src/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecCoul_VdwLJ_GeomW3P1_avx_128_fma_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_128_fma_single.h"
  34 #include "kernelutil_x86_avx_128_fma_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_avx_128_fma_single
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_avx_128_fma_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  76     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  77     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  78     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  79     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  80     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  87     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  88     __m128           dummy_mask,cutoff_mask;
  89     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  90     __m128           one     = _mm_set1_ps(1.0);
  91     __m128           two     = _mm_set1_ps(2.0);
  92     x                = xx[0];
  93     f                = ff[0];
  94
  95     nri              = nlist->nri;
  96     iinr             = nlist->iinr;
  97     jindex           = nlist->jindex;
  98     jjnr             = nlist->jjnr;
  99     shiftidx         = nlist->shift;
 100     gid              = nlist->gid;
 101     shiftvec         = fr->shift_vec[0];
 102     fshift           = fr->fshift[0];
 103     facel            = _mm_set1_ps(fr->epsfac);
 104     charge           = mdatoms->chargeA;
 105     nvdwtype         = fr->ntype;
 106     vdwparam         = fr->nbfp;
 107     vdwtype          = mdatoms->typeA;
 108
 109     /* Setup water-specific parameters */
 110     inr              = nlist->iinr[0];
 111     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 112     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 113     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 114     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 115
 116     /* Avoid stupid compiler warnings */
 117     jnrA = jnrB = jnrC = jnrD = 0;
 118     j_coord_offsetA = 0;
 119     j_coord_offsetB = 0;
 120     j_coord_offsetC = 0;
 121     j_coord_offsetD = 0;
 122
 123     outeriter        = 0;
 124     inneriter        = 0;
 125
 126     for(iidx=0;iidx<4*DIM;iidx++)
 127     {
 128         scratch[iidx] = 0.0;
 129     }
 130
 131     /* Start outer loop over neighborlists */
 132     for(iidx=0; iidx<nri; iidx++)
 133     {
 134         /* Load shift vector for this list */
 135         i_shift_offset   = DIM*shiftidx[iidx];
 136
 137         /* Load limits for loop over neighbors */
 138         j_index_start    = jindex[iidx];
 139         j_index_end      = jindex[iidx+1];
 140
 141         /* Get outer coordinate index */
 142         inr              = iinr[iidx];
 143         i_coord_offset   = DIM*inr;
 144
 145         /* Load i particle coords and add shift vector */
 146         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 147                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 148
 149         fix0             = _mm_setzero_ps();
 150         fiy0             = _mm_setzero_ps();
 151         fiz0             = _mm_setzero_ps();
 152         fix1             = _mm_setzero_ps();
 153         fiy1             = _mm_setzero_ps();
 154         fiz1             = _mm_setzero_ps();
 155         fix2             = _mm_setzero_ps();
 156         fiy2             = _mm_setzero_ps();
 157         fiz2             = _mm_setzero_ps();
 158
 159         /* Reset potential sums */
 160         velecsum         = _mm_setzero_ps();
 161         vvdwsum          = _mm_setzero_ps();
 162
 163         /* Start inner kernel loop */
 164         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 165         {
 166
 167             /* Get j neighbor index, and coordinate index */
 168             jnrA             = jjnr[jidx];
 169             jnrB             = jjnr[jidx+1];
 170             jnrC             = jjnr[jidx+2];
 171             jnrD             = jjnr[jidx+3];
 172             j_coord_offsetA  = DIM*jnrA;
 173             j_coord_offsetB  = DIM*jnrB;
 174             j_coord_offsetC  = DIM*jnrC;
 175             j_coord_offsetD  = DIM*jnrD;
 176
 177             /* load j atom coordinates */
 178             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 179                                               x+j_coord_offsetC,x+j_coord_offsetD,
 180                                               &jx0,&jy0,&jz0);
 181
 182             /* Calculate displacement vector */
 183             dx00             = _mm_sub_ps(ix0,jx0);
 184             dy00             = _mm_sub_ps(iy0,jy0);
 185             dz00             = _mm_sub_ps(iz0,jz0);
 186             dx10             = _mm_sub_ps(ix1,jx0);
 187             dy10             = _mm_sub_ps(iy1,jy0);
 188             dz10             = _mm_sub_ps(iz1,jz0);
 189             dx20             = _mm_sub_ps(ix2,jx0);
 190             dy20             = _mm_sub_ps(iy2,jy0);
 191             dz20             = _mm_sub_ps(iz2,jz0);
 192
 193             /* Calculate squared distance and things based on it */
 194             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 195             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 196             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 197
 198             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 199             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 200             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 201
 202             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 203             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 204             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 205
 206             /* Load parameters for j particles */
 207             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 208                                                               charge+jnrC+0,charge+jnrD+0);
 209             vdwjidx0A        = 2*vdwtype[jnrA+0];
 210             vdwjidx0B        = 2*vdwtype[jnrB+0];
 211             vdwjidx0C        = 2*vdwtype[jnrC+0];
 212             vdwjidx0D        = 2*vdwtype[jnrD+0];
 213
 214             fjx0             = _mm_setzero_ps();
 215             fjy0             = _mm_setzero_ps();
 216             fjz0             = _mm_setzero_ps();
 217
 218             /**************************
 219              * CALCULATE INTERACTIONS *
 220              **************************/
 221
 222             /* Compute parameters for interactions between i and j atoms */
 223             qq00             = _mm_mul_ps(iq0,jq0);
 224             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 225                                          vdwparam+vdwioffset0+vdwjidx0B,
 226                                          vdwparam+vdwioffset0+vdwjidx0C,
 227                                          vdwparam+vdwioffset0+vdwjidx0D,
 228                                          &c6_00,&c12_00);
 229
 230             /* COULOMB ELECTROSTATICS */
 231             velec            = _mm_mul_ps(qq00,rinv00);
 232             felec            = _mm_mul_ps(velec,rinvsq00);
 233
 234             /* LENNARD-JONES DISPERSION/REPULSION */
 235
 236             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 237             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 238             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 239             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 240             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 241
 242             /* Update potential sum for this i atom from the interaction with this j atom. */
 243             velecsum         = _mm_add_ps(velecsum,velec);
 244             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 245
 246             fscal            = _mm_add_ps(felec,fvdw);
 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              * CALCULATE INTERACTIONS *
 259              **************************/
 260
 261             /* Compute parameters for interactions between i and j atoms */
 262             qq10             = _mm_mul_ps(iq1,jq0);
 263
 264             /* COULOMB ELECTROSTATICS */
 265             velec            = _mm_mul_ps(qq10,rinv10);
 266             felec            = _mm_mul_ps(velec,rinvsq10);
 267
 268             /* Update potential sum for this i atom from the interaction with this j atom. */
 269             velecsum         = _mm_add_ps(velecsum,velec);
 270
 271             fscal            = felec;
 272
 273              /* Update vectorial force */
 274             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 275             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 276             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 277
 278             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 279             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 280             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 281
 282             /**************************
 283              * CALCULATE INTERACTIONS *
 284              **************************/
 285
 286             /* Compute parameters for interactions between i and j atoms */
 287             qq20             = _mm_mul_ps(iq2,jq0);
 288
 289             /* COULOMB ELECTROSTATICS */
 290             velec            = _mm_mul_ps(qq20,rinv20);
 291             felec            = _mm_mul_ps(velec,rinvsq20);
 292
 293             /* Update potential sum for this i atom from the interaction with this j atom. */
 294             velecsum         = _mm_add_ps(velecsum,velec);
 295
 296             fscal            = felec;
 297
 298              /* Update vectorial force */
 299             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 300             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 301             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 302
 303             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 304             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 305             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 306
 307             fjptrA             = f+j_coord_offsetA;
 308             fjptrB             = f+j_coord_offsetB;
 309             fjptrC             = f+j_coord_offsetC;
 310             fjptrD             = f+j_coord_offsetD;
 311
 312             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 313
 314             /* Inner loop uses 105 flops */
 315         }
 316
 317         if(jidx<j_index_end)
 318         {
 319
 320             /* Get j neighbor index, and coordinate index */
 321             jnrlistA         = jjnr[jidx];
 322             jnrlistB         = jjnr[jidx+1];
 323             jnrlistC         = jjnr[jidx+2];
 324             jnrlistD         = jjnr[jidx+3];
 325             /* Sign of each element will be negative for non-real atoms.
 326              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 327              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 328              */
 329             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 330             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 331             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 332             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 333             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 334             j_coord_offsetA  = DIM*jnrA;
 335             j_coord_offsetB  = DIM*jnrB;
 336             j_coord_offsetC  = DIM*jnrC;
 337             j_coord_offsetD  = DIM*jnrD;
 338
 339             /* load j atom coordinates */
 340             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 341                                               x+j_coord_offsetC,x+j_coord_offsetD,
 342                                               &jx0,&jy0,&jz0);
 343
 344             /* Calculate displacement vector */
 345             dx00             = _mm_sub_ps(ix0,jx0);
 346             dy00             = _mm_sub_ps(iy0,jy0);
 347             dz00             = _mm_sub_ps(iz0,jz0);
 348             dx10             = _mm_sub_ps(ix1,jx0);
 349             dy10             = _mm_sub_ps(iy1,jy0);
 350             dz10             = _mm_sub_ps(iz1,jz0);
 351             dx20             = _mm_sub_ps(ix2,jx0);
 352             dy20             = _mm_sub_ps(iy2,jy0);
 353             dz20             = _mm_sub_ps(iz2,jz0);
 354
 355             /* Calculate squared distance and things based on it */
 356             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 357             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 358             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 359
 360             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 361             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 362             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 363
 364             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 365             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 366             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 367
 368             /* Load parameters for j particles */
 369             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 370                                                               charge+jnrC+0,charge+jnrD+0);
 371             vdwjidx0A        = 2*vdwtype[jnrA+0];
 372             vdwjidx0B        = 2*vdwtype[jnrB+0];
 373             vdwjidx0C        = 2*vdwtype[jnrC+0];
 374             vdwjidx0D        = 2*vdwtype[jnrD+0];
 375
 376             fjx0             = _mm_setzero_ps();
 377             fjy0             = _mm_setzero_ps();
 378             fjz0             = _mm_setzero_ps();
 379
 380             /**************************
 381              * CALCULATE INTERACTIONS *
 382              **************************/
 383
 384             /* Compute parameters for interactions between i and j atoms */
 385             qq00             = _mm_mul_ps(iq0,jq0);
 386             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 387                                          vdwparam+vdwioffset0+vdwjidx0B,
 388                                          vdwparam+vdwioffset0+vdwjidx0C,
 389                                          vdwparam+vdwioffset0+vdwjidx0D,
 390                                          &c6_00,&c12_00);
 391
 392             /* COULOMB ELECTROSTATICS */
 393             velec            = _mm_mul_ps(qq00,rinv00);
 394             felec            = _mm_mul_ps(velec,rinvsq00);
 395
 396             /* LENNARD-JONES DISPERSION/REPULSION */
 397
 398             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 399             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 400             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 401             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 402             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 403
 404             /* Update potential sum for this i atom from the interaction with this j atom. */
 405             velec            = _mm_andnot_ps(dummy_mask,velec);
 406             velecsum         = _mm_add_ps(velecsum,velec);
 407             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 408             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 409
 410             fscal            = _mm_add_ps(felec,fvdw);
 411
 412             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 413
 414              /* Update vectorial force */
 415             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 416             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 417             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 418
 419             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 420             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 421             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 422
 423             /**************************
 424              * CALCULATE INTERACTIONS *
 425              **************************/
 426
 427             /* Compute parameters for interactions between i and j atoms */
 428             qq10             = _mm_mul_ps(iq1,jq0);
 429
 430             /* COULOMB ELECTROSTATICS */
 431             velec            = _mm_mul_ps(qq10,rinv10);
 432             felec            = _mm_mul_ps(velec,rinvsq10);
 433
 434             /* Update potential sum for this i atom from the interaction with this j atom. */
 435             velec            = _mm_andnot_ps(dummy_mask,velec);
 436             velecsum         = _mm_add_ps(velecsum,velec);
 437
 438             fscal            = felec;
 439
 440             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 441
 442              /* Update vectorial force */
 443             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 444             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 445             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 446
 447             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 448             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 449             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 450
 451             /**************************
 452              * CALCULATE INTERACTIONS *
 453              **************************/
 454
 455             /* Compute parameters for interactions between i and j atoms */
 456             qq20             = _mm_mul_ps(iq2,jq0);
 457
 458             /* COULOMB ELECTROSTATICS */
 459             velec            = _mm_mul_ps(qq20,rinv20);
 460             felec            = _mm_mul_ps(velec,rinvsq20);
 461
 462             /* Update potential sum for this i atom from the interaction with this j atom. */
 463             velec            = _mm_andnot_ps(dummy_mask,velec);
 464             velecsum         = _mm_add_ps(velecsum,velec);
 465
 466             fscal            = felec;
 467
 468             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 469
 470              /* Update vectorial force */
 471             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 472             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 473             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 474
 475             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 476             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 477             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 478
 479             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 480             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 481             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 482             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 483
 484             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 485
 486             /* Inner loop uses 105 flops */
 487         }
 488
 489         /* End of innermost loop */
 490
 491         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 492                                               f+i_coord_offset,fshift+i_shift_offset);
 493
 494         ggid                        = gid[iidx];
 495         /* Update potential energies */
 496         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 497         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 498
 499         /* Increment number of inner iterations */
 500         inneriter                  += j_index_end - j_index_start;
 501
 502         /* Outer loop uses 20 flops */
 503     }
 504
 505     /* Increment number of outer iterations */
 506     outeriter        += nri;
 507
 508     /* Update outer/inner flops */
 509
 510     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*105);
 511 }
 512 /*
 513  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_avx_128_fma_single
 514  * Electrostatics interaction: Coulomb
 515  * VdW interaction:            LennardJones
 516  * Geometry:                   Water3-Particle
 517  * Calculate force/pot:        Force
 518  */
 519 void
 520 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_avx_128_fma_single
 521                     (t_nblist * gmx_restrict                nlist,
 522                      rvec * gmx_restrict                    xx,
 523                      rvec * gmx_restrict                    ff,
 524                      t_forcerec * gmx_restrict              fr,
 525                      t_mdatoms * gmx_restrict               mdatoms,
 526                      nb_kernel_data_t * gmx_restrict        kernel_data,
 527                      t_nrnb * gmx_restrict                  nrnb)
 528 {
 529     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 530      * just 0 for non-waters.
 531      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 532      * jnr indices corresponding to data put in the four positions in the SIMD register.
 533      */
 534     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 535     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 536     int              jnrA,jnrB,jnrC,jnrD;
 537     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 538     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 539     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 540     real             rcutoff_scalar;
 541     real             *shiftvec,*fshift,*x,*f;
 542     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 543     real             scratch[4*DIM];
 544     __m128           fscal,rcutoff,rcutoff2,jidxall;
 545     int              vdwioffset0;
 546     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 547     int              vdwioffset1;
 548     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 549     int              vdwioffset2;
 550     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 551     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 552     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 553     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 554     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 555     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 556     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 557     real             *charge;
 558     int              nvdwtype;
 559     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 560     int              *vdwtype;
 561     real             *vdwparam;
 562     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 563     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 564     __m128           dummy_mask,cutoff_mask;
 565     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 566     __m128           one     = _mm_set1_ps(1.0);
 567     __m128           two     = _mm_set1_ps(2.0);
 568     x                = xx[0];
 569     f                = ff[0];
 570
 571     nri              = nlist->nri;
 572     iinr             = nlist->iinr;
 573     jindex           = nlist->jindex;
 574     jjnr             = nlist->jjnr;
 575     shiftidx         = nlist->shift;
 576     gid              = nlist->gid;
 577     shiftvec         = fr->shift_vec[0];
 578     fshift           = fr->fshift[0];
 579     facel            = _mm_set1_ps(fr->epsfac);
 580     charge           = mdatoms->chargeA;
 581     nvdwtype         = fr->ntype;
 582     vdwparam         = fr->nbfp;
 583     vdwtype          = mdatoms->typeA;
 584
 585     /* Setup water-specific parameters */
 586     inr              = nlist->iinr[0];
 587     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 588     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 589     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 590     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 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_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 623                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 624
 625         fix0             = _mm_setzero_ps();
 626         fiy0             = _mm_setzero_ps();
 627         fiz0             = _mm_setzero_ps();
 628         fix1             = _mm_setzero_ps();
 629         fiy1             = _mm_setzero_ps();
 630         fiz1             = _mm_setzero_ps();
 631         fix2             = _mm_setzero_ps();
 632         fiy2             = _mm_setzero_ps();
 633         fiz2             = _mm_setzero_ps();
 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_mm_load_1rvec_4ptr_swizzle_ps(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             dx00             = _mm_sub_ps(ix0,jx0);
 656             dy00             = _mm_sub_ps(iy0,jy0);
 657             dz00             = _mm_sub_ps(iz0,jz0);
 658             dx10             = _mm_sub_ps(ix1,jx0);
 659             dy10             = _mm_sub_ps(iy1,jy0);
 660             dz10             = _mm_sub_ps(iz1,jz0);
 661             dx20             = _mm_sub_ps(ix2,jx0);
 662             dy20             = _mm_sub_ps(iy2,jy0);
 663             dz20             = _mm_sub_ps(iz2,jz0);
 664
 665             /* Calculate squared distance and things based on it */
 666             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 667             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 668             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 669
 670             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 671             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 672             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 673
 674             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 675             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 676             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 677
 678             /* Load parameters for j particles */
 679             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 680                                                               charge+jnrC+0,charge+jnrD+0);
 681             vdwjidx0A        = 2*vdwtype[jnrA+0];
 682             vdwjidx0B        = 2*vdwtype[jnrB+0];
 683             vdwjidx0C        = 2*vdwtype[jnrC+0];
 684             vdwjidx0D        = 2*vdwtype[jnrD+0];
 685
 686             fjx0             = _mm_setzero_ps();
 687             fjy0             = _mm_setzero_ps();
 688             fjz0             = _mm_setzero_ps();
 689
 690             /**************************
 691              * CALCULATE INTERACTIONS *
 692              **************************/
 693
 694             /* Compute parameters for interactions between i and j atoms */
 695             qq00             = _mm_mul_ps(iq0,jq0);
 696             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 697                                          vdwparam+vdwioffset0+vdwjidx0B,
 698                                          vdwparam+vdwioffset0+vdwjidx0C,
 699                                          vdwparam+vdwioffset0+vdwjidx0D,
 700                                          &c6_00,&c12_00);
 701
 702             /* COULOMB ELECTROSTATICS */
 703             velec            = _mm_mul_ps(qq00,rinv00);
 704             felec            = _mm_mul_ps(velec,rinvsq00);
 705
 706             /* LENNARD-JONES DISPERSION/REPULSION */
 707
 708             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 709             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 710
 711             fscal            = _mm_add_ps(felec,fvdw);
 712
 713              /* Update vectorial force */
 714             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 715             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 716             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 717
 718             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 719             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 720             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 721
 722             /**************************
 723              * CALCULATE INTERACTIONS *
 724              **************************/
 725
 726             /* Compute parameters for interactions between i and j atoms */
 727             qq10             = _mm_mul_ps(iq1,jq0);
 728
 729             /* COULOMB ELECTROSTATICS */
 730             velec            = _mm_mul_ps(qq10,rinv10);
 731             felec            = _mm_mul_ps(velec,rinvsq10);
 732
 733             fscal            = felec;
 734
 735              /* Update vectorial force */
 736             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 737             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 738             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 739
 740             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 741             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 742             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 743
 744             /**************************
 745              * CALCULATE INTERACTIONS *
 746              **************************/
 747
 748             /* Compute parameters for interactions between i and j atoms */
 749             qq20             = _mm_mul_ps(iq2,jq0);
 750
 751             /* COULOMB ELECTROSTATICS */
 752             velec            = _mm_mul_ps(qq20,rinv20);
 753             felec            = _mm_mul_ps(velec,rinvsq20);
 754
 755             fscal            = felec;
 756
 757              /* Update vectorial force */
 758             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 759             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 760             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 761
 762             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 763             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 764             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 765
 766             fjptrA             = f+j_coord_offsetA;
 767             fjptrB             = f+j_coord_offsetB;
 768             fjptrC             = f+j_coord_offsetC;
 769             fjptrD             = f+j_coord_offsetD;
 770
 771             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 772
 773             /* Inner loop uses 97 flops */
 774         }
 775
 776         if(jidx<j_index_end)
 777         {
 778
 779             /* Get j neighbor index, and coordinate index */
 780             jnrlistA         = jjnr[jidx];
 781             jnrlistB         = jjnr[jidx+1];
 782             jnrlistC         = jjnr[jidx+2];
 783             jnrlistD         = jjnr[jidx+3];
 784             /* Sign of each element will be negative for non-real atoms.
 785              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 786              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 787              */
 788             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 789             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 790             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 791             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 792             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 793             j_coord_offsetA  = DIM*jnrA;
 794             j_coord_offsetB  = DIM*jnrB;
 795             j_coord_offsetC  = DIM*jnrC;
 796             j_coord_offsetD  = DIM*jnrD;
 797
 798             /* load j atom coordinates */
 799             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 800                                               x+j_coord_offsetC,x+j_coord_offsetD,
 801                                               &jx0,&jy0,&jz0);
 802
 803             /* Calculate displacement vector */
 804             dx00             = _mm_sub_ps(ix0,jx0);
 805             dy00             = _mm_sub_ps(iy0,jy0);
 806             dz00             = _mm_sub_ps(iz0,jz0);
 807             dx10             = _mm_sub_ps(ix1,jx0);
 808             dy10             = _mm_sub_ps(iy1,jy0);
 809             dz10             = _mm_sub_ps(iz1,jz0);
 810             dx20             = _mm_sub_ps(ix2,jx0);
 811             dy20             = _mm_sub_ps(iy2,jy0);
 812             dz20             = _mm_sub_ps(iz2,jz0);
 813
 814             /* Calculate squared distance and things based on it */
 815             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 816             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 817             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 818
 819             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 820             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 821             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 822
 823             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 824             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 825             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 826
 827             /* Load parameters for j particles */
 828             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 829                                                               charge+jnrC+0,charge+jnrD+0);
 830             vdwjidx0A        = 2*vdwtype[jnrA+0];
 831             vdwjidx0B        = 2*vdwtype[jnrB+0];
 832             vdwjidx0C        = 2*vdwtype[jnrC+0];
 833             vdwjidx0D        = 2*vdwtype[jnrD+0];
 834
 835             fjx0             = _mm_setzero_ps();
 836             fjy0             = _mm_setzero_ps();
 837             fjz0             = _mm_setzero_ps();
 838
 839             /**************************
 840              * CALCULATE INTERACTIONS *
 841              **************************/
 842
 843             /* Compute parameters for interactions between i and j atoms */
 844             qq00             = _mm_mul_ps(iq0,jq0);
 845             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 846                                          vdwparam+vdwioffset0+vdwjidx0B,
 847                                          vdwparam+vdwioffset0+vdwjidx0C,
 848                                          vdwparam+vdwioffset0+vdwjidx0D,
 849                                          &c6_00,&c12_00);
 850
 851             /* COULOMB ELECTROSTATICS */
 852             velec            = _mm_mul_ps(qq00,rinv00);
 853             felec            = _mm_mul_ps(velec,rinvsq00);
 854
 855             /* LENNARD-JONES DISPERSION/REPULSION */
 856
 857             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 858             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 859
 860             fscal            = _mm_add_ps(felec,fvdw);
 861
 862             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 863
 864              /* Update vectorial force */
 865             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 866             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 867             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 868
 869             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 870             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 871             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 872
 873             /**************************
 874              * CALCULATE INTERACTIONS *
 875              **************************/
 876
 877             /* Compute parameters for interactions between i and j atoms */
 878             qq10             = _mm_mul_ps(iq1,jq0);
 879
 880             /* COULOMB ELECTROSTATICS */
 881             velec            = _mm_mul_ps(qq10,rinv10);
 882             felec            = _mm_mul_ps(velec,rinvsq10);
 883
 884             fscal            = felec;
 885
 886             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 887
 888              /* Update vectorial force */
 889             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 890             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 891             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 892
 893             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 894             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 895             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 896
 897             /**************************
 898              * CALCULATE INTERACTIONS *
 899              **************************/
 900
 901             /* Compute parameters for interactions between i and j atoms */
 902             qq20             = _mm_mul_ps(iq2,jq0);
 903
 904             /* COULOMB ELECTROSTATICS */
 905             velec            = _mm_mul_ps(qq20,rinv20);
 906             felec            = _mm_mul_ps(velec,rinvsq20);
 907
 908             fscal            = felec;
 909
 910             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 911
 912              /* Update vectorial force */
 913             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 914             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 915             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 916
 917             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 918             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 919             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 920
 921             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 922             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 923             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 924             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 925
 926             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 927
 928             /* Inner loop uses 97 flops */
 929         }
 930
 931         /* End of innermost loop */
 932
 933         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 934                                               f+i_coord_offset,fshift+i_shift_offset);
 935
 936         /* Increment number of inner iterations */
 937         inneriter                  += j_index_end - j_index_start;
 938
 939         /* Outer loop uses 18 flops */
 940     }
 941
 942     /* Increment number of outer iterations */
 943     outeriter        += nri;
 944
 945     /* Update outer/inner flops */
 946
 947     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*97);
 948 }