443             /* Update vectorial force */
 444             fix2             = _mm256_add_pd(fix2,tx);
 445             fiy2             = _mm256_add_pd(fiy2,ty);
 446             fiz2             = _mm256_add_pd(fiz2,tz);
 447
 448             fjx0             = _mm256_add_pd(fjx0,tx);
 449             fjy0             = _mm256_add_pd(fjy0,ty);
 450             fjz0             = _mm256_add_pd(fjz0,tz);
 451
 452             /**************************
 453              * CALCULATE INTERACTIONS *
 454              **************************/
 455
 456             /* Compute parameters for interactions between i and j atoms */
 457             qq30             = _mm256_mul_pd(iq3,jq0);
 458
 459             /* REACTION-FIELD ELECTROSTATICS */
 460             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
 461             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 462
 463             /* Update potential sum for this i atom from the interaction with this j atom. */
 464             velec            = _mm256_andnot_pd(dummy_mask,velec);
 465             velecsum         = _mm256_add_pd(velecsum,velec);
 466
 467             fscal            = felec;
 468
 469             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 470
 471             /* Calculate temporary vectorial force */
 472             tx               = _mm256_mul_pd(fscal,dx30);
 473             ty               = _mm256_mul_pd(fscal,dy30);
 474             tz               = _mm256_mul_pd(fscal,dz30);
 475
 476             /* Update vectorial force */
 477             fix3             = _mm256_add_pd(fix3,tx);
 478             fiy3             = _mm256_add_pd(fiy3,ty);
 479             fiz3             = _mm256_add_pd(fiz3,tz);
 480
 481             fjx0             = _mm256_add_pd(fjx0,tx);
 482             fjy0             = _mm256_add_pd(fjy0,ty);
 483             fjz0             = _mm256_add_pd(fjz0,tz);
 484
 485             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 486             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 487             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 488             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 489
 490             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 491
 492             /* Inner loop uses 99 flops */
 493         }
 494
 495         /* End of innermost loop */
 496
 497         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 498                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
 499
 500         ggid                        = gid[iidx];
 501         /* Update potential energies */
 502         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 503
 504         /* Increment number of inner iterations */
 505         inneriter                  += j_index_end - j_index_start;
 506
 507         /* Outer loop uses 19 flops */
 508     }
 509
 510     /* Increment number of outer iterations */
 511     outeriter        += nri;
 512
 513     /* Update outer/inner flops */
 514
 515     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*99);
 516 }
 517 /*
 518  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW4P1_F_avx_256_double
 519  * Electrostatics interaction: ReactionField
 520  * VdW interaction:            None
 521  * Geometry:                   Water4-Particle
 522  * Calculate force/pot:        Force
 523  */
 524 void
 525 nb_kernel_ElecRF_VdwNone_GeomW4P1_F_avx_256_double
 526                     (t_nblist                    * gmx_restrict       nlist,
 527                      rvec                        * gmx_restrict          xx,
 528                      rvec                        * gmx_restrict          ff,
 529                      t_forcerec                  * gmx_restrict          fr,
 530                      t_mdatoms                   * gmx_restrict     mdatoms,
 531                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 532                      t_nrnb                      * gmx_restrict        nrnb)
 533 {
 534     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 535      * just 0 for non-waters.
 536      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 537      * jnr indices corresponding to data put in the four positions in the SIMD register.
 538      */
 539     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 540     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 541     int              jnrA,jnrB,jnrC,jnrD;
 542     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 543     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 544     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 545     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 546     real             rcutoff_scalar;
 547     real             *shiftvec,*fshift,*x,*f;
 548     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 549     real             scratch[4*DIM];
 550     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 551     real *           vdwioffsetptr1;
 552     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 553     real *           vdwioffsetptr2;
 554     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 555     real *           vdwioffsetptr3;
 556     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 557     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 558     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 559     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 560     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 561     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 562     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 563     real             *charge;
 564     __m256d          dummy_mask,cutoff_mask;
 565     __m128           tmpmask0,tmpmask1;
 566     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 567     __m256d          one     = _mm256_set1_pd(1.0);
 568     __m256d          two     = _mm256_set1_pd(2.0);
 569     x                = xx[0];
 570     f                = ff[0];
 571
 572     nri              = nlist->nri;
 573     iinr             = nlist->iinr;
 574     jindex           = nlist->jindex;
 575     jjnr             = nlist->jjnr;
 576     shiftidx         = nlist->shift;
 577     gid              = nlist->gid;
 578     shiftvec         = fr->shift_vec[0];
 579     fshift           = fr->fshift[0];
 580     facel            = _mm256_set1_pd(fr->epsfac);
 581     charge           = mdatoms->chargeA;
 582     krf              = _mm256_set1_pd(fr->ic->k_rf);
 583     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 584     crf              = _mm256_set1_pd(fr->ic->c_rf);
 585
 586     /* Setup water-specific parameters */
 587     inr              = nlist->iinr[0];
 588     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 589     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 590     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 591
 592     /* Avoid stupid compiler warnings */
 593     jnrA = jnrB = jnrC = jnrD = 0;
 594     j_coord_offsetA = 0;
 595     j_coord_offsetB = 0;
 596     j_coord_offsetC = 0;
 597     j_coord_offsetD = 0;
 598
 599     outeriter        = 0;
 600     inneriter        = 0;
 601
 602     for(iidx=0;iidx<4*DIM;iidx++)
 603     {
 604         scratch[iidx] = 0.0;
 605     }
 606
 607     /* Start outer loop over neighborlists */
 608     for(iidx=0; iidx<nri; iidx++)
 609     {
 610         /* Load shift vector for this list */
 611         i_shift_offset   = DIM*shiftidx[iidx];
 612
 613         /* Load limits for loop over neighbors */
 614         j_index_start    = jindex[iidx];
 615         j_index_end      = jindex[iidx+1];
 616
 617         /* Get outer coordinate index */
 618         inr              = iinr[iidx];
 619         i_coord_offset   = DIM*inr;
 620
 621         /* Load i particle coords and add shift vector */
 622         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 623                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 624
 625         fix1             = _mm256_setzero_pd();
 626         fiy1             = _mm256_setzero_pd();
 627         fiz1             = _mm256_setzero_pd();
 628         fix2             = _mm256_setzero_pd();
 629         fiy2             = _mm256_setzero_pd();
 630         fiz2             = _mm256_setzero_pd();
 631         fix3             = _mm256_setzero_pd();
 632         fiy3             = _mm256_setzero_pd();
 633         fiz3             = _mm256_setzero_pd();
 634
 635         /* Start inner kernel loop */
 636         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 637         {
 638
 639             /* Get j neighbor index, and coordinate index */
 640             jnrA             = jjnr[jidx];
 641             jnrB             = jjnr[jidx+1];
 642             jnrC             = jjnr[jidx+2];
 643             jnrD             = jjnr[jidx+3];
 644             j_coord_offsetA  = DIM*jnrA;
 645             j_coord_offsetB  = DIM*jnrB;
 646             j_coord_offsetC  = DIM*jnrC;
 647             j_coord_offsetD  = DIM*jnrD;
 648
 649             /* load j atom coordinates */
 650             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 651                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 652                                                  &jx0,&jy0,&jz0);
 653
 654             /* Calculate displacement vector */
 655             dx10             = _mm256_sub_pd(ix1,jx0);
 656             dy10             = _mm256_sub_pd(iy1,jy0);
 657             dz10             = _mm256_sub_pd(iz1,jz0);
 658             dx20             = _mm256_sub_pd(ix2,jx0);
 659             dy20             = _mm256_sub_pd(iy2,jy0);
 660             dz20             = _mm256_sub_pd(iz2,jz0);
 661             dx30             = _mm256_sub_pd(ix3,jx0);
 662             dy30             = _mm256_sub_pd(iy3,jy0);
 663             dz30             = _mm256_sub_pd(iz3,jz0);
 664
 665             /* Calculate squared distance and things based on it */
 666             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 667             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 668             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 669
 670             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 671             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 672             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 673
 674             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 675             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 676             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 677
 678             /* Load parameters for j particles */
 679             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 680                                                                  charge+jnrC+0,charge+jnrD+0);
 681
 682             fjx0             = _mm256_setzero_pd();
 683             fjy0             = _mm256_setzero_pd();
 684             fjz0             = _mm256_setzero_pd();
 685
 686             /**************************
 687              * CALCULATE INTERACTIONS *
 688              **************************/
 689
 690             /* Compute parameters for interactions between i and j atoms */
 691             qq10             = _mm256_mul_pd(iq1,jq0);
 692
 693             /* REACTION-FIELD ELECTROSTATICS */
 694             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 695
 696             fscal            = felec;
 697
 698             /* Calculate temporary vectorial force */
 699             tx               = _mm256_mul_pd(fscal,dx10);
 700             ty               = _mm256_mul_pd(fscal,dy10);
 701             tz               = _mm256_mul_pd(fscal,dz10);
 702
 703             /* Update vectorial force */
 704             fix1             = _mm256_add_pd(fix1,tx);
 705             fiy1             = _mm256_add_pd(fiy1,ty);
 706             fiz1             = _mm256_add_pd(fiz1,tz);
 707
 708             fjx0             = _mm256_add_pd(fjx0,tx);
 709             fjy0             = _mm256_add_pd(fjy0,ty);
 710             fjz0             = _mm256_add_pd(fjz0,tz);
 711
 712             /**************************
 713              * CALCULATE INTERACTIONS *
 714              **************************/
 715
 716             /* Compute parameters for interactions between i and j atoms */
 717             qq20             = _mm256_mul_pd(iq2,jq0);
 718
 719             /* REACTION-FIELD ELECTROSTATICS */
 720             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 721
 722             fscal            = felec;
 723
 724             /* Calculate temporary vectorial force */
 725             tx               = _mm256_mul_pd(fscal,dx20);
 726             ty               = _mm256_mul_pd(fscal,dy20);
 727             tz               = _mm256_mul_pd(fscal,dz20);
 728
 729             /* Update vectorial force */
 730             fix2             = _mm256_add_pd(fix2,tx);
 731             fiy2             = _mm256_add_pd(fiy2,ty);
 732             fiz2             = _mm256_add_pd(fiz2,tz);
 733
 734             fjx0             = _mm256_add_pd(fjx0,tx);
 735             fjy0             = _mm256_add_pd(fjy0,ty);
 736             fjz0             = _mm256_add_pd(fjz0,tz);
 737
 738             /**************************
 739              * CALCULATE INTERACTIONS *
 740              **************************/
 741
 742             /* Compute parameters for interactions between i and j atoms */
 743             qq30             = _mm256_mul_pd(iq3,jq0);
 744
 745             /* REACTION-FIELD ELECTROSTATICS */
 746             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 747
 748             fscal            = felec;
 749
 750             /* Calculate temporary vectorial force */
 751             tx               = _mm256_mul_pd(fscal,dx30);
 752             ty               = _mm256_mul_pd(fscal,dy30);
 753             tz               = _mm256_mul_pd(fscal,dz30);
 754
 755             /* Update vectorial force */
 756             fix3             = _mm256_add_pd(fix3,tx);
 757             fiy3             = _mm256_add_pd(fiy3,ty);
 758             fiz3             = _mm256_add_pd(fiz3,tz);
 759
 760             fjx0             = _mm256_add_pd(fjx0,tx);
 761             fjy0             = _mm256_add_pd(fjy0,ty);
 762             fjz0             = _mm256_add_pd(fjz0,tz);
 763
 764             fjptrA             = f+j_coord_offsetA;
 765             fjptrB             = f+j_coord_offsetB;
 766             fjptrC             = f+j_coord_offsetC;
 767             fjptrD             = f+j_coord_offsetD;
 768
 769             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 770
 771             /* Inner loop uses 84 flops */
 772         }
 773
 774         if(jidx<j_index_end)
 775         {
 776
 777             /* Get j neighbor index, and coordinate index */
 778             jnrlistA         = jjnr[jidx];
 779             jnrlistB         = jjnr[jidx+1];
 780             jnrlistC         = jjnr[jidx+2];
 781             jnrlistD         = jjnr[jidx+3];
 782             /* Sign of each element will be negative for non-real atoms.
 783              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 784              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 785              */
 786             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 787
 788             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 789             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 790             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 791
 792             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 793             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 794             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 795             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 796             j_coord_offsetA  = DIM*jnrA;
 797             j_coord_offsetB  = DIM*jnrB;
 798             j_coord_offsetC  = DIM*jnrC;
 799             j_coord_offsetD  = DIM*jnrD;
 800
 801             /* load j atom coordinates */
 802             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 803                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 804                                                  &jx0,&jy0,&jz0);
 805
 806             /* Calculate displacement vector */
 807             dx10             = _mm256_sub_pd(ix1,jx0);
 808             dy10             = _mm256_sub_pd(iy1,jy0);
 809             dz10             = _mm256_sub_pd(iz1,jz0);
 810             dx20             = _mm256_sub_pd(ix2,jx0);
 811             dy20             = _mm256_sub_pd(iy2,jy0);
 812             dz20             = _mm256_sub_pd(iz2,jz0);
 813             dx30             = _mm256_sub_pd(ix3,jx0);
 814             dy30             = _mm256_sub_pd(iy3,jy0);
 815             dz30             = _mm256_sub_pd(iz3,jz0);
 816
 817             /* Calculate squared distance and things based on it */
 818             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 819             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 820             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 821
 822             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 823             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 824             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 825
 826             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 827             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 828             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 829
 830             /* Load parameters for j particles */
 831             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 832                                                                  charge+jnrC+0,charge+jnrD+0);
 833
 834             fjx0             = _mm256_setzero_pd();
 835             fjy0             = _mm256_setzero_pd();
 836             fjz0             = _mm256_setzero_pd();
 837
 838             /**************************
 839              * CALCULATE INTERACTIONS *
 840              **************************/
 841
 842             /* Compute parameters for interactions between i and j atoms */
 843             qq10             = _mm256_mul_pd(iq1,jq0);
 844
 845             /* REACTION-FIELD ELECTROSTATICS */
 846             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 847
 848             fscal            = felec;
 849
 850             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 851
 852             /* Calculate temporary vectorial force */
 853             tx               = _mm256_mul_pd(fscal,dx10);
 854             ty               = _mm256_mul_pd(fscal,dy10);
 855             tz               = _mm256_mul_pd(fscal,dz10);
 856
 857             /* Update vectorial force */
 858             fix1             = _mm256_add_pd(fix1,tx);
 859             fiy1             = _mm256_add_pd(fiy1,ty);
 860             fiz1             = _mm256_add_pd(fiz1,tz);
 861
 862             fjx0             = _mm256_add_pd(fjx0,tx);
 863             fjy0             = _mm256_add_pd(fjy0,ty);
 864             fjz0             = _mm256_add_pd(fjz0,tz);
 865
 866             /**************************
 867              * CALCULATE INTERACTIONS *
 868              **************************/
 869
 870             /* Compute parameters for interactions between i and j atoms */
 871             qq20             = _mm256_mul_pd(iq2,jq0);
 872
 873             /* REACTION-FIELD ELECTROSTATICS */
 874             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 875
 876             fscal            = felec;
 877
 878             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 879
 880             /* Calculate temporary vectorial force */
 881             tx               = _mm256_mul_pd(fscal,dx20);
 882             ty               = _mm256_mul_pd(fscal,dy20);
 883             tz               = _mm256_mul_pd(fscal,dz20);
 884
 885             /* Update vectorial force */
 886             fix2             = _mm256_add_pd(fix2,tx);
 887             fiy2             = _mm256_add_pd(fiy2,ty);
 888             fiz2             = _mm256_add_pd(fiz2,tz);
 889
 890             fjx0             = _mm256_add_pd(fjx0,tx);
 891             fjy0             = _mm256_add_pd(fjy0,ty);
 892             fjz0             = _mm256_add_pd(fjz0,tz);
 893
 894             /**************************
 895              * CALCULATE INTERACTIONS *
 896              **************************/
 897
 898             /* Compute parameters for interactions between i and j atoms */
 899             qq30             = _mm256_mul_pd(iq3,jq0);
 900
 901             /* REACTION-FIELD ELECTROSTATICS */
 902             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 903
 904             fscal            = felec;
 905
 906             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 907
 908             /* Calculate temporary vectorial force */
 909             tx               = _mm256_mul_pd(fscal,dx30);
 910             ty               = _mm256_mul_pd(fscal,dy30);
 911             tz               = _mm256_mul_pd(fscal,dz30);
 912
 913             /* Update vectorial force */
 914             fix3             = _mm256_add_pd(fix3,tx);
 915             fiy3             = _mm256_add_pd(fiy3,ty);
 916             fiz3             = _mm256_add_pd(fiz3,tz);
 917
 918             fjx0             = _mm256_add_pd(fjx0,tx);
 919             fjy0             = _mm256_add_pd(fjy0,ty);
 920             fjz0             = _mm256_add_pd(fjz0,tz);
 921
 922             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 923             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 924             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 925             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 926
 927             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 928
 929             /* Inner loop uses 84 flops */
 930         }
 931
 932         /* End of innermost loop */
 933
 934         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 935                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
 936
 937         /* Increment number of inner iterations */
 938         inneriter                  += j_index_end - j_index_start;
 939
 940         /* Outer loop uses 18 flops */
 941     }
 942
 943     /* Increment number of outer iterations */
 944     outeriter        += nri;
 945
 946     /* Update outer/inner flops */
 947
 948     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*84);
 949 }