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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_VF_avx_256_double
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_VF_avx_256_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  89     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     int              nvdwtype;
  94     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  98     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  99     __m128i          vfitab;
 100     __m128i          ifour       = _mm_set1_epi32(4);
 101     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 102     real             *vftab;
 103     __m256d          dummy_mask,cutoff_mask;
 104     __m128           tmpmask0,tmpmask1;
 105     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 106     __m256d          one     = _mm256_set1_pd(1.0);
 107     __m256d          two     = _mm256_set1_pd(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm256_set1_pd(fr->epsfac);
 120     charge           = mdatoms->chargeA;
 121     nvdwtype         = fr->ntype;
 122     vdwparam         = fr->nbfp;
 123     vdwtype          = mdatoms->typeA;
 124
 125     vftab            = kernel_data->table_elec_vdw->data;
 126     vftabscale       = _mm256_set1_pd(kernel_data->table_elec_vdw->scale);
 127
 128     /* Avoid stupid compiler warnings */
 129     jnrA = jnrB = jnrC = jnrD = 0;
 130     j_coord_offsetA = 0;
 131     j_coord_offsetB = 0;
 132     j_coord_offsetC = 0;
 133     j_coord_offsetD = 0;
 134
 135     outeriter        = 0;
 136     inneriter        = 0;
 137
 138     for(iidx=0;iidx<4*DIM;iidx++)
 139     {
 140         scratch[iidx] = 0.0;
 141     }
 142
 143     /* Start outer loop over neighborlists */
 144     for(iidx=0; iidx<nri; iidx++)
 145     {
 146         /* Load shift vector for this list */
 147         i_shift_offset   = DIM*shiftidx[iidx];
 148
 149         /* Load limits for loop over neighbors */
 150         j_index_start    = jindex[iidx];
 151         j_index_end      = jindex[iidx+1];
 152
 153         /* Get outer coordinate index */
 154         inr              = iinr[iidx];
 155         i_coord_offset   = DIM*inr;
 156
 157         /* Load i particle coords and add shift vector */
 158         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 159
 160         fix0             = _mm256_setzero_pd();
 161         fiy0             = _mm256_setzero_pd();
 162         fiz0             = _mm256_setzero_pd();
 163
 164         /* Load parameters for i particles */
 165         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 166         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 167
 168         /* Reset potential sums */
 169         velecsum         = _mm256_setzero_pd();
 170         vvdwsum          = _mm256_setzero_pd();
 171
 172         /* Start inner kernel loop */
 173         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 174         {
 175
 176             /* Get j neighbor index, and coordinate index */
 177             jnrA             = jjnr[jidx];
 178             jnrB             = jjnr[jidx+1];
 179             jnrC             = jjnr[jidx+2];
 180             jnrD             = jjnr[jidx+3];
 181             j_coord_offsetA  = DIM*jnrA;
 182             j_coord_offsetB  = DIM*jnrB;
 183             j_coord_offsetC  = DIM*jnrC;
 184             j_coord_offsetD  = DIM*jnrD;
 185
 186             /* load j atom coordinates */
 187             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 188                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 189                                                  &jx0,&jy0,&jz0);
 190
 191             /* Calculate displacement vector */
 192             dx00             = _mm256_sub_pd(ix0,jx0);
 193             dy00             = _mm256_sub_pd(iy0,jy0);
 194             dz00             = _mm256_sub_pd(iz0,jz0);
 195
 196             /* Calculate squared distance and things based on it */
 197             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 198
 199             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 200
 201             /* Load parameters for j particles */
 202             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 203                                                                  charge+jnrC+0,charge+jnrD+0);
 204             vdwjidx0A        = 2*vdwtype[jnrA+0];
 205             vdwjidx0B        = 2*vdwtype[jnrB+0];
 206             vdwjidx0C        = 2*vdwtype[jnrC+0];
 207             vdwjidx0D        = 2*vdwtype[jnrD+0];
 208
 209             /**************************
 210              * CALCULATE INTERACTIONS *
 211              **************************/
 212
 213             r00              = _mm256_mul_pd(rsq00,rinv00);
 214
 215             /* Compute parameters for interactions between i and j atoms */
 216             qq00             = _mm256_mul_pd(iq0,jq0);
 217             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 218                                             vdwioffsetptr0+vdwjidx0B,
 219                                             vdwioffsetptr0+vdwjidx0C,
 220                                             vdwioffsetptr0+vdwjidx0D,
 221                                             &c6_00,&c12_00);
 222
 223             /* Calculate table index by multiplying r with table scale and truncate to integer */
 224             rt               = _mm256_mul_pd(r00,vftabscale);
 225             vfitab           = _mm256_cvttpd_epi32(rt);
 226             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 227             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 228
 229             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 230             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 231             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 232             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 233             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 234             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 235             Heps             = _mm256_mul_pd(vfeps,H);
 236             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 237             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 238             velec            = _mm256_mul_pd(qq00,VV);
 239             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 240             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 241
 242             /* CUBIC SPLINE TABLE DISPERSION */
 243             vfitab           = _mm_add_epi32(vfitab,ifour);
 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             /* Update potential sum for this i atom from the interaction with this j atom. */
 273             velecsum         = _mm256_add_pd(velecsum,velec);
 274             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 275
 276             fscal            = _mm256_add_pd(felec,fvdw);
 277
 278             /* Calculate temporary vectorial force */
 279             tx               = _mm256_mul_pd(fscal,dx00);
 280             ty               = _mm256_mul_pd(fscal,dy00);
 281             tz               = _mm256_mul_pd(fscal,dz00);
 282
 283             /* Update vectorial force */
 284             fix0             = _mm256_add_pd(fix0,tx);
 285             fiy0             = _mm256_add_pd(fiy0,ty);
 286             fiz0             = _mm256_add_pd(fiz0,tz);
 287
 288             fjptrA             = f+j_coord_offsetA;
 289             fjptrB             = f+j_coord_offsetB;
 290             fjptrC             = f+j_coord_offsetC;
 291             fjptrD             = f+j_coord_offsetD;
 292             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 293
 294             /* Inner loop uses 73 flops */
 295         }
 296
 297         if(jidx<j_index_end)
 298         {
 299
 300             /* Get j neighbor index, and coordinate index */
 301             jnrlistA         = jjnr[jidx];
 302             jnrlistB         = jjnr[jidx+1];
 303             jnrlistC         = jjnr[jidx+2];
 304             jnrlistD         = jjnr[jidx+3];
 305             /* Sign of each element will be negative for non-real atoms.
 306              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 307              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 308              */
 309             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 310
 311             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 312             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 313             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 314
 315             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 316             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 317             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 318             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 319             j_coord_offsetA  = DIM*jnrA;
 320             j_coord_offsetB  = DIM*jnrB;
 321             j_coord_offsetC  = DIM*jnrC;
 322             j_coord_offsetD  = DIM*jnrD;
 323
 324             /* load j atom coordinates */
 325             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 326                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 327                                                  &jx0,&jy0,&jz0);
 328
 329             /* Calculate displacement vector */
 330             dx00             = _mm256_sub_pd(ix0,jx0);
 331             dy00             = _mm256_sub_pd(iy0,jy0);
 332             dz00             = _mm256_sub_pd(iz0,jz0);
 333
 334             /* Calculate squared distance and things based on it */
 335             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 336
 337             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 338
 339             /* Load parameters for j particles */
 340             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 341                                                                  charge+jnrC+0,charge+jnrD+0);
 342             vdwjidx0A        = 2*vdwtype[jnrA+0];
 343             vdwjidx0B        = 2*vdwtype[jnrB+0];
 344             vdwjidx0C        = 2*vdwtype[jnrC+0];
 345             vdwjidx0D        = 2*vdwtype[jnrD+0];
 346
 347             /**************************
 348              * CALCULATE INTERACTIONS *
 349              **************************/
 350
 351             r00              = _mm256_mul_pd(rsq00,rinv00);
 352             r00              = _mm256_andnot_pd(dummy_mask,r00);
 353
 354             /* Compute parameters for interactions between i and j atoms */
 355             qq00             = _mm256_mul_pd(iq0,jq0);
 356             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 357                                             vdwioffsetptr0+vdwjidx0B,
 358                                             vdwioffsetptr0+vdwjidx0C,
 359                                             vdwioffsetptr0+vdwjidx0D,
 360                                             &c6_00,&c12_00);
 361
 362             /* Calculate table index by multiplying r with table scale and truncate to integer */
 363             rt               = _mm256_mul_pd(r00,vftabscale);
 364             vfitab           = _mm256_cvttpd_epi32(rt);
 365             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 366             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 367
 368             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 369             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 370             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 371             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 372             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 373             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 374             Heps             = _mm256_mul_pd(vfeps,H);
 375             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 376             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 377             velec            = _mm256_mul_pd(qq00,VV);
 378             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 379             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 380
 381             /* CUBIC SPLINE TABLE DISPERSION */
 382             vfitab           = _mm_add_epi32(vfitab,ifour);
 383             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 384             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 385             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 386             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 387             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 388             Heps             = _mm256_mul_pd(vfeps,H);
 389             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 390             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 391             vvdw6            = _mm256_mul_pd(c6_00,VV);
 392             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 393             fvdw6            = _mm256_mul_pd(c6_00,FF);
 394
 395             /* CUBIC SPLINE TABLE REPULSION */
 396             vfitab           = _mm_add_epi32(vfitab,ifour);
 397             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 398             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 399             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 400             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 401             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 402             Heps             = _mm256_mul_pd(vfeps,H);
 403             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 404             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 405             vvdw12           = _mm256_mul_pd(c12_00,VV);
 406             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 407             fvdw12           = _mm256_mul_pd(c12_00,FF);
 408             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 409             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 410
 411             /* Update potential sum for this i atom from the interaction with this j atom. */
 412             velec            = _mm256_andnot_pd(dummy_mask,velec);
 413             velecsum         = _mm256_add_pd(velecsum,velec);
 414             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 415             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 416
 417             fscal            = _mm256_add_pd(felec,fvdw);
 418
 419             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 420
 421             /* Calculate temporary vectorial force */
 422             tx               = _mm256_mul_pd(fscal,dx00);
 423             ty               = _mm256_mul_pd(fscal,dy00);
 424             tz               = _mm256_mul_pd(fscal,dz00);
 425
 426             /* Update vectorial force */
 427             fix0             = _mm256_add_pd(fix0,tx);
 428             fiy0             = _mm256_add_pd(fiy0,ty);
 429             fiz0             = _mm256_add_pd(fiz0,tz);
 430
 431             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 432             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 433             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 434             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 435             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 436
 437             /* Inner loop uses 74 flops */
 438         }
 439
 440         /* End of innermost loop */
 441
 442         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 443                                                  f+i_coord_offset,fshift+i_shift_offset);
 444
 445         ggid                        = gid[iidx];
 446         /* Update potential energies */
 447         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 448         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 449
 450         /* Increment number of inner iterations */
 451         inneriter                  += j_index_end - j_index_start;
 452
 453         /* Outer loop uses 9 flops */
 454     }
 455
 456     /* Increment number of outer iterations */
 457     outeriter        += nri;
 458
 459     /* Update outer/inner flops */
 460
 461     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*74);
 462 }
 463 /*
 464  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_F_avx_256_double
 465  * Electrostatics interaction: CubicSplineTable
 466  * VdW interaction:            CubicSplineTable
 467  * Geometry:                   Particle-Particle
 468  * Calculate force/pot:        Force
 469  */
 470 void
 471 nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_F_avx_256_double
 472                     (t_nblist                    * gmx_restrict       nlist,
 473                      rvec                        * gmx_restrict          xx,
 474                      rvec                        * gmx_restrict          ff,
 475                      t_forcerec                  * gmx_restrict          fr,
 476                      t_mdatoms                   * gmx_restrict     mdatoms,
 477                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 478                      t_nrnb                      * gmx_restrict        nrnb)
 479 {
 480     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 481      * just 0 for non-waters.
 482      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 483      * jnr indices corresponding to data put in the four positions in the SIMD register.
 484      */
 485     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 486     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 487     int              jnrA,jnrB,jnrC,jnrD;
 488     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 489     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 490     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 491     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 492     real             rcutoff_scalar;
 493     real             *shiftvec,*fshift,*x,*f;
 494     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 495     real             scratch[4*DIM];
 496     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 497     real *           vdwioffsetptr0;
 498     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 499     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 500     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 501     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 502     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 503     real             *charge;
 504     int              nvdwtype;
 505     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 506     int              *vdwtype;
 507     real             *vdwparam;
 508     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 509     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 510     __m128i          vfitab;
 511     __m128i          ifour       = _mm_set1_epi32(4);
 512     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 513     real             *vftab;
 514     __m256d          dummy_mask,cutoff_mask;
 515     __m128           tmpmask0,tmpmask1;
 516     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 517     __m256d          one     = _mm256_set1_pd(1.0);
 518     __m256d          two     = _mm256_set1_pd(2.0);
 519     x                = xx[0];
 520     f                = ff[0];
 521
 522     nri              = nlist->nri;
 523     iinr             = nlist->iinr;
 524     jindex           = nlist->jindex;
 525     jjnr             = nlist->jjnr;
 526     shiftidx         = nlist->shift;
 527     gid              = nlist->gid;
 528     shiftvec         = fr->shift_vec[0];
 529     fshift           = fr->fshift[0];
 530     facel            = _mm256_set1_pd(fr->epsfac);
 531     charge           = mdatoms->chargeA;
 532     nvdwtype         = fr->ntype;
 533     vdwparam         = fr->nbfp;
 534     vdwtype          = mdatoms->typeA;
 535
 536     vftab            = kernel_data->table_elec_vdw->data;
 537     vftabscale       = _mm256_set1_pd(kernel_data->table_elec_vdw->scale);
 538
 539     /* Avoid stupid compiler warnings */
 540     jnrA = jnrB = jnrC = jnrD = 0;
 541     j_coord_offsetA = 0;
 542     j_coord_offsetB = 0;
 543     j_coord_offsetC = 0;
 544     j_coord_offsetD = 0;
 545
 546     outeriter        = 0;
 547     inneriter        = 0;
 548
 549     for(iidx=0;iidx<4*DIM;iidx++)
 550     {
 551         scratch[iidx] = 0.0;
 552     }
 553
 554     /* Start outer loop over neighborlists */
 555     for(iidx=0; iidx<nri; iidx++)
 556     {
 557         /* Load shift vector for this list */
 558         i_shift_offset   = DIM*shiftidx[iidx];
 559
 560         /* Load limits for loop over neighbors */
 561         j_index_start    = jindex[iidx];
 562         j_index_end      = jindex[iidx+1];
 563
 564         /* Get outer coordinate index */
 565         inr              = iinr[iidx];
 566         i_coord_offset   = DIM*inr;
 567
 568         /* Load i particle coords and add shift vector */
 569         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 570
 571         fix0             = _mm256_setzero_pd();
 572         fiy0             = _mm256_setzero_pd();
 573         fiz0             = _mm256_setzero_pd();
 574
 575         /* Load parameters for i particles */
 576         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 577         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 578
 579         /* Start inner kernel loop */
 580         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 581         {
 582
 583             /* Get j neighbor index, and coordinate index */
 584             jnrA             = jjnr[jidx];
 585             jnrB             = jjnr[jidx+1];
 586             jnrC             = jjnr[jidx+2];
 587             jnrD             = jjnr[jidx+3];
 588             j_coord_offsetA  = DIM*jnrA;
 589             j_coord_offsetB  = DIM*jnrB;
 590             j_coord_offsetC  = DIM*jnrC;
 591             j_coord_offsetD  = DIM*jnrD;
 592
 593             /* load j atom coordinates */
 594             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 595                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 596                                                  &jx0,&jy0,&jz0);
 597
 598             /* Calculate displacement vector */
 599             dx00             = _mm256_sub_pd(ix0,jx0);
 600             dy00             = _mm256_sub_pd(iy0,jy0);
 601             dz00             = _mm256_sub_pd(iz0,jz0);
 602
 603             /* Calculate squared distance and things based on it */
 604             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 605
 606             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 607
 608             /* Load parameters for j particles */
 609             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 610                                                                  charge+jnrC+0,charge+jnrD+0);
 611             vdwjidx0A        = 2*vdwtype[jnrA+0];
 612             vdwjidx0B        = 2*vdwtype[jnrB+0];
 613             vdwjidx0C        = 2*vdwtype[jnrC+0];
 614             vdwjidx0D        = 2*vdwtype[jnrD+0];
 615
 616             /**************************
 617              * CALCULATE INTERACTIONS *
 618              **************************/
 619
 620             r00              = _mm256_mul_pd(rsq00,rinv00);
 621
 622             /* Compute parameters for interactions between i and j atoms */
 623             qq00             = _mm256_mul_pd(iq0,jq0);
 624             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 625                                             vdwioffsetptr0+vdwjidx0B,
 626                                             vdwioffsetptr0+vdwjidx0C,
 627                                             vdwioffsetptr0+vdwjidx0D,
 628                                             &c6_00,&c12_00);
 629
 630             /* Calculate table index by multiplying r with table scale and truncate to integer */
 631             rt               = _mm256_mul_pd(r00,vftabscale);
 632             vfitab           = _mm256_cvttpd_epi32(rt);
 633             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 634             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 635
 636             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 637             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 638             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 639             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 640             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 641             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 642             Heps             = _mm256_mul_pd(vfeps,H);
 643             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 644             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 645             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 646
 647             /* CUBIC SPLINE TABLE DISPERSION */
 648             vfitab           = _mm_add_epi32(vfitab,ifour);
 649             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 650             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 651             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 652             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 653             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 654             Heps             = _mm256_mul_pd(vfeps,H);
 655             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 656             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 657             fvdw6            = _mm256_mul_pd(c6_00,FF);
 658
 659             /* CUBIC SPLINE TABLE REPULSION */
 660             vfitab           = _mm_add_epi32(vfitab,ifour);
 661             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 662             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 663             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 664             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 665             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 666             Heps             = _mm256_mul_pd(vfeps,H);
 667             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 668             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 669             fvdw12           = _mm256_mul_pd(c12_00,FF);
 670             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 671
 672             fscal            = _mm256_add_pd(felec,fvdw);
 673
 674             /* Calculate temporary vectorial force */
 675             tx               = _mm256_mul_pd(fscal,dx00);
 676             ty               = _mm256_mul_pd(fscal,dy00);
 677             tz               = _mm256_mul_pd(fscal,dz00);
 678
 679             /* Update vectorial force */
 680             fix0             = _mm256_add_pd(fix0,tx);
 681             fiy0             = _mm256_add_pd(fiy0,ty);
 682             fiz0             = _mm256_add_pd(fiz0,tz);
 683
 684             fjptrA             = f+j_coord_offsetA;
 685             fjptrB             = f+j_coord_offsetB;
 686             fjptrC             = f+j_coord_offsetC;
 687             fjptrD             = f+j_coord_offsetD;
 688             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 689
 690             /* Inner loop uses 61 flops */
 691         }
 692
 693         if(jidx<j_index_end)
 694         {
 695
 696             /* Get j neighbor index, and coordinate index */
 697             jnrlistA         = jjnr[jidx];
 698             jnrlistB         = jjnr[jidx+1];
 699             jnrlistC         = jjnr[jidx+2];
 700             jnrlistD         = jjnr[jidx+3];
 701             /* Sign of each element will be negative for non-real atoms.
 702              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 703              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 704              */
 705             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 706
 707             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 708             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 709             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 710
 711             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 712             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 713             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 714             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 715             j_coord_offsetA  = DIM*jnrA;
 716             j_coord_offsetB  = DIM*jnrB;
 717             j_coord_offsetC  = DIM*jnrC;
 718             j_coord_offsetD  = DIM*jnrD;
 719
 720             /* load j atom coordinates */
 721             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 722                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 723                                                  &jx0,&jy0,&jz0);
 724
 725             /* Calculate displacement vector */
 726             dx00             = _mm256_sub_pd(ix0,jx0);
 727             dy00             = _mm256_sub_pd(iy0,jy0);
 728             dz00             = _mm256_sub_pd(iz0,jz0);
 729
 730             /* Calculate squared distance and things based on it */
 731             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 732
 733             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 734
 735             /* Load parameters for j particles */
 736             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 737                                                                  charge+jnrC+0,charge+jnrD+0);
 738             vdwjidx0A        = 2*vdwtype[jnrA+0];
 739             vdwjidx0B        = 2*vdwtype[jnrB+0];
 740             vdwjidx0C        = 2*vdwtype[jnrC+0];
 741             vdwjidx0D        = 2*vdwtype[jnrD+0];
 742
 743             /**************************
 744              * CALCULATE INTERACTIONS *
 745              **************************/
 746
 747             r00              = _mm256_mul_pd(rsq00,rinv00);
 748             r00              = _mm256_andnot_pd(dummy_mask,r00);
 749
 750             /* Compute parameters for interactions between i and j atoms */
 751             qq00             = _mm256_mul_pd(iq0,jq0);
 752             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 753                                             vdwioffsetptr0+vdwjidx0B,
 754                                             vdwioffsetptr0+vdwjidx0C,
 755                                             vdwioffsetptr0+vdwjidx0D,
 756                                             &c6_00,&c12_00);
 757
 758             /* Calculate table index by multiplying r with table scale and truncate to integer */
 759             rt               = _mm256_mul_pd(r00,vftabscale);
 760             vfitab           = _mm256_cvttpd_epi32(rt);
 761             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 762             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 763
 764             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 765             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 766             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 767             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 768             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 769             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 770             Heps             = _mm256_mul_pd(vfeps,H);
 771             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 772             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 773             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 774
 775             /* CUBIC SPLINE TABLE DISPERSION */
 776             vfitab           = _mm_add_epi32(vfitab,ifour);
 777             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 778             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 779             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 780             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 781             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 782             Heps             = _mm256_mul_pd(vfeps,H);
 783             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 784             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 785             fvdw6            = _mm256_mul_pd(c6_00,FF);
 786
 787             /* CUBIC SPLINE TABLE REPULSION */
 788             vfitab           = _mm_add_epi32(vfitab,ifour);
 789             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 790             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 791             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 792             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 793             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 794             Heps             = _mm256_mul_pd(vfeps,H);
 795             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 796             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 797             fvdw12           = _mm256_mul_pd(c12_00,FF);
 798             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 799
 800             fscal            = _mm256_add_pd(felec,fvdw);
 801
 802             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 803
 804             /* Calculate temporary vectorial force */
 805             tx               = _mm256_mul_pd(fscal,dx00);
 806             ty               = _mm256_mul_pd(fscal,dy00);
 807             tz               = _mm256_mul_pd(fscal,dz00);
 808
 809             /* Update vectorial force */
 810             fix0             = _mm256_add_pd(fix0,tx);
 811             fiy0             = _mm256_add_pd(fiy0,ty);
 812             fiz0             = _mm256_add_pd(fiz0,tz);
 813
 814             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 815             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 816             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 817             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 818             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 819
 820             /* Inner loop uses 62 flops */
 821         }
 822
 823         /* End of innermost loop */
 824
 825         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 826                                                  f+i_coord_offset,fshift+i_shift_offset);
 827
 828         /* Increment number of inner iterations */
 829         inneriter                  += j_index_end - j_index_start;
 830
 831         /* Outer loop uses 7 flops */
 832     }
 833
 834     /* Increment number of outer iterations */
 835     outeriter        += nri;
 836
 837     /* Update outer/inner flops */
 838
 839     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*62);
 840 }