src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_avx_256_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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_ElecCSTab_VdwCSTab_GeomP1P1_VF_avx_256_double
  51  * Electrostatics interaction: CubicSplineTable
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCSTab_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     nvdwtype         = fr->ntype;
 119     vdwparam         = fr->nbfp;
 120     vdwtype          = mdatoms->typeA;
 121
 122     vftab            = kernel_data->table_elec_vdw->data;
 123     vftabscale       = _mm256_set1_pd(kernel_data->table_elec_vdw->scale);
 124
 125     /* Avoid stupid compiler warnings */
 126     jnrA = jnrB = jnrC = jnrD = 0;
 127     j_coord_offsetA = 0;
 128     j_coord_offsetB = 0;
 129     j_coord_offsetC = 0;
 130     j_coord_offsetD = 0;
 131
 132     outeriter        = 0;
 133     inneriter        = 0;
 134
 135     for(iidx=0;iidx<4*DIM;iidx++)
 136     {
 137         scratch[iidx] = 0.0;
 138     }
 139
 140     /* Start outer loop over neighborlists */
 141     for(iidx=0; iidx<nri; iidx++)
 142     {
 143         /* Load shift vector for this list */
 144         i_shift_offset   = DIM*shiftidx[iidx];
 145
 146         /* Load limits for loop over neighbors */
 147         j_index_start    = jindex[iidx];
 148         j_index_end      = jindex[iidx+1];
 149
 150         /* Get outer coordinate index */
 151         inr              = iinr[iidx];
 152         i_coord_offset   = DIM*inr;
 153
 154         /* Load i particle coords and add shift vector */
 155         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 156
 157         fix0             = _mm256_setzero_pd();
 158         fiy0             = _mm256_setzero_pd();
 159         fiz0             = _mm256_setzero_pd();
 160
 161         /* Load parameters for i particles */
 162         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 163         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 164
 165         /* Reset potential sums */
 166         velecsum         = _mm256_setzero_pd();
 167         vvdwsum          = _mm256_setzero_pd();
 168
 169         /* Start inner kernel loop */
 170         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 171         {
 172
 173             /* Get j neighbor index, and coordinate index */
 174             jnrA             = jjnr[jidx];
 175             jnrB             = jjnr[jidx+1];
 176             jnrC             = jjnr[jidx+2];
 177             jnrD             = jjnr[jidx+3];
 178             j_coord_offsetA  = DIM*jnrA;
 179             j_coord_offsetB  = DIM*jnrB;
 180             j_coord_offsetC  = DIM*jnrC;
 181             j_coord_offsetD  = DIM*jnrD;
 182
 183             /* load j atom coordinates */
 184             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 185                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 186                                                  &jx0,&jy0,&jz0);
 187
 188             /* Calculate displacement vector */
 189             dx00             = _mm256_sub_pd(ix0,jx0);
 190             dy00             = _mm256_sub_pd(iy0,jy0);
 191             dz00             = _mm256_sub_pd(iz0,jz0);
 192
 193             /* Calculate squared distance and things based on it */
 194             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 195
 196             rinv00           = avx256_invsqrt_d(rsq00);
 197
 198             /* Load parameters for j particles */
 199             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 200                                                                  charge+jnrC+0,charge+jnrD+0);
 201             vdwjidx0A        = 2*vdwtype[jnrA+0];
 202             vdwjidx0B        = 2*vdwtype[jnrB+0];
 203             vdwjidx0C        = 2*vdwtype[jnrC+0];
 204             vdwjidx0D        = 2*vdwtype[jnrD+0];
 205
 206             /**************************
 207              * CALCULATE INTERACTIONS *
 208              **************************/
 209
 210             r00              = _mm256_mul_pd(rsq00,rinv00);
 211
 212             /* Compute parameters for interactions between i and j atoms */
 213             qq00             = _mm256_mul_pd(iq0,jq0);
 214             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 215                                             vdwioffsetptr0+vdwjidx0B,
 216                                             vdwioffsetptr0+vdwjidx0C,
 217                                             vdwioffsetptr0+vdwjidx0D,
 218                                             &c6_00,&c12_00);
 219
 220             /* Calculate table index by multiplying r with table scale and truncate to integer */
 221             rt               = _mm256_mul_pd(r00,vftabscale);
 222             vfitab           = _mm256_cvttpd_epi32(rt);
 223             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 224             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 225
 226             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 227             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 228             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 229             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 230             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 231             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 232             Heps             = _mm256_mul_pd(vfeps,H);
 233             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 234             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 235             velec            = _mm256_mul_pd(qq00,VV);
 236             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 237             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 238
 239             /* CUBIC SPLINE TABLE DISPERSION */
 240             vfitab           = _mm_add_epi32(vfitab,ifour);
 241             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 242             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 243             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 244             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 245             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 246             Heps             = _mm256_mul_pd(vfeps,H);
 247             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 248             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 249             vvdw6            = _mm256_mul_pd(c6_00,VV);
 250             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 251             fvdw6            = _mm256_mul_pd(c6_00,FF);
 252
 253             /* CUBIC SPLINE TABLE REPULSION */
 254             vfitab           = _mm_add_epi32(vfitab,ifour);
 255             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 256             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 257             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 258             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 259             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 260             Heps             = _mm256_mul_pd(vfeps,H);
 261             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 262             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 263             vvdw12           = _mm256_mul_pd(c12_00,VV);
 264             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 265             fvdw12           = _mm256_mul_pd(c12_00,FF);
 266             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 267             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 268
 269             /* Update potential sum for this i atom from the interaction with this j atom. */
 270             velecsum         = _mm256_add_pd(velecsum,velec);
 271             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 272
 273             fscal            = _mm256_add_pd(felec,fvdw);
 274
 275             /* Calculate temporary vectorial force */
 276             tx               = _mm256_mul_pd(fscal,dx00);
 277             ty               = _mm256_mul_pd(fscal,dy00);
 278             tz               = _mm256_mul_pd(fscal,dz00);
 279
 280             /* Update vectorial force */
 281             fix0             = _mm256_add_pd(fix0,tx);
 282             fiy0             = _mm256_add_pd(fiy0,ty);
 283             fiz0             = _mm256_add_pd(fiz0,tz);
 284
 285             fjptrA             = f+j_coord_offsetA;
 286             fjptrB             = f+j_coord_offsetB;
 287             fjptrC             = f+j_coord_offsetC;
 288             fjptrD             = f+j_coord_offsetD;
 289             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 290
 291             /* Inner loop uses 73 flops */
 292         }
 293
 294         if(jidx<j_index_end)
 295         {
 296
 297             /* Get j neighbor index, and coordinate index */
 298             jnrlistA         = jjnr[jidx];
 299             jnrlistB         = jjnr[jidx+1];
 300             jnrlistC         = jjnr[jidx+2];
 301             jnrlistD         = jjnr[jidx+3];
 302             /* Sign of each element will be negative for non-real atoms.
 303              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 304              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 305              */
 306             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 307
 308             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 309             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 310             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 311
 312             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 313             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 314             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 315             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 316             j_coord_offsetA  = DIM*jnrA;
 317             j_coord_offsetB  = DIM*jnrB;
 318             j_coord_offsetC  = DIM*jnrC;
 319             j_coord_offsetD  = DIM*jnrD;
 320
 321             /* load j atom coordinates */
 322             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 323                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 324                                                  &jx0,&jy0,&jz0);
 325
 326             /* Calculate displacement vector */
 327             dx00             = _mm256_sub_pd(ix0,jx0);
 328             dy00             = _mm256_sub_pd(iy0,jy0);
 329             dz00             = _mm256_sub_pd(iz0,jz0);
 330
 331             /* Calculate squared distance and things based on it */
 332             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 333
 334             rinv00           = avx256_invsqrt_d(rsq00);
 335
 336             /* Load parameters for j particles */
 337             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 338                                                                  charge+jnrC+0,charge+jnrD+0);
 339             vdwjidx0A        = 2*vdwtype[jnrA+0];
 340             vdwjidx0B        = 2*vdwtype[jnrB+0];
 341             vdwjidx0C        = 2*vdwtype[jnrC+0];
 342             vdwjidx0D        = 2*vdwtype[jnrD+0];
 343
 344             /**************************
 345              * CALCULATE INTERACTIONS *
 346              **************************/
 347
 348             r00              = _mm256_mul_pd(rsq00,rinv00);
 349             r00              = _mm256_andnot_pd(dummy_mask,r00);
 350
 351             /* Compute parameters for interactions between i and j atoms */
 352             qq00             = _mm256_mul_pd(iq0,jq0);
 353             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 354                                             vdwioffsetptr0+vdwjidx0B,
 355                                             vdwioffsetptr0+vdwjidx0C,
 356                                             vdwioffsetptr0+vdwjidx0D,
 357                                             &c6_00,&c12_00);
 358
 359             /* Calculate table index by multiplying r with table scale and truncate to integer */
 360             rt               = _mm256_mul_pd(r00,vftabscale);
 361             vfitab           = _mm256_cvttpd_epi32(rt);
 362             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 363             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 364
 365             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 366             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 367             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 368             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 369             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 370             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 371             Heps             = _mm256_mul_pd(vfeps,H);
 372             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 373             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 374             velec            = _mm256_mul_pd(qq00,VV);
 375             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 376             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 377
 378             /* CUBIC SPLINE TABLE DISPERSION */
 379             vfitab           = _mm_add_epi32(vfitab,ifour);
 380             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 381             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 382             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 383             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 384             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 385             Heps             = _mm256_mul_pd(vfeps,H);
 386             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 387             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 388             vvdw6            = _mm256_mul_pd(c6_00,VV);
 389             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 390             fvdw6            = _mm256_mul_pd(c6_00,FF);
 391
 392             /* CUBIC SPLINE TABLE REPULSION */
 393             vfitab           = _mm_add_epi32(vfitab,ifour);
 394             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 395             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 396             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 397             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 398             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 399             Heps             = _mm256_mul_pd(vfeps,H);
 400             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 401             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 402             vvdw12           = _mm256_mul_pd(c12_00,VV);
 403             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 404             fvdw12           = _mm256_mul_pd(c12_00,FF);
 405             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 406             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 407
 408             /* Update potential sum for this i atom from the interaction with this j atom. */
 409             velec            = _mm256_andnot_pd(dummy_mask,velec);
 410             velecsum         = _mm256_add_pd(velecsum,velec);
 411             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 412             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 413
 414             fscal            = _mm256_add_pd(felec,fvdw);
 415
 416             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 417
 418             /* Calculate temporary vectorial force */
 419             tx               = _mm256_mul_pd(fscal,dx00);
 420             ty               = _mm256_mul_pd(fscal,dy00);
 421             tz               = _mm256_mul_pd(fscal,dz00);
 422
 423             /* Update vectorial force */
 424             fix0             = _mm256_add_pd(fix0,tx);
 425             fiy0             = _mm256_add_pd(fiy0,ty);
 426             fiz0             = _mm256_add_pd(fiz0,tz);
 427
 428             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 429             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 430             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 431             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 432             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 433
 434             /* Inner loop uses 74 flops */
 435         }
 436
 437         /* End of innermost loop */
 438
 439         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 440                                                  f+i_coord_offset,fshift+i_shift_offset);
 441
 442         ggid                        = gid[iidx];
 443         /* Update potential energies */
 444         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 445         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 446
 447         /* Increment number of inner iterations */
 448         inneriter                  += j_index_end - j_index_start;
 449
 450         /* Outer loop uses 9 flops */
 451     }
 452
 453     /* Increment number of outer iterations */
 454     outeriter        += nri;
 455
 456     /* Update outer/inner flops */
 457
 458     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*74);
 459 }
 460 /*
 461  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_F_avx_256_double
 462  * Electrostatics interaction: CubicSplineTable
 463  * VdW interaction:            CubicSplineTable
 464  * Geometry:                   Particle-Particle
 465  * Calculate force/pot:        Force
 466  */
 467 void
 468 nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_F_avx_256_double
 469                     (t_nblist                    * gmx_restrict       nlist,
 470                      rvec                        * gmx_restrict          xx,
 471                      rvec                        * gmx_restrict          ff,
 472                      struct t_forcerec           * gmx_restrict          fr,
 473                      t_mdatoms                   * gmx_restrict     mdatoms,
 474                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 475                      t_nrnb                      * gmx_restrict        nrnb)
 476 {
 477     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 478      * just 0 for non-waters.
 479      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 480      * jnr indices corresponding to data put in the four positions in the SIMD register.
 481      */
 482     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 483     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 484     int              jnrA,jnrB,jnrC,jnrD;
 485     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 486     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 487     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 488     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 489     real             rcutoff_scalar;
 490     real             *shiftvec,*fshift,*x,*f;
 491     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 492     real             scratch[4*DIM];
 493     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 494     real *           vdwioffsetptr0;
 495     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 496     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 497     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 498     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 499     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 500     real             *charge;
 501     int              nvdwtype;
 502     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 503     int              *vdwtype;
 504     real             *vdwparam;
 505     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 506     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 507     __m128i          vfitab;
 508     __m128i          ifour       = _mm_set1_epi32(4);
 509     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 510     real             *vftab;
 511     __m256d          dummy_mask,cutoff_mask;
 512     __m128           tmpmask0,tmpmask1;
 513     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 514     __m256d          one     = _mm256_set1_pd(1.0);
 515     __m256d          two     = _mm256_set1_pd(2.0);
 516     x                = xx[0];
 517     f                = ff[0];
 518
 519     nri              = nlist->nri;
 520     iinr             = nlist->iinr;
 521     jindex           = nlist->jindex;
 522     jjnr             = nlist->jjnr;
 523     shiftidx         = nlist->shift;
 524     gid              = nlist->gid;
 525     shiftvec         = fr->shift_vec[0];
 526     fshift           = fr->fshift[0];
 527     facel            = _mm256_set1_pd(fr->ic->epsfac);
 528     charge           = mdatoms->chargeA;
 529     nvdwtype         = fr->ntype;
 530     vdwparam         = fr->nbfp;
 531     vdwtype          = mdatoms->typeA;
 532
 533     vftab            = kernel_data->table_elec_vdw->data;
 534     vftabscale       = _mm256_set1_pd(kernel_data->table_elec_vdw->scale);
 535
 536     /* Avoid stupid compiler warnings */
 537     jnrA = jnrB = jnrC = jnrD = 0;
 538     j_coord_offsetA = 0;
 539     j_coord_offsetB = 0;
 540     j_coord_offsetC = 0;
 541     j_coord_offsetD = 0;
 542
 543     outeriter        = 0;
 544     inneriter        = 0;
 545
 546     for(iidx=0;iidx<4*DIM;iidx++)
 547     {
 548         scratch[iidx] = 0.0;
 549     }
 550
 551     /* Start outer loop over neighborlists */
 552     for(iidx=0; iidx<nri; iidx++)
 553     {
 554         /* Load shift vector for this list */
 555         i_shift_offset   = DIM*shiftidx[iidx];
 556
 557         /* Load limits for loop over neighbors */
 558         j_index_start    = jindex[iidx];
 559         j_index_end      = jindex[iidx+1];
 560
 561         /* Get outer coordinate index */
 562         inr              = iinr[iidx];
 563         i_coord_offset   = DIM*inr;
 564
 565         /* Load i particle coords and add shift vector */
 566         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 567
 568         fix0             = _mm256_setzero_pd();
 569         fiy0             = _mm256_setzero_pd();
 570         fiz0             = _mm256_setzero_pd();
 571
 572         /* Load parameters for i particles */
 573         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 574         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 575
 576         /* Start inner kernel loop */
 577         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 578         {
 579
 580             /* Get j neighbor index, and coordinate index */
 581             jnrA             = jjnr[jidx];
 582             jnrB             = jjnr[jidx+1];
 583             jnrC             = jjnr[jidx+2];
 584             jnrD             = jjnr[jidx+3];
 585             j_coord_offsetA  = DIM*jnrA;
 586             j_coord_offsetB  = DIM*jnrB;
 587             j_coord_offsetC  = DIM*jnrC;
 588             j_coord_offsetD  = DIM*jnrD;
 589
 590             /* load j atom coordinates */
 591             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 592                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 593                                                  &jx0,&jy0,&jz0);
 594
 595             /* Calculate displacement vector */
 596             dx00             = _mm256_sub_pd(ix0,jx0);
 597             dy00             = _mm256_sub_pd(iy0,jy0);
 598             dz00             = _mm256_sub_pd(iz0,jz0);
 599
 600             /* Calculate squared distance and things based on it */
 601             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 602
 603             rinv00           = avx256_invsqrt_d(rsq00);
 604
 605             /* Load parameters for j particles */
 606             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 607                                                                  charge+jnrC+0,charge+jnrD+0);
 608             vdwjidx0A        = 2*vdwtype[jnrA+0];
 609             vdwjidx0B        = 2*vdwtype[jnrB+0];
 610             vdwjidx0C        = 2*vdwtype[jnrC+0];
 611             vdwjidx0D        = 2*vdwtype[jnrD+0];
 612
 613             /**************************
 614              * CALCULATE INTERACTIONS *
 615              **************************/
 616
 617             r00              = _mm256_mul_pd(rsq00,rinv00);
 618
 619             /* Compute parameters for interactions between i and j atoms */
 620             qq00             = _mm256_mul_pd(iq0,jq0);
 621             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 622                                             vdwioffsetptr0+vdwjidx0B,
 623                                             vdwioffsetptr0+vdwjidx0C,
 624                                             vdwioffsetptr0+vdwjidx0D,
 625                                             &c6_00,&c12_00);
 626
 627             /* Calculate table index by multiplying r with table scale and truncate to integer */
 628             rt               = _mm256_mul_pd(r00,vftabscale);
 629             vfitab           = _mm256_cvttpd_epi32(rt);
 630             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 631             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 632
 633             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 634             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 635             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 636             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 637             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 638             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 639             Heps             = _mm256_mul_pd(vfeps,H);
 640             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 641             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 642             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 643
 644             /* CUBIC SPLINE TABLE DISPERSION */
 645             vfitab           = _mm_add_epi32(vfitab,ifour);
 646             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 647             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 648             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 649             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 650             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 651             Heps             = _mm256_mul_pd(vfeps,H);
 652             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 653             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 654             fvdw6            = _mm256_mul_pd(c6_00,FF);
 655
 656             /* CUBIC SPLINE TABLE REPULSION */
 657             vfitab           = _mm_add_epi32(vfitab,ifour);
 658             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 659             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 660             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 661             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 662             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 663             Heps             = _mm256_mul_pd(vfeps,H);
 664             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 665             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 666             fvdw12           = _mm256_mul_pd(c12_00,FF);
 667             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 668
 669             fscal            = _mm256_add_pd(felec,fvdw);
 670
 671             /* Calculate temporary vectorial force */
 672             tx               = _mm256_mul_pd(fscal,dx00);
 673             ty               = _mm256_mul_pd(fscal,dy00);
 674             tz               = _mm256_mul_pd(fscal,dz00);
 675
 676             /* Update vectorial force */
 677             fix0             = _mm256_add_pd(fix0,tx);
 678             fiy0             = _mm256_add_pd(fiy0,ty);
 679             fiz0             = _mm256_add_pd(fiz0,tz);
 680
 681             fjptrA             = f+j_coord_offsetA;
 682             fjptrB             = f+j_coord_offsetB;
 683             fjptrC             = f+j_coord_offsetC;
 684             fjptrD             = f+j_coord_offsetD;
 685             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 686
 687             /* Inner loop uses 61 flops */
 688         }
 689
 690         if(jidx<j_index_end)
 691         {
 692
 693             /* Get j neighbor index, and coordinate index */
 694             jnrlistA         = jjnr[jidx];
 695             jnrlistB         = jjnr[jidx+1];
 696             jnrlistC         = jjnr[jidx+2];
 697             jnrlistD         = jjnr[jidx+3];
 698             /* Sign of each element will be negative for non-real atoms.
 699              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 700              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 701              */
 702             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 703
 704             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 705             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 706             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 707
 708             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 709             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 710             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 711             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 712             j_coord_offsetA  = DIM*jnrA;
 713             j_coord_offsetB  = DIM*jnrB;
 714             j_coord_offsetC  = DIM*jnrC;
 715             j_coord_offsetD  = DIM*jnrD;
 716
 717             /* load j atom coordinates */
 718             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 719                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 720                                                  &jx0,&jy0,&jz0);
 721
 722             /* Calculate displacement vector */
 723             dx00             = _mm256_sub_pd(ix0,jx0);
 724             dy00             = _mm256_sub_pd(iy0,jy0);
 725             dz00             = _mm256_sub_pd(iz0,jz0);
 726
 727             /* Calculate squared distance and things based on it */
 728             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 729
 730             rinv00           = avx256_invsqrt_d(rsq00);
 731
 732             /* Load parameters for j particles */
 733             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 734                                                                  charge+jnrC+0,charge+jnrD+0);
 735             vdwjidx0A        = 2*vdwtype[jnrA+0];
 736             vdwjidx0B        = 2*vdwtype[jnrB+0];
 737             vdwjidx0C        = 2*vdwtype[jnrC+0];
 738             vdwjidx0D        = 2*vdwtype[jnrD+0];
 739
 740             /**************************
 741              * CALCULATE INTERACTIONS *
 742              **************************/
 743
 744             r00              = _mm256_mul_pd(rsq00,rinv00);
 745             r00              = _mm256_andnot_pd(dummy_mask,r00);
 746
 747             /* Compute parameters for interactions between i and j atoms */
 748             qq00             = _mm256_mul_pd(iq0,jq0);
 749             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 750                                             vdwioffsetptr0+vdwjidx0B,
 751                                             vdwioffsetptr0+vdwjidx0C,
 752                                             vdwioffsetptr0+vdwjidx0D,
 753                                             &c6_00,&c12_00);
 754
 755             /* Calculate table index by multiplying r with table scale and truncate to integer */
 756             rt               = _mm256_mul_pd(r00,vftabscale);
 757             vfitab           = _mm256_cvttpd_epi32(rt);
 758             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 759             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 760
 761             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 762             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 763             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 764             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 765             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 766             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 767             Heps             = _mm256_mul_pd(vfeps,H);
 768             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 769             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 770             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 771
 772             /* CUBIC SPLINE TABLE DISPERSION */
 773             vfitab           = _mm_add_epi32(vfitab,ifour);
 774             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 775             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 776             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 777             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 778             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 779             Heps             = _mm256_mul_pd(vfeps,H);
 780             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 781             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 782             fvdw6            = _mm256_mul_pd(c6_00,FF);
 783
 784             /* CUBIC SPLINE TABLE REPULSION */
 785             vfitab           = _mm_add_epi32(vfitab,ifour);
 786             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 787             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 788             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 789             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 790             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 791             Heps             = _mm256_mul_pd(vfeps,H);
 792             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 793             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 794             fvdw12           = _mm256_mul_pd(c12_00,FF);
 795             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 796
 797             fscal            = _mm256_add_pd(felec,fvdw);
 798
 799             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 800
 801             /* Calculate temporary vectorial force */
 802             tx               = _mm256_mul_pd(fscal,dx00);
 803             ty               = _mm256_mul_pd(fscal,dy00);
 804             tz               = _mm256_mul_pd(fscal,dz00);
 805
 806             /* Update vectorial force */
 807             fix0             = _mm256_add_pd(fix0,tx);
 808             fiy0             = _mm256_add_pd(fiy0,ty);
 809             fiz0             = _mm256_add_pd(fiz0,tz);
 810
 811             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 812             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 813             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 814             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 815             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 816
 817             /* Inner loop uses 62 flops */
 818         }
 819
 820         /* End of innermost loop */
 821
 822         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 823                                                  f+i_coord_offset,fshift+i_shift_offset);
 824
 825         /* Increment number of inner iterations */
 826         inneriter                  += j_index_end - j_index_start;
 827
 828         /* Outer loop uses 7 flops */
 829     }
 830
 831     /* Increment number of outer iterations */
 832     outeriter        += nri;
 833
 834     /* Update outer/inner flops */
 835
 836     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*62);
 837 }