src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_avx_256_double.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_256_double 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_256_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_VF_avx_256_double
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_VF_avx_256_double
  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, 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              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     real *           vdwioffsetptr0;
  84     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  86     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  89     real             *charge;
  90     int              nvdwtype;
  91     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  92     int              *vdwtype;
  93     real             *vdwparam;
  94     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  95     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  96     __m128i          vfitab;
  97     __m128i          ifour       = _mm_set1_epi32(4);
  98     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  99     real             *vftab;
 100     __m256d          dummy_mask,cutoff_mask;
 101     __m128           tmpmask0,tmpmask1;
 102     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 103     __m256d          one     = _mm256_set1_pd(1.0);
 104     __m256d          two     = _mm256_set1_pd(2.0);
 105     x                = xx[0];
 106     f                = ff[0];
 107
 108     nri              = nlist->nri;
 109     iinr             = nlist->iinr;
 110     jindex           = nlist->jindex;
 111     jjnr             = nlist->jjnr;
 112     shiftidx         = nlist->shift;
 113     gid              = nlist->gid;
 114     shiftvec         = fr->shift_vec[0];
 115     fshift           = fr->fshift[0];
 116     facel            = _mm256_set1_pd(fr->ic->epsfac);
 117     charge           = mdatoms->chargeA;
 118     krf              = _mm256_set1_pd(fr->ic->k_rf);
 119     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 120     crf              = _mm256_set1_pd(fr->ic->c_rf);
 121     nvdwtype         = fr->ntype;
 122     vdwparam         = fr->nbfp;
 123     vdwtype          = mdatoms->typeA;
 124
 125     vftab            = kernel_data->table_vdw->data;
 126     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 127
 128     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 129     rcutoff_scalar   = fr->ic->rcoulomb;
 130     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 131     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 132
 133     /* Avoid stupid compiler warnings */
 134     jnrA = jnrB = jnrC = jnrD = 0;
 135     j_coord_offsetA = 0;
 136     j_coord_offsetB = 0;
 137     j_coord_offsetC = 0;
 138     j_coord_offsetD = 0;
 139
 140     outeriter        = 0;
 141     inneriter        = 0;
 142
 143     for(iidx=0;iidx<4*DIM;iidx++)
 144     {
 145         scratch[iidx] = 0.0;
 146     }
 147
 148     /* Start outer loop over neighborlists */
 149     for(iidx=0; iidx<nri; iidx++)
 150     {
 151         /* Load shift vector for this list */
 152         i_shift_offset   = DIM*shiftidx[iidx];
 153
 154         /* Load limits for loop over neighbors */
 155         j_index_start    = jindex[iidx];
 156         j_index_end      = jindex[iidx+1];
 157
 158         /* Get outer coordinate index */
 159         inr              = iinr[iidx];
 160         i_coord_offset   = DIM*inr;
 161
 162         /* Load i particle coords and add shift vector */
 163         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 164
 165         fix0             = _mm256_setzero_pd();
 166         fiy0             = _mm256_setzero_pd();
 167         fiz0             = _mm256_setzero_pd();
 168
 169         /* Load parameters for i particles */
 170         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 171         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 172
 173         /* Reset potential sums */
 174         velecsum         = _mm256_setzero_pd();
 175         vvdwsum          = _mm256_setzero_pd();
 176
 177         /* Start inner kernel loop */
 178         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 179         {
 180
 181             /* Get j neighbor index, and coordinate index */
 182             jnrA             = jjnr[jidx];
 183             jnrB             = jjnr[jidx+1];
 184             jnrC             = jjnr[jidx+2];
 185             jnrD             = jjnr[jidx+3];
 186             j_coord_offsetA  = DIM*jnrA;
 187             j_coord_offsetB  = DIM*jnrB;
 188             j_coord_offsetC  = DIM*jnrC;
 189             j_coord_offsetD  = DIM*jnrD;
 190
 191             /* load j atom coordinates */
 192             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 193                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 194                                                  &jx0,&jy0,&jz0);
 195
 196             /* Calculate displacement vector */
 197             dx00             = _mm256_sub_pd(ix0,jx0);
 198             dy00             = _mm256_sub_pd(iy0,jy0);
 199             dz00             = _mm256_sub_pd(iz0,jz0);
 200
 201             /* Calculate squared distance and things based on it */
 202             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 203
 204             rinv00           = avx256_invsqrt_d(rsq00);
 205
 206             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 207
 208             /* Load parameters for j particles */
 209             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 210                                                                  charge+jnrC+0,charge+jnrD+0);
 211             vdwjidx0A        = 2*vdwtype[jnrA+0];
 212             vdwjidx0B        = 2*vdwtype[jnrB+0];
 213             vdwjidx0C        = 2*vdwtype[jnrC+0];
 214             vdwjidx0D        = 2*vdwtype[jnrD+0];
 215
 216             /**************************
 217              * CALCULATE INTERACTIONS *
 218              **************************/
 219
 220             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 221             {
 222
 223             r00              = _mm256_mul_pd(rsq00,rinv00);
 224
 225             /* Compute parameters for interactions between i and j atoms */
 226             qq00             = _mm256_mul_pd(iq0,jq0);
 227             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 228                                             vdwioffsetptr0+vdwjidx0B,
 229                                             vdwioffsetptr0+vdwjidx0C,
 230                                             vdwioffsetptr0+vdwjidx0D,
 231                                             &c6_00,&c12_00);
 232
 233             /* Calculate table index by multiplying r with table scale and truncate to integer */
 234             rt               = _mm256_mul_pd(r00,vftabscale);
 235             vfitab           = _mm256_cvttpd_epi32(rt);
 236             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 237             vfitab           = _mm_slli_epi32(vfitab,3);
 238
 239             /* REACTION-FIELD ELECTROSTATICS */
 240             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 241             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 242
 243             /* CUBIC SPLINE TABLE DISPERSION */
 244             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 245             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 246             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 247             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 248             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 249             Heps             = _mm256_mul_pd(vfeps,H);
 250             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 251             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 252             vvdw6            = _mm256_mul_pd(c6_00,VV);
 253             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 254             fvdw6            = _mm256_mul_pd(c6_00,FF);
 255
 256             /* CUBIC SPLINE TABLE REPULSION */
 257             vfitab           = _mm_add_epi32(vfitab,ifour);
 258             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 259             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 260             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 261             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 262             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 263             Heps             = _mm256_mul_pd(vfeps,H);
 264             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 265             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 266             vvdw12           = _mm256_mul_pd(c12_00,VV);
 267             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 268             fvdw12           = _mm256_mul_pd(c12_00,FF);
 269             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 270             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 271
 272             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 273
 274             /* Update potential sum for this i atom from the interaction with this j atom. */
 275             velec            = _mm256_and_pd(velec,cutoff_mask);
 276             velecsum         = _mm256_add_pd(velecsum,velec);
 277             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 278             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 279
 280             fscal            = _mm256_add_pd(felec,fvdw);
 281
 282             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 283
 284             /* Calculate temporary vectorial force */
 285             tx               = _mm256_mul_pd(fscal,dx00);
 286             ty               = _mm256_mul_pd(fscal,dy00);
 287             tz               = _mm256_mul_pd(fscal,dz00);
 288
 289             /* Update vectorial force */
 290             fix0             = _mm256_add_pd(fix0,tx);
 291             fiy0             = _mm256_add_pd(fiy0,ty);
 292             fiz0             = _mm256_add_pd(fiz0,tz);
 293
 294             fjptrA             = f+j_coord_offsetA;
 295             fjptrB             = f+j_coord_offsetB;
 296             fjptrC             = f+j_coord_offsetC;
 297             fjptrD             = f+j_coord_offsetD;
 298             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 299
 300             }
 301
 302             /* Inner loop uses 72 flops */
 303         }
 304
 305         if(jidx<j_index_end)
 306         {
 307
 308             /* Get j neighbor index, and coordinate index */
 309             jnrlistA         = jjnr[jidx];
 310             jnrlistB         = jjnr[jidx+1];
 311             jnrlistC         = jjnr[jidx+2];
 312             jnrlistD         = jjnr[jidx+3];
 313             /* Sign of each element will be negative for non-real atoms.
 314              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 315              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 316              */
 317             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 318
 319             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 320             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 321             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 322
 323             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 324             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 325             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 326             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 327             j_coord_offsetA  = DIM*jnrA;
 328             j_coord_offsetB  = DIM*jnrB;
 329             j_coord_offsetC  = DIM*jnrC;
 330             j_coord_offsetD  = DIM*jnrD;
 331
 332             /* load j atom coordinates */
 333             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 334                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 335                                                  &jx0,&jy0,&jz0);
 336
 337             /* Calculate displacement vector */
 338             dx00             = _mm256_sub_pd(ix0,jx0);
 339             dy00             = _mm256_sub_pd(iy0,jy0);
 340             dz00             = _mm256_sub_pd(iz0,jz0);
 341
 342             /* Calculate squared distance and things based on it */
 343             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 344
 345             rinv00           = avx256_invsqrt_d(rsq00);
 346
 347             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 348
 349             /* Load parameters for j particles */
 350             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 351                                                                  charge+jnrC+0,charge+jnrD+0);
 352             vdwjidx0A        = 2*vdwtype[jnrA+0];
 353             vdwjidx0B        = 2*vdwtype[jnrB+0];
 354             vdwjidx0C        = 2*vdwtype[jnrC+0];
 355             vdwjidx0D        = 2*vdwtype[jnrD+0];
 356
 357             /**************************
 358              * CALCULATE INTERACTIONS *
 359              **************************/
 360
 361             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 362             {
 363
 364             r00              = _mm256_mul_pd(rsq00,rinv00);
 365             r00              = _mm256_andnot_pd(dummy_mask,r00);
 366
 367             /* Compute parameters for interactions between i and j atoms */
 368             qq00             = _mm256_mul_pd(iq0,jq0);
 369             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 370                                             vdwioffsetptr0+vdwjidx0B,
 371                                             vdwioffsetptr0+vdwjidx0C,
 372                                             vdwioffsetptr0+vdwjidx0D,
 373                                             &c6_00,&c12_00);
 374
 375             /* Calculate table index by multiplying r with table scale and truncate to integer */
 376             rt               = _mm256_mul_pd(r00,vftabscale);
 377             vfitab           = _mm256_cvttpd_epi32(rt);
 378             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 379             vfitab           = _mm_slli_epi32(vfitab,3);
 380
 381             /* REACTION-FIELD ELECTROSTATICS */
 382             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 383             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 384
 385             /* CUBIC SPLINE TABLE DISPERSION */
 386             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 387             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 388             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 389             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 390             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 391             Heps             = _mm256_mul_pd(vfeps,H);
 392             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 393             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 394             vvdw6            = _mm256_mul_pd(c6_00,VV);
 395             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 396             fvdw6            = _mm256_mul_pd(c6_00,FF);
 397
 398             /* CUBIC SPLINE TABLE REPULSION */
 399             vfitab           = _mm_add_epi32(vfitab,ifour);
 400             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 401             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 402             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 403             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 404             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 405             Heps             = _mm256_mul_pd(vfeps,H);
 406             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 407             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 408             vvdw12           = _mm256_mul_pd(c12_00,VV);
 409             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 410             fvdw12           = _mm256_mul_pd(c12_00,FF);
 411             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 412             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 413
 414             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 415
 416             /* Update potential sum for this i atom from the interaction with this j atom. */
 417             velec            = _mm256_and_pd(velec,cutoff_mask);
 418             velec            = _mm256_andnot_pd(dummy_mask,velec);
 419             velecsum         = _mm256_add_pd(velecsum,velec);
 420             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 421             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 422             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 423
 424             fscal            = _mm256_add_pd(felec,fvdw);
 425
 426             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 427
 428             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 429
 430             /* Calculate temporary vectorial force */
 431             tx               = _mm256_mul_pd(fscal,dx00);
 432             ty               = _mm256_mul_pd(fscal,dy00);
 433             tz               = _mm256_mul_pd(fscal,dz00);
 434
 435             /* Update vectorial force */
 436             fix0             = _mm256_add_pd(fix0,tx);
 437             fiy0             = _mm256_add_pd(fiy0,ty);
 438             fiz0             = _mm256_add_pd(fiz0,tz);
 439
 440             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 441             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 442             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 443             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 444             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 445
 446             }
 447
 448             /* Inner loop uses 73 flops */
 449         }
 450
 451         /* End of innermost loop */
 452
 453         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 454                                                  f+i_coord_offset,fshift+i_shift_offset);
 455
 456         ggid                        = gid[iidx];
 457         /* Update potential energies */
 458         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 459         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 460
 461         /* Increment number of inner iterations */
 462         inneriter                  += j_index_end - j_index_start;
 463
 464         /* Outer loop uses 9 flops */
 465     }
 466
 467     /* Increment number of outer iterations */
 468     outeriter        += nri;
 469
 470     /* Update outer/inner flops */
 471
 472     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*73);
 473 }
 474 /*
 475  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_F_avx_256_double
 476  * Electrostatics interaction: ReactionField
 477  * VdW interaction:            CubicSplineTable
 478  * Geometry:                   Particle-Particle
 479  * Calculate force/pot:        Force
 480  */
 481 void
 482 nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_F_avx_256_double
 483                     (t_nblist                    * gmx_restrict       nlist,
 484                      rvec                        * gmx_restrict          xx,
 485                      rvec                        * gmx_restrict          ff,
 486                      struct t_forcerec           * gmx_restrict          fr,
 487                      t_mdatoms                   * gmx_restrict     mdatoms,
 488                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 489                      t_nrnb                      * gmx_restrict        nrnb)
 490 {
 491     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 492      * just 0 for non-waters.
 493      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 494      * jnr indices corresponding to data put in the four positions in the SIMD register.
 495      */
 496     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 497     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 498     int              jnrA,jnrB,jnrC,jnrD;
 499     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 500     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 501     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 502     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 503     real             rcutoff_scalar;
 504     real             *shiftvec,*fshift,*x,*f;
 505     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 506     real             scratch[4*DIM];
 507     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 508     real *           vdwioffsetptr0;
 509     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 510     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 511     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 512     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 513     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 514     real             *charge;
 515     int              nvdwtype;
 516     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 517     int              *vdwtype;
 518     real             *vdwparam;
 519     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 520     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 521     __m128i          vfitab;
 522     __m128i          ifour       = _mm_set1_epi32(4);
 523     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 524     real             *vftab;
 525     __m256d          dummy_mask,cutoff_mask;
 526     __m128           tmpmask0,tmpmask1;
 527     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 528     __m256d          one     = _mm256_set1_pd(1.0);
 529     __m256d          two     = _mm256_set1_pd(2.0);
 530     x                = xx[0];
 531     f                = ff[0];
 532
 533     nri              = nlist->nri;
 534     iinr             = nlist->iinr;
 535     jindex           = nlist->jindex;
 536     jjnr             = nlist->jjnr;
 537     shiftidx         = nlist->shift;
 538     gid              = nlist->gid;
 539     shiftvec         = fr->shift_vec[0];
 540     fshift           = fr->fshift[0];
 541     facel            = _mm256_set1_pd(fr->ic->epsfac);
 542     charge           = mdatoms->chargeA;
 543     krf              = _mm256_set1_pd(fr->ic->k_rf);
 544     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 545     crf              = _mm256_set1_pd(fr->ic->c_rf);
 546     nvdwtype         = fr->ntype;
 547     vdwparam         = fr->nbfp;
 548     vdwtype          = mdatoms->typeA;
 549
 550     vftab            = kernel_data->table_vdw->data;
 551     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 552
 553     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 554     rcutoff_scalar   = fr->ic->rcoulomb;
 555     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 556     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 557
 558     /* Avoid stupid compiler warnings */
 559     jnrA = jnrB = jnrC = jnrD = 0;
 560     j_coord_offsetA = 0;
 561     j_coord_offsetB = 0;
 562     j_coord_offsetC = 0;
 563     j_coord_offsetD = 0;
 564
 565     outeriter        = 0;
 566     inneriter        = 0;
 567
 568     for(iidx=0;iidx<4*DIM;iidx++)
 569     {
 570         scratch[iidx] = 0.0;
 571     }
 572
 573     /* Start outer loop over neighborlists */
 574     for(iidx=0; iidx<nri; iidx++)
 575     {
 576         /* Load shift vector for this list */
 577         i_shift_offset   = DIM*shiftidx[iidx];
 578
 579         /* Load limits for loop over neighbors */
 580         j_index_start    = jindex[iidx];
 581         j_index_end      = jindex[iidx+1];
 582
 583         /* Get outer coordinate index */
 584         inr              = iinr[iidx];
 585         i_coord_offset   = DIM*inr;
 586
 587         /* Load i particle coords and add shift vector */
 588         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 589
 590         fix0             = _mm256_setzero_pd();
 591         fiy0             = _mm256_setzero_pd();
 592         fiz0             = _mm256_setzero_pd();
 593
 594         /* Load parameters for i particles */
 595         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 596         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 597
 598         /* Start inner kernel loop */
 599         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 600         {
 601
 602             /* Get j neighbor index, and coordinate index */
 603             jnrA             = jjnr[jidx];
 604             jnrB             = jjnr[jidx+1];
 605             jnrC             = jjnr[jidx+2];
 606             jnrD             = jjnr[jidx+3];
 607             j_coord_offsetA  = DIM*jnrA;
 608             j_coord_offsetB  = DIM*jnrB;
 609             j_coord_offsetC  = DIM*jnrC;
 610             j_coord_offsetD  = DIM*jnrD;
 611
 612             /* load j atom coordinates */
 613             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 614                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 615                                                  &jx0,&jy0,&jz0);
 616
 617             /* Calculate displacement vector */
 618             dx00             = _mm256_sub_pd(ix0,jx0);
 619             dy00             = _mm256_sub_pd(iy0,jy0);
 620             dz00             = _mm256_sub_pd(iz0,jz0);
 621
 622             /* Calculate squared distance and things based on it */
 623             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 624
 625             rinv00           = avx256_invsqrt_d(rsq00);
 626
 627             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 628
 629             /* Load parameters for j particles */
 630             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 631                                                                  charge+jnrC+0,charge+jnrD+0);
 632             vdwjidx0A        = 2*vdwtype[jnrA+0];
 633             vdwjidx0B        = 2*vdwtype[jnrB+0];
 634             vdwjidx0C        = 2*vdwtype[jnrC+0];
 635             vdwjidx0D        = 2*vdwtype[jnrD+0];
 636
 637             /**************************
 638              * CALCULATE INTERACTIONS *
 639              **************************/
 640
 641             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 642             {
 643
 644             r00              = _mm256_mul_pd(rsq00,rinv00);
 645
 646             /* Compute parameters for interactions between i and j atoms */
 647             qq00             = _mm256_mul_pd(iq0,jq0);
 648             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 649                                             vdwioffsetptr0+vdwjidx0B,
 650                                             vdwioffsetptr0+vdwjidx0C,
 651                                             vdwioffsetptr0+vdwjidx0D,
 652                                             &c6_00,&c12_00);
 653
 654             /* Calculate table index by multiplying r with table scale and truncate to integer */
 655             rt               = _mm256_mul_pd(r00,vftabscale);
 656             vfitab           = _mm256_cvttpd_epi32(rt);
 657             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 658             vfitab           = _mm_slli_epi32(vfitab,3);
 659
 660             /* REACTION-FIELD ELECTROSTATICS */
 661             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 662
 663             /* CUBIC SPLINE TABLE DISPERSION */
 664             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 665             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 666             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 667             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 668             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 669             Heps             = _mm256_mul_pd(vfeps,H);
 670             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 671             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 672             fvdw6            = _mm256_mul_pd(c6_00,FF);
 673
 674             /* CUBIC SPLINE TABLE REPULSION */
 675             vfitab           = _mm_add_epi32(vfitab,ifour);
 676             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 677             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 678             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 679             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 680             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 681             Heps             = _mm256_mul_pd(vfeps,H);
 682             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 683             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 684             fvdw12           = _mm256_mul_pd(c12_00,FF);
 685             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 686
 687             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 688
 689             fscal            = _mm256_add_pd(felec,fvdw);
 690
 691             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 692
 693             /* Calculate temporary vectorial force */
 694             tx               = _mm256_mul_pd(fscal,dx00);
 695             ty               = _mm256_mul_pd(fscal,dy00);
 696             tz               = _mm256_mul_pd(fscal,dz00);
 697
 698             /* Update vectorial force */
 699             fix0             = _mm256_add_pd(fix0,tx);
 700             fiy0             = _mm256_add_pd(fiy0,ty);
 701             fiz0             = _mm256_add_pd(fiz0,tz);
 702
 703             fjptrA             = f+j_coord_offsetA;
 704             fjptrB             = f+j_coord_offsetB;
 705             fjptrC             = f+j_coord_offsetC;
 706             fjptrD             = f+j_coord_offsetD;
 707             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 708
 709             }
 710
 711             /* Inner loop uses 57 flops */
 712         }
 713
 714         if(jidx<j_index_end)
 715         {
 716
 717             /* Get j neighbor index, and coordinate index */
 718             jnrlistA         = jjnr[jidx];
 719             jnrlistB         = jjnr[jidx+1];
 720             jnrlistC         = jjnr[jidx+2];
 721             jnrlistD         = jjnr[jidx+3];
 722             /* Sign of each element will be negative for non-real atoms.
 723              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 724              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 725              */
 726             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 727
 728             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 729             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 730             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 731
 732             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 733             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 734             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 735             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 736             j_coord_offsetA  = DIM*jnrA;
 737             j_coord_offsetB  = DIM*jnrB;
 738             j_coord_offsetC  = DIM*jnrC;
 739             j_coord_offsetD  = DIM*jnrD;
 740
 741             /* load j atom coordinates */
 742             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 743                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 744                                                  &jx0,&jy0,&jz0);
 745
 746             /* Calculate displacement vector */
 747             dx00             = _mm256_sub_pd(ix0,jx0);
 748             dy00             = _mm256_sub_pd(iy0,jy0);
 749             dz00             = _mm256_sub_pd(iz0,jz0);
 750
 751             /* Calculate squared distance and things based on it */
 752             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 753
 754             rinv00           = avx256_invsqrt_d(rsq00);
 755
 756             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 757
 758             /* Load parameters for j particles */
 759             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 760                                                                  charge+jnrC+0,charge+jnrD+0);
 761             vdwjidx0A        = 2*vdwtype[jnrA+0];
 762             vdwjidx0B        = 2*vdwtype[jnrB+0];
 763             vdwjidx0C        = 2*vdwtype[jnrC+0];
 764             vdwjidx0D        = 2*vdwtype[jnrD+0];
 765
 766             /**************************
 767              * CALCULATE INTERACTIONS *
 768              **************************/
 769
 770             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 771             {
 772
 773             r00              = _mm256_mul_pd(rsq00,rinv00);
 774             r00              = _mm256_andnot_pd(dummy_mask,r00);
 775
 776             /* Compute parameters for interactions between i and j atoms */
 777             qq00             = _mm256_mul_pd(iq0,jq0);
 778             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 779                                             vdwioffsetptr0+vdwjidx0B,
 780                                             vdwioffsetptr0+vdwjidx0C,
 781                                             vdwioffsetptr0+vdwjidx0D,
 782                                             &c6_00,&c12_00);
 783
 784             /* Calculate table index by multiplying r with table scale and truncate to integer */
 785             rt               = _mm256_mul_pd(r00,vftabscale);
 786             vfitab           = _mm256_cvttpd_epi32(rt);
 787             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 788             vfitab           = _mm_slli_epi32(vfitab,3);
 789
 790             /* REACTION-FIELD ELECTROSTATICS */
 791             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 792
 793             /* CUBIC SPLINE TABLE DISPERSION */
 794             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 795             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 796             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 797             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 798             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 799             Heps             = _mm256_mul_pd(vfeps,H);
 800             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 801             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 802             fvdw6            = _mm256_mul_pd(c6_00,FF);
 803
 804             /* CUBIC SPLINE TABLE REPULSION */
 805             vfitab           = _mm_add_epi32(vfitab,ifour);
 806             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 807             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 808             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 809             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 810             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 811             Heps             = _mm256_mul_pd(vfeps,H);
 812             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 813             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 814             fvdw12           = _mm256_mul_pd(c12_00,FF);
 815             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 816
 817             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 818
 819             fscal            = _mm256_add_pd(felec,fvdw);
 820
 821             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 822
 823             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 824
 825             /* Calculate temporary vectorial force */
 826             tx               = _mm256_mul_pd(fscal,dx00);
 827             ty               = _mm256_mul_pd(fscal,dy00);
 828             tz               = _mm256_mul_pd(fscal,dz00);
 829
 830             /* Update vectorial force */
 831             fix0             = _mm256_add_pd(fix0,tx);
 832             fiy0             = _mm256_add_pd(fiy0,ty);
 833             fiz0             = _mm256_add_pd(fiz0,tz);
 834
 835             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 836             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 837             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 838             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 839             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 840
 841             }
 842
 843             /* Inner loop uses 58 flops */
 844         }
 845
 846         /* End of innermost loop */
 847
 848         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 849                                                  f+i_coord_offset,fshift+i_shift_offset);
 850
 851         /* Increment number of inner iterations */
 852         inneriter                  += j_index_end - j_index_start;
 853
 854         /* Outer loop uses 7 flops */
 855     }
 856
 857     /* Increment number of outer iterations */
 858     outeriter        += nri;
 859
 860     /* Update outer/inner flops */
 861
 862     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*58);
 863 }