src/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_avx_256_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_double 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_256_double.h"
  34 #include "kernelutil_x86_avx_256_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_VF_avx_256_double
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_VF_avx_256_double
  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, 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              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  63     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  64     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  65     real             rcutoff_scalar;
  66     real             *shiftvec,*fshift,*x,*f;
  67     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  68     real             scratch[4*DIM];
  69     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  70     real *           vdwioffsetptr0;
  71     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  72     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  73     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  75     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  76     real             *charge;
  77     int              nvdwtype;
  78     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  79     int              *vdwtype;
  80     real             *vdwparam;
  81     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  82     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  83     __m128i          vfitab;
  84     __m128i          ifour       = _mm_set1_epi32(4);
  85     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  86     real             *vftab;
  87     __m256d          dummy_mask,cutoff_mask;
  88     __m128           tmpmask0,tmpmask1;
  89     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
  90     __m256d          one     = _mm256_set1_pd(1.0);
  91     __m256d          two     = _mm256_set1_pd(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            = _mm256_set1_pd(fr->epsfac);
 104     charge           = mdatoms->chargeA;
 105     nvdwtype         = fr->ntype;
 106     vdwparam         = fr->nbfp;
 107     vdwtype          = mdatoms->typeA;
 108
 109     vftab            = kernel_data->table_elec_vdw->data;
 110     vftabscale       = _mm256_set1_pd(kernel_data->table_elec_vdw->scale);
 111
 112     /* Avoid stupid compiler warnings */
 113     jnrA = jnrB = jnrC = jnrD = 0;
 114     j_coord_offsetA = 0;
 115     j_coord_offsetB = 0;
 116     j_coord_offsetC = 0;
 117     j_coord_offsetD = 0;
 118
 119     outeriter        = 0;
 120     inneriter        = 0;
 121
 122     for(iidx=0;iidx<4*DIM;iidx++)
 123     {
 124         scratch[iidx] = 0.0;
 125     }
 126
 127     /* Start outer loop over neighborlists */
 128     for(iidx=0; iidx<nri; iidx++)
 129     {
 130         /* Load shift vector for this list */
 131         i_shift_offset   = DIM*shiftidx[iidx];
 132
 133         /* Load limits for loop over neighbors */
 134         j_index_start    = jindex[iidx];
 135         j_index_end      = jindex[iidx+1];
 136
 137         /* Get outer coordinate index */
 138         inr              = iinr[iidx];
 139         i_coord_offset   = DIM*inr;
 140
 141         /* Load i particle coords and add shift vector */
 142         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 143
 144         fix0             = _mm256_setzero_pd();
 145         fiy0             = _mm256_setzero_pd();
 146         fiz0             = _mm256_setzero_pd();
 147
 148         /* Load parameters for i particles */
 149         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 150         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 151
 152         /* Reset potential sums */
 153         velecsum         = _mm256_setzero_pd();
 154         vvdwsum          = _mm256_setzero_pd();
 155
 156         /* Start inner kernel loop */
 157         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 158         {
 159
 160             /* Get j neighbor index, and coordinate index */
 161             jnrA             = jjnr[jidx];
 162             jnrB             = jjnr[jidx+1];
 163             jnrC             = jjnr[jidx+2];
 164             jnrD             = jjnr[jidx+3];
 165             j_coord_offsetA  = DIM*jnrA;
 166             j_coord_offsetB  = DIM*jnrB;
 167             j_coord_offsetC  = DIM*jnrC;
 168             j_coord_offsetD  = DIM*jnrD;
 169
 170             /* load j atom coordinates */
 171             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 172                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 173                                                  &jx0,&jy0,&jz0);
 174
 175             /* Calculate displacement vector */
 176             dx00             = _mm256_sub_pd(ix0,jx0);
 177             dy00             = _mm256_sub_pd(iy0,jy0);
 178             dz00             = _mm256_sub_pd(iz0,jz0);
 179
 180             /* Calculate squared distance and things based on it */
 181             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 182
 183             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 184
 185             /* Load parameters for j particles */
 186             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 187                                                                  charge+jnrC+0,charge+jnrD+0);
 188             vdwjidx0A        = 2*vdwtype[jnrA+0];
 189             vdwjidx0B        = 2*vdwtype[jnrB+0];
 190             vdwjidx0C        = 2*vdwtype[jnrC+0];
 191             vdwjidx0D        = 2*vdwtype[jnrD+0];
 192
 193             /**************************
 194              * CALCULATE INTERACTIONS *
 195              **************************/
 196
 197             r00              = _mm256_mul_pd(rsq00,rinv00);
 198
 199             /* Compute parameters for interactions between i and j atoms */
 200             qq00             = _mm256_mul_pd(iq0,jq0);
 201             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 202                                             vdwioffsetptr0+vdwjidx0B,
 203                                             vdwioffsetptr0+vdwjidx0C,
 204                                             vdwioffsetptr0+vdwjidx0D,
 205                                             &c6_00,&c12_00);
 206
 207             /* Calculate table index by multiplying r with table scale and truncate to integer */
 208             rt               = _mm256_mul_pd(r00,vftabscale);
 209             vfitab           = _mm256_cvttpd_epi32(rt);
 210             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 211             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 212
 213             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 214             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 215             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 216             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 217             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 218             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 219             Heps             = _mm256_mul_pd(vfeps,H);
 220             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 221             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 222             velec            = _mm256_mul_pd(qq00,VV);
 223             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 224             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 225
 226             /* CUBIC SPLINE TABLE DISPERSION */
 227             vfitab           = _mm_add_epi32(vfitab,ifour);
 228             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 229             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 230             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 231             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 232             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 233             Heps             = _mm256_mul_pd(vfeps,H);
 234             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 235             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 236             vvdw6            = _mm256_mul_pd(c6_00,VV);
 237             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 238             fvdw6            = _mm256_mul_pd(c6_00,FF);
 239
 240             /* CUBIC SPLINE TABLE REPULSION */
 241             vfitab           = _mm_add_epi32(vfitab,ifour);
 242             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 243             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 244             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 245             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 246             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 247             Heps             = _mm256_mul_pd(vfeps,H);
 248             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 249             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 250             vvdw12           = _mm256_mul_pd(c12_00,VV);
 251             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 252             fvdw12           = _mm256_mul_pd(c12_00,FF);
 253             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 254             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 255
 256             /* Update potential sum for this i atom from the interaction with this j atom. */
 257             velecsum         = _mm256_add_pd(velecsum,velec);
 258             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 259
 260             fscal            = _mm256_add_pd(felec,fvdw);
 261
 262             /* Calculate temporary vectorial force */
 263             tx               = _mm256_mul_pd(fscal,dx00);
 264             ty               = _mm256_mul_pd(fscal,dy00);
 265             tz               = _mm256_mul_pd(fscal,dz00);
 266
 267             /* Update vectorial force */
 268             fix0             = _mm256_add_pd(fix0,tx);
 269             fiy0             = _mm256_add_pd(fiy0,ty);
 270             fiz0             = _mm256_add_pd(fiz0,tz);
 271
 272             fjptrA             = f+j_coord_offsetA;
 273             fjptrB             = f+j_coord_offsetB;
 274             fjptrC             = f+j_coord_offsetC;
 275             fjptrD             = f+j_coord_offsetD;
 276             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 277
 278             /* Inner loop uses 73 flops */
 279         }
 280
 281         if(jidx<j_index_end)
 282         {
 283
 284             /* Get j neighbor index, and coordinate index */
 285             jnrlistA         = jjnr[jidx];
 286             jnrlistB         = jjnr[jidx+1];
 287             jnrlistC         = jjnr[jidx+2];
 288             jnrlistD         = jjnr[jidx+3];
 289             /* Sign of each element will be negative for non-real atoms.
 290              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 291              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 292              */
 293             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 294
 295             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 296             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 297             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 298
 299             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 300             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 301             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 302             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 303             j_coord_offsetA  = DIM*jnrA;
 304             j_coord_offsetB  = DIM*jnrB;
 305             j_coord_offsetC  = DIM*jnrC;
 306             j_coord_offsetD  = DIM*jnrD;
 307
 308             /* load j atom coordinates */
 309             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 310                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 311                                                  &jx0,&jy0,&jz0);
 312
 313             /* Calculate displacement vector */
 314             dx00             = _mm256_sub_pd(ix0,jx0);
 315             dy00             = _mm256_sub_pd(iy0,jy0);
 316             dz00             = _mm256_sub_pd(iz0,jz0);
 317
 318             /* Calculate squared distance and things based on it */
 319             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 320
 321             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 322
 323             /* Load parameters for j particles */
 324             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 325                                                                  charge+jnrC+0,charge+jnrD+0);
 326             vdwjidx0A        = 2*vdwtype[jnrA+0];
 327             vdwjidx0B        = 2*vdwtype[jnrB+0];
 328             vdwjidx0C        = 2*vdwtype[jnrC+0];
 329             vdwjidx0D        = 2*vdwtype[jnrD+0];
 330
 331             /**************************
 332              * CALCULATE INTERACTIONS *
 333              **************************/
 334
 335             r00              = _mm256_mul_pd(rsq00,rinv00);
 336             r00              = _mm256_andnot_pd(dummy_mask,r00);
 337
 338             /* Compute parameters for interactions between i and j atoms */
 339             qq00             = _mm256_mul_pd(iq0,jq0);
 340             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 341                                             vdwioffsetptr0+vdwjidx0B,
 342                                             vdwioffsetptr0+vdwjidx0C,
 343                                             vdwioffsetptr0+vdwjidx0D,
 344                                             &c6_00,&c12_00);
 345
 346             /* Calculate table index by multiplying r with table scale and truncate to integer */
 347             rt               = _mm256_mul_pd(r00,vftabscale);
 348             vfitab           = _mm256_cvttpd_epi32(rt);
 349             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 350             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 351
 352             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 353             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 354             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 355             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 356             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 357             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 358             Heps             = _mm256_mul_pd(vfeps,H);
 359             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 360             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 361             velec            = _mm256_mul_pd(qq00,VV);
 362             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 363             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 364
 365             /* CUBIC SPLINE TABLE DISPERSION */
 366             vfitab           = _mm_add_epi32(vfitab,ifour);
 367             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 368             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 369             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 370             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 371             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 372             Heps             = _mm256_mul_pd(vfeps,H);
 373             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 374             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 375             vvdw6            = _mm256_mul_pd(c6_00,VV);
 376             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 377             fvdw6            = _mm256_mul_pd(c6_00,FF);
 378
 379             /* CUBIC SPLINE TABLE REPULSION */
 380             vfitab           = _mm_add_epi32(vfitab,ifour);
 381             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 382             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 383             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 384             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 385             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 386             Heps             = _mm256_mul_pd(vfeps,H);
 387             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 388             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 389             vvdw12           = _mm256_mul_pd(c12_00,VV);
 390             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 391             fvdw12           = _mm256_mul_pd(c12_00,FF);
 392             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 393             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 394
 395             /* Update potential sum for this i atom from the interaction with this j atom. */
 396             velec            = _mm256_andnot_pd(dummy_mask,velec);
 397             velecsum         = _mm256_add_pd(velecsum,velec);
 398             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 399             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 400
 401             fscal            = _mm256_add_pd(felec,fvdw);
 402
 403             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 404
 405             /* Calculate temporary vectorial force */
 406             tx               = _mm256_mul_pd(fscal,dx00);
 407             ty               = _mm256_mul_pd(fscal,dy00);
 408             tz               = _mm256_mul_pd(fscal,dz00);
 409
 410             /* Update vectorial force */
 411             fix0             = _mm256_add_pd(fix0,tx);
 412             fiy0             = _mm256_add_pd(fiy0,ty);
 413             fiz0             = _mm256_add_pd(fiz0,tz);
 414
 415             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 416             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 417             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 418             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 419             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 420
 421             /* Inner loop uses 74 flops */
 422         }
 423
 424         /* End of innermost loop */
 425
 426         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 427                                                  f+i_coord_offset,fshift+i_shift_offset);
 428
 429         ggid                        = gid[iidx];
 430         /* Update potential energies */
 431         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 432         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 433
 434         /* Increment number of inner iterations */
 435         inneriter                  += j_index_end - j_index_start;
 436
 437         /* Outer loop uses 9 flops */
 438     }
 439
 440     /* Increment number of outer iterations */
 441     outeriter        += nri;
 442
 443     /* Update outer/inner flops */
 444
 445     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*74);
 446 }
 447 /*
 448  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_F_avx_256_double
 449  * Electrostatics interaction: CubicSplineTable
 450  * VdW interaction:            CubicSplineTable
 451  * Geometry:                   Particle-Particle
 452  * Calculate force/pot:        Force
 453  */
 454 void
 455 nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_F_avx_256_double
 456                     (t_nblist * gmx_restrict                nlist,
 457                      rvec * gmx_restrict                    xx,
 458                      rvec * gmx_restrict                    ff,
 459                      t_forcerec * gmx_restrict              fr,
 460                      t_mdatoms * gmx_restrict               mdatoms,
 461                      nb_kernel_data_t * gmx_restrict        kernel_data,
 462                      t_nrnb * gmx_restrict                  nrnb)
 463 {
 464     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 465      * just 0 for non-waters.
 466      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 467      * jnr indices corresponding to data put in the four positions in the SIMD register.
 468      */
 469     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 470     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 471     int              jnrA,jnrB,jnrC,jnrD;
 472     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 473     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 474     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 475     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 476     real             rcutoff_scalar;
 477     real             *shiftvec,*fshift,*x,*f;
 478     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 479     real             scratch[4*DIM];
 480     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 481     real *           vdwioffsetptr0;
 482     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 483     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 484     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 485     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 486     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 487     real             *charge;
 488     int              nvdwtype;
 489     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 490     int              *vdwtype;
 491     real             *vdwparam;
 492     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 493     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 494     __m128i          vfitab;
 495     __m128i          ifour       = _mm_set1_epi32(4);
 496     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 497     real             *vftab;
 498     __m256d          dummy_mask,cutoff_mask;
 499     __m128           tmpmask0,tmpmask1;
 500     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 501     __m256d          one     = _mm256_set1_pd(1.0);
 502     __m256d          two     = _mm256_set1_pd(2.0);
 503     x                = xx[0];
 504     f                = ff[0];
 505
 506     nri              = nlist->nri;
 507     iinr             = nlist->iinr;
 508     jindex           = nlist->jindex;
 509     jjnr             = nlist->jjnr;
 510     shiftidx         = nlist->shift;
 511     gid              = nlist->gid;
 512     shiftvec         = fr->shift_vec[0];
 513     fshift           = fr->fshift[0];
 514     facel            = _mm256_set1_pd(fr->epsfac);
 515     charge           = mdatoms->chargeA;
 516     nvdwtype         = fr->ntype;
 517     vdwparam         = fr->nbfp;
 518     vdwtype          = mdatoms->typeA;
 519
 520     vftab            = kernel_data->table_elec_vdw->data;
 521     vftabscale       = _mm256_set1_pd(kernel_data->table_elec_vdw->scale);
 522
 523     /* Avoid stupid compiler warnings */
 524     jnrA = jnrB = jnrC = jnrD = 0;
 525     j_coord_offsetA = 0;
 526     j_coord_offsetB = 0;
 527     j_coord_offsetC = 0;
 528     j_coord_offsetD = 0;
 529
 530     outeriter        = 0;
 531     inneriter        = 0;
 532
 533     for(iidx=0;iidx<4*DIM;iidx++)
 534     {
 535         scratch[iidx] = 0.0;
 536     }
 537
 538     /* Start outer loop over neighborlists */
 539     for(iidx=0; iidx<nri; iidx++)
 540     {
 541         /* Load shift vector for this list */
 542         i_shift_offset   = DIM*shiftidx[iidx];
 543
 544         /* Load limits for loop over neighbors */
 545         j_index_start    = jindex[iidx];
 546         j_index_end      = jindex[iidx+1];
 547
 548         /* Get outer coordinate index */
 549         inr              = iinr[iidx];
 550         i_coord_offset   = DIM*inr;
 551
 552         /* Load i particle coords and add shift vector */
 553         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 554
 555         fix0             = _mm256_setzero_pd();
 556         fiy0             = _mm256_setzero_pd();
 557         fiz0             = _mm256_setzero_pd();
 558
 559         /* Load parameters for i particles */
 560         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 561         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 562
 563         /* Start inner kernel loop */
 564         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 565         {
 566
 567             /* Get j neighbor index, and coordinate index */
 568             jnrA             = jjnr[jidx];
 569             jnrB             = jjnr[jidx+1];
 570             jnrC             = jjnr[jidx+2];
 571             jnrD             = jjnr[jidx+3];
 572             j_coord_offsetA  = DIM*jnrA;
 573             j_coord_offsetB  = DIM*jnrB;
 574             j_coord_offsetC  = DIM*jnrC;
 575             j_coord_offsetD  = DIM*jnrD;
 576
 577             /* load j atom coordinates */
 578             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 579                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 580                                                  &jx0,&jy0,&jz0);
 581
 582             /* Calculate displacement vector */
 583             dx00             = _mm256_sub_pd(ix0,jx0);
 584             dy00             = _mm256_sub_pd(iy0,jy0);
 585             dz00             = _mm256_sub_pd(iz0,jz0);
 586
 587             /* Calculate squared distance and things based on it */
 588             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 589
 590             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 591
 592             /* Load parameters for j particles */
 593             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 594                                                                  charge+jnrC+0,charge+jnrD+0);
 595             vdwjidx0A        = 2*vdwtype[jnrA+0];
 596             vdwjidx0B        = 2*vdwtype[jnrB+0];
 597             vdwjidx0C        = 2*vdwtype[jnrC+0];
 598             vdwjidx0D        = 2*vdwtype[jnrD+0];
 599
 600             /**************************
 601              * CALCULATE INTERACTIONS *
 602              **************************/
 603
 604             r00              = _mm256_mul_pd(rsq00,rinv00);
 605
 606             /* Compute parameters for interactions between i and j atoms */
 607             qq00             = _mm256_mul_pd(iq0,jq0);
 608             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 609                                             vdwioffsetptr0+vdwjidx0B,
 610                                             vdwioffsetptr0+vdwjidx0C,
 611                                             vdwioffsetptr0+vdwjidx0D,
 612                                             &c6_00,&c12_00);
 613
 614             /* Calculate table index by multiplying r with table scale and truncate to integer */
 615             rt               = _mm256_mul_pd(r00,vftabscale);
 616             vfitab           = _mm256_cvttpd_epi32(rt);
 617             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 618             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 619
 620             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 621             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 622             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 623             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 624             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 625             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 626             Heps             = _mm256_mul_pd(vfeps,H);
 627             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 628             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 629             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 630
 631             /* CUBIC SPLINE TABLE DISPERSION */
 632             vfitab           = _mm_add_epi32(vfitab,ifour);
 633             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 634             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 635             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 636             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 637             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 638             Heps             = _mm256_mul_pd(vfeps,H);
 639             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 640             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 641             fvdw6            = _mm256_mul_pd(c6_00,FF);
 642
 643             /* CUBIC SPLINE TABLE REPULSION */
 644             vfitab           = _mm_add_epi32(vfitab,ifour);
 645             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 646             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 647             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 648             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 649             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 650             Heps             = _mm256_mul_pd(vfeps,H);
 651             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 652             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 653             fvdw12           = _mm256_mul_pd(c12_00,FF);
 654             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 655
 656             fscal            = _mm256_add_pd(felec,fvdw);
 657
 658             /* Calculate temporary vectorial force */
 659             tx               = _mm256_mul_pd(fscal,dx00);
 660             ty               = _mm256_mul_pd(fscal,dy00);
 661             tz               = _mm256_mul_pd(fscal,dz00);
 662
 663             /* Update vectorial force */
 664             fix0             = _mm256_add_pd(fix0,tx);
 665             fiy0             = _mm256_add_pd(fiy0,ty);
 666             fiz0             = _mm256_add_pd(fiz0,tz);
 667
 668             fjptrA             = f+j_coord_offsetA;
 669             fjptrB             = f+j_coord_offsetB;
 670             fjptrC             = f+j_coord_offsetC;
 671             fjptrD             = f+j_coord_offsetD;
 672             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 673
 674             /* Inner loop uses 61 flops */
 675         }
 676
 677         if(jidx<j_index_end)
 678         {
 679
 680             /* Get j neighbor index, and coordinate index */
 681             jnrlistA         = jjnr[jidx];
 682             jnrlistB         = jjnr[jidx+1];
 683             jnrlistC         = jjnr[jidx+2];
 684             jnrlistD         = jjnr[jidx+3];
 685             /* Sign of each element will be negative for non-real atoms.
 686              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 687              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 688              */
 689             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 690
 691             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 692             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 693             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 694
 695             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 696             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 697             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 698             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 699             j_coord_offsetA  = DIM*jnrA;
 700             j_coord_offsetB  = DIM*jnrB;
 701             j_coord_offsetC  = DIM*jnrC;
 702             j_coord_offsetD  = DIM*jnrD;
 703
 704             /* load j atom coordinates */
 705             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 706                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 707                                                  &jx0,&jy0,&jz0);
 708
 709             /* Calculate displacement vector */
 710             dx00             = _mm256_sub_pd(ix0,jx0);
 711             dy00             = _mm256_sub_pd(iy0,jy0);
 712             dz00             = _mm256_sub_pd(iz0,jz0);
 713
 714             /* Calculate squared distance and things based on it */
 715             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 716
 717             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 718
 719             /* Load parameters for j particles */
 720             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 721                                                                  charge+jnrC+0,charge+jnrD+0);
 722             vdwjidx0A        = 2*vdwtype[jnrA+0];
 723             vdwjidx0B        = 2*vdwtype[jnrB+0];
 724             vdwjidx0C        = 2*vdwtype[jnrC+0];
 725             vdwjidx0D        = 2*vdwtype[jnrD+0];
 726
 727             /**************************
 728              * CALCULATE INTERACTIONS *
 729              **************************/
 730
 731             r00              = _mm256_mul_pd(rsq00,rinv00);
 732             r00              = _mm256_andnot_pd(dummy_mask,r00);
 733
 734             /* Compute parameters for interactions between i and j atoms */
 735             qq00             = _mm256_mul_pd(iq0,jq0);
 736             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 737                                             vdwioffsetptr0+vdwjidx0B,
 738                                             vdwioffsetptr0+vdwjidx0C,
 739                                             vdwioffsetptr0+vdwjidx0D,
 740                                             &c6_00,&c12_00);
 741
 742             /* Calculate table index by multiplying r with table scale and truncate to integer */
 743             rt               = _mm256_mul_pd(r00,vftabscale);
 744             vfitab           = _mm256_cvttpd_epi32(rt);
 745             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 746             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 747
 748             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 749             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 750             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 751             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 752             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 753             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 754             Heps             = _mm256_mul_pd(vfeps,H);
 755             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 756             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 757             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 758
 759             /* CUBIC SPLINE TABLE DISPERSION */
 760             vfitab           = _mm_add_epi32(vfitab,ifour);
 761             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 762             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 763             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 764             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 765             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 766             Heps             = _mm256_mul_pd(vfeps,H);
 767             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 768             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 769             fvdw6            = _mm256_mul_pd(c6_00,FF);
 770
 771             /* CUBIC SPLINE TABLE REPULSION */
 772             vfitab           = _mm_add_epi32(vfitab,ifour);
 773             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 774             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 775             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 776             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 777             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 778             Heps             = _mm256_mul_pd(vfeps,H);
 779             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 780             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 781             fvdw12           = _mm256_mul_pd(c12_00,FF);
 782             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 783
 784             fscal            = _mm256_add_pd(felec,fvdw);
 785
 786             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 787
 788             /* Calculate temporary vectorial force */
 789             tx               = _mm256_mul_pd(fscal,dx00);
 790             ty               = _mm256_mul_pd(fscal,dy00);
 791             tz               = _mm256_mul_pd(fscal,dz00);
 792
 793             /* Update vectorial force */
 794             fix0             = _mm256_add_pd(fix0,tx);
 795             fiy0             = _mm256_add_pd(fiy0,ty);
 796             fiz0             = _mm256_add_pd(fiz0,tz);
 797
 798             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 799             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 800             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 801             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 802             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 803
 804             /* Inner loop uses 62 flops */
 805         }
 806
 807         /* End of innermost loop */
 808
 809         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 810                                                  f+i_coord_offset,fshift+i_shift_offset);
 811
 812         /* Increment number of inner iterations */
 813         inneriter                  += j_index_end - j_index_start;
 814
 815         /* Outer loop uses 7 flops */
 816     }
 817
 818     /* Increment number of outer iterations */
 819     outeriter        += nri;
 820
 821     /* Update outer/inner flops */
 822
 823     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*62);
 824 }