src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEw_VdwLJEw_GeomP1P1_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, 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/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/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_ElecEw_VdwLJEw_GeomP1P1_VF_avx_256_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJEw_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     real *           vdwgridioffsetptr0;
  88     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  93     real             *charge;
  94     int              nvdwtype;
  95     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  96     int              *vdwtype;
  97     real             *vdwparam;
  98     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  99     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 100     __m256d           c6grid_00;
 101     real             *vdwgridparam;
 102     __m256d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 103     __m256d           one_half  = _mm256_set1_pd(0.5);
 104     __m256d           minus_one = _mm256_set1_pd(-1.0);
 105     __m128i          ewitab;
 106     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 107     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 108     real             *ewtab;
 109     __m256d          dummy_mask,cutoff_mask;
 110     __m128           tmpmask0,tmpmask1;
 111     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 112     __m256d          one     = _mm256_set1_pd(1.0);
 113     __m256d          two     = _mm256_set1_pd(2.0);
 114     x                = xx[0];
 115     f                = ff[0];
 116
 117     nri              = nlist->nri;
 118     iinr             = nlist->iinr;
 119     jindex           = nlist->jindex;
 120     jjnr             = nlist->jjnr;
 121     shiftidx         = nlist->shift;
 122     gid              = nlist->gid;
 123     shiftvec         = fr->shift_vec[0];
 124     fshift           = fr->fshift[0];
 125     facel            = _mm256_set1_pd(fr->epsfac);
 126     charge           = mdatoms->chargeA;
 127     nvdwtype         = fr->ntype;
 128     vdwparam         = fr->nbfp;
 129     vdwtype          = mdatoms->typeA;
 130     vdwgridparam     = fr->ljpme_c6grid;
 131     sh_lj_ewald      = _mm256_set1_pd(fr->ic->sh_lj_ewald);
 132     ewclj            = _mm256_set1_pd(fr->ewaldcoeff_lj);
 133     ewclj2           = _mm256_mul_pd(minus_one,_mm256_mul_pd(ewclj,ewclj));
 134
 135     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 136     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 137     beta2            = _mm256_mul_pd(beta,beta);
 138     beta3            = _mm256_mul_pd(beta,beta2);
 139
 140     ewtab            = fr->ic->tabq_coul_FDV0;
 141     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 142     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 143
 144     /* Avoid stupid compiler warnings */
 145     jnrA = jnrB = jnrC = jnrD = 0;
 146     j_coord_offsetA = 0;
 147     j_coord_offsetB = 0;
 148     j_coord_offsetC = 0;
 149     j_coord_offsetD = 0;
 150
 151     outeriter        = 0;
 152     inneriter        = 0;
 153
 154     for(iidx=0;iidx<4*DIM;iidx++)
 155     {
 156         scratch[iidx] = 0.0;
 157     }
 158
 159     /* Start outer loop over neighborlists */
 160     for(iidx=0; iidx<nri; iidx++)
 161     {
 162         /* Load shift vector for this list */
 163         i_shift_offset   = DIM*shiftidx[iidx];
 164
 165         /* Load limits for loop over neighbors */
 166         j_index_start    = jindex[iidx];
 167         j_index_end      = jindex[iidx+1];
 168
 169         /* Get outer coordinate index */
 170         inr              = iinr[iidx];
 171         i_coord_offset   = DIM*inr;
 172
 173         /* Load i particle coords and add shift vector */
 174         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 175
 176         fix0             = _mm256_setzero_pd();
 177         fiy0             = _mm256_setzero_pd();
 178         fiz0             = _mm256_setzero_pd();
 179
 180         /* Load parameters for i particles */
 181         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 182         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 183         vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 184
 185         /* Reset potential sums */
 186         velecsum         = _mm256_setzero_pd();
 187         vvdwsum          = _mm256_setzero_pd();
 188
 189         /* Start inner kernel loop */
 190         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 191         {
 192
 193             /* Get j neighbor index, and coordinate index */
 194             jnrA             = jjnr[jidx];
 195             jnrB             = jjnr[jidx+1];
 196             jnrC             = jjnr[jidx+2];
 197             jnrD             = jjnr[jidx+3];
 198             j_coord_offsetA  = DIM*jnrA;
 199             j_coord_offsetB  = DIM*jnrB;
 200             j_coord_offsetC  = DIM*jnrC;
 201             j_coord_offsetD  = DIM*jnrD;
 202
 203             /* load j atom coordinates */
 204             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 205                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 206                                                  &jx0,&jy0,&jz0);
 207
 208             /* Calculate displacement vector */
 209             dx00             = _mm256_sub_pd(ix0,jx0);
 210             dy00             = _mm256_sub_pd(iy0,jy0);
 211             dz00             = _mm256_sub_pd(iz0,jz0);
 212
 213             /* Calculate squared distance and things based on it */
 214             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 215
 216             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 217
 218             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 219
 220             /* Load parameters for j particles */
 221             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 222                                                                  charge+jnrC+0,charge+jnrD+0);
 223             vdwjidx0A        = 2*vdwtype[jnrA+0];
 224             vdwjidx0B        = 2*vdwtype[jnrB+0];
 225             vdwjidx0C        = 2*vdwtype[jnrC+0];
 226             vdwjidx0D        = 2*vdwtype[jnrD+0];
 227
 228             /**************************
 229              * CALCULATE INTERACTIONS *
 230              **************************/
 231
 232             r00              = _mm256_mul_pd(rsq00,rinv00);
 233
 234             /* Compute parameters for interactions between i and j atoms */
 235             qq00             = _mm256_mul_pd(iq0,jq0);
 236             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 237                                             vdwioffsetptr0+vdwjidx0B,
 238                                             vdwioffsetptr0+vdwjidx0C,
 239                                             vdwioffsetptr0+vdwjidx0D,
 240                                             &c6_00,&c12_00);
 241
 242             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 243                                                                   vdwgridioffsetptr0+vdwjidx0B,
 244                                                                   vdwgridioffsetptr0+vdwjidx0C,
 245                                                                   vdwgridioffsetptr0+vdwjidx0D);
 246
 247             /* EWALD ELECTROSTATICS */
 248
 249             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 250             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 251             ewitab           = _mm256_cvttpd_epi32(ewrt);
 252             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 253             ewitab           = _mm_slli_epi32(ewitab,2);
 254             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 255             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 256             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 257             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 258             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 259             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 260             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 261             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 262             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 263
 264             /* Analytical LJ-PME */
 265             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 266             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 267             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 268             exponent         = gmx_simd_exp_d(ewcljrsq);
 269             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 270             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 271             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 272             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
 273             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 274             vvdw             = _mm256_sub_pd(_mm256_mul_pd(vvdw12,one_twelfth),_mm256_mul_pd(vvdw6,one_sixth));
 275             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 276             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,_mm256_sub_pd(vvdw6,_mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6)))),rinvsq00);
 277
 278             /* Update potential sum for this i atom from the interaction with this j atom. */
 279             velecsum         = _mm256_add_pd(velecsum,velec);
 280             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 281
 282             fscal            = _mm256_add_pd(felec,fvdw);
 283
 284             /* Calculate temporary vectorial force */
 285             tx               = _mm256_mul_pd(fscal,dx00);
 286             ty               = _mm256_mul_pd(fscal,dy00);
 287             tz               = _mm256_mul_pd(fscal,dz00);
 288
 289             /* Update vectorial force */
 290             fix0             = _mm256_add_pd(fix0,tx);
 291             fiy0             = _mm256_add_pd(fiy0,ty);
 292             fiz0             = _mm256_add_pd(fiz0,tz);
 293
 294             fjptrA             = f+j_coord_offsetA;
 295             fjptrB             = f+j_coord_offsetB;
 296             fjptrC             = f+j_coord_offsetC;
 297             fjptrD             = f+j_coord_offsetD;
 298             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 299
 300             /* Inner loop uses 72 flops */
 301         }
 302
 303         if(jidx<j_index_end)
 304         {
 305
 306             /* Get j neighbor index, and coordinate index */
 307             jnrlistA         = jjnr[jidx];
 308             jnrlistB         = jjnr[jidx+1];
 309             jnrlistC         = jjnr[jidx+2];
 310             jnrlistD         = jjnr[jidx+3];
 311             /* Sign of each element will be negative for non-real atoms.
 312              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 313              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 314              */
 315             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 316
 317             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 318             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 319             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 320
 321             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 322             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 323             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 324             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 325             j_coord_offsetA  = DIM*jnrA;
 326             j_coord_offsetB  = DIM*jnrB;
 327             j_coord_offsetC  = DIM*jnrC;
 328             j_coord_offsetD  = DIM*jnrD;
 329
 330             /* load j atom coordinates */
 331             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 332                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 333                                                  &jx0,&jy0,&jz0);
 334
 335             /* Calculate displacement vector */
 336             dx00             = _mm256_sub_pd(ix0,jx0);
 337             dy00             = _mm256_sub_pd(iy0,jy0);
 338             dz00             = _mm256_sub_pd(iz0,jz0);
 339
 340             /* Calculate squared distance and things based on it */
 341             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 342
 343             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 344
 345             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 346
 347             /* Load parameters for j particles */
 348             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 349                                                                  charge+jnrC+0,charge+jnrD+0);
 350             vdwjidx0A        = 2*vdwtype[jnrA+0];
 351             vdwjidx0B        = 2*vdwtype[jnrB+0];
 352             vdwjidx0C        = 2*vdwtype[jnrC+0];
 353             vdwjidx0D        = 2*vdwtype[jnrD+0];
 354
 355             /**************************
 356              * CALCULATE INTERACTIONS *
 357              **************************/
 358
 359             r00              = _mm256_mul_pd(rsq00,rinv00);
 360             r00              = _mm256_andnot_pd(dummy_mask,r00);
 361
 362             /* Compute parameters for interactions between i and j atoms */
 363             qq00             = _mm256_mul_pd(iq0,jq0);
 364             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 365                                             vdwioffsetptr0+vdwjidx0B,
 366                                             vdwioffsetptr0+vdwjidx0C,
 367                                             vdwioffsetptr0+vdwjidx0D,
 368                                             &c6_00,&c12_00);
 369
 370             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 371                                                                   vdwgridioffsetptr0+vdwjidx0B,
 372                                                                   vdwgridioffsetptr0+vdwjidx0C,
 373                                                                   vdwgridioffsetptr0+vdwjidx0D);
 374
 375             /* EWALD ELECTROSTATICS */
 376
 377             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 378             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 379             ewitab           = _mm256_cvttpd_epi32(ewrt);
 380             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 381             ewitab           = _mm_slli_epi32(ewitab,2);
 382             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 383             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 384             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 385             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 386             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 387             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 388             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 389             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 390             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 391
 392             /* Analytical LJ-PME */
 393             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 394             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 395             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 396             exponent         = gmx_simd_exp_d(ewcljrsq);
 397             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 398             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 399             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 400             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
 401             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 402             vvdw             = _mm256_sub_pd(_mm256_mul_pd(vvdw12,one_twelfth),_mm256_mul_pd(vvdw6,one_sixth));
 403             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 404             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,_mm256_sub_pd(vvdw6,_mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6)))),rinvsq00);
 405
 406             /* Update potential sum for this i atom from the interaction with this j atom. */
 407             velec            = _mm256_andnot_pd(dummy_mask,velec);
 408             velecsum         = _mm256_add_pd(velecsum,velec);
 409             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 410             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 411
 412             fscal            = _mm256_add_pd(felec,fvdw);
 413
 414             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 415
 416             /* Calculate temporary vectorial force */
 417             tx               = _mm256_mul_pd(fscal,dx00);
 418             ty               = _mm256_mul_pd(fscal,dy00);
 419             tz               = _mm256_mul_pd(fscal,dz00);
 420
 421             /* Update vectorial force */
 422             fix0             = _mm256_add_pd(fix0,tx);
 423             fiy0             = _mm256_add_pd(fiy0,ty);
 424             fiz0             = _mm256_add_pd(fiz0,tz);
 425
 426             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 427             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 428             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 429             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 430             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 431
 432             /* Inner loop uses 73 flops */
 433         }
 434
 435         /* End of innermost loop */
 436
 437         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 438                                                  f+i_coord_offset,fshift+i_shift_offset);
 439
 440         ggid                        = gid[iidx];
 441         /* Update potential energies */
 442         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 443         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 444
 445         /* Increment number of inner iterations */
 446         inneriter                  += j_index_end - j_index_start;
 447
 448         /* Outer loop uses 9 flops */
 449     }
 450
 451     /* Increment number of outer iterations */
 452     outeriter        += nri;
 453
 454     /* Update outer/inner flops */
 455
 456     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*73);
 457 }
 458 /*
 459  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomP1P1_F_avx_256_double
 460  * Electrostatics interaction: Ewald
 461  * VdW interaction:            LJEwald
 462  * Geometry:                   Particle-Particle
 463  * Calculate force/pot:        Force
 464  */
 465 void
 466 nb_kernel_ElecEw_VdwLJEw_GeomP1P1_F_avx_256_double
 467                     (t_nblist                    * gmx_restrict       nlist,
 468                      rvec                        * gmx_restrict          xx,
 469                      rvec                        * gmx_restrict          ff,
 470                      t_forcerec                  * gmx_restrict          fr,
 471                      t_mdatoms                   * gmx_restrict     mdatoms,
 472                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 473                      t_nrnb                      * gmx_restrict        nrnb)
 474 {
 475     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 476      * just 0 for non-waters.
 477      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 478      * jnr indices corresponding to data put in the four positions in the SIMD register.
 479      */
 480     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 481     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 482     int              jnrA,jnrB,jnrC,jnrD;
 483     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 484     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 485     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 486     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 487     real             rcutoff_scalar;
 488     real             *shiftvec,*fshift,*x,*f;
 489     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 490     real             scratch[4*DIM];
 491     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 492     real *           vdwioffsetptr0;
 493     real *           vdwgridioffsetptr0;
 494     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 495     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 496     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 497     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 498     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 499     real             *charge;
 500     int              nvdwtype;
 501     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 502     int              *vdwtype;
 503     real             *vdwparam;
 504     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 505     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 506     __m256d           c6grid_00;
 507     real             *vdwgridparam;
 508     __m256d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 509     __m256d           one_half  = _mm256_set1_pd(0.5);
 510     __m256d           minus_one = _mm256_set1_pd(-1.0);
 511     __m128i          ewitab;
 512     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 513     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 514     real             *ewtab;
 515     __m256d          dummy_mask,cutoff_mask;
 516     __m128           tmpmask0,tmpmask1;
 517     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 518     __m256d          one     = _mm256_set1_pd(1.0);
 519     __m256d          two     = _mm256_set1_pd(2.0);
 520     x                = xx[0];
 521     f                = ff[0];
 522
 523     nri              = nlist->nri;
 524     iinr             = nlist->iinr;
 525     jindex           = nlist->jindex;
 526     jjnr             = nlist->jjnr;
 527     shiftidx         = nlist->shift;
 528     gid              = nlist->gid;
 529     shiftvec         = fr->shift_vec[0];
 530     fshift           = fr->fshift[0];
 531     facel            = _mm256_set1_pd(fr->epsfac);
 532     charge           = mdatoms->chargeA;
 533     nvdwtype         = fr->ntype;
 534     vdwparam         = fr->nbfp;
 535     vdwtype          = mdatoms->typeA;
 536     vdwgridparam     = fr->ljpme_c6grid;
 537     sh_lj_ewald      = _mm256_set1_pd(fr->ic->sh_lj_ewald);
 538     ewclj            = _mm256_set1_pd(fr->ewaldcoeff_lj);
 539     ewclj2           = _mm256_mul_pd(minus_one,_mm256_mul_pd(ewclj,ewclj));
 540
 541     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 542     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 543     beta2            = _mm256_mul_pd(beta,beta);
 544     beta3            = _mm256_mul_pd(beta,beta2);
 545
 546     ewtab            = fr->ic->tabq_coul_F;
 547     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 548     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 549
 550     /* Avoid stupid compiler warnings */
 551     jnrA = jnrB = jnrC = jnrD = 0;
 552     j_coord_offsetA = 0;
 553     j_coord_offsetB = 0;
 554     j_coord_offsetC = 0;
 555     j_coord_offsetD = 0;
 556
 557     outeriter        = 0;
 558     inneriter        = 0;
 559
 560     for(iidx=0;iidx<4*DIM;iidx++)
 561     {
 562         scratch[iidx] = 0.0;
 563     }
 564
 565     /* Start outer loop over neighborlists */
 566     for(iidx=0; iidx<nri; iidx++)
 567     {
 568         /* Load shift vector for this list */
 569         i_shift_offset   = DIM*shiftidx[iidx];
 570
 571         /* Load limits for loop over neighbors */
 572         j_index_start    = jindex[iidx];
 573         j_index_end      = jindex[iidx+1];
 574
 575         /* Get outer coordinate index */
 576         inr              = iinr[iidx];
 577         i_coord_offset   = DIM*inr;
 578
 579         /* Load i particle coords and add shift vector */
 580         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 581
 582         fix0             = _mm256_setzero_pd();
 583         fiy0             = _mm256_setzero_pd();
 584         fiz0             = _mm256_setzero_pd();
 585
 586         /* Load parameters for i particles */
 587         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 588         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 589         vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 590
 591         /* Start inner kernel loop */
 592         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 593         {
 594
 595             /* Get j neighbor index, and coordinate index */
 596             jnrA             = jjnr[jidx];
 597             jnrB             = jjnr[jidx+1];
 598             jnrC             = jjnr[jidx+2];
 599             jnrD             = jjnr[jidx+3];
 600             j_coord_offsetA  = DIM*jnrA;
 601             j_coord_offsetB  = DIM*jnrB;
 602             j_coord_offsetC  = DIM*jnrC;
 603             j_coord_offsetD  = DIM*jnrD;
 604
 605             /* load j atom coordinates */
 606             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 607                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 608                                                  &jx0,&jy0,&jz0);
 609
 610             /* Calculate displacement vector */
 611             dx00             = _mm256_sub_pd(ix0,jx0);
 612             dy00             = _mm256_sub_pd(iy0,jy0);
 613             dz00             = _mm256_sub_pd(iz0,jz0);
 614
 615             /* Calculate squared distance and things based on it */
 616             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 617
 618             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 619
 620             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 621
 622             /* Load parameters for j particles */
 623             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 624                                                                  charge+jnrC+0,charge+jnrD+0);
 625             vdwjidx0A        = 2*vdwtype[jnrA+0];
 626             vdwjidx0B        = 2*vdwtype[jnrB+0];
 627             vdwjidx0C        = 2*vdwtype[jnrC+0];
 628             vdwjidx0D        = 2*vdwtype[jnrD+0];
 629
 630             /**************************
 631              * CALCULATE INTERACTIONS *
 632              **************************/
 633
 634             r00              = _mm256_mul_pd(rsq00,rinv00);
 635
 636             /* Compute parameters for interactions between i and j atoms */
 637             qq00             = _mm256_mul_pd(iq0,jq0);
 638             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 639                                             vdwioffsetptr0+vdwjidx0B,
 640                                             vdwioffsetptr0+vdwjidx0C,
 641                                             vdwioffsetptr0+vdwjidx0D,
 642                                             &c6_00,&c12_00);
 643
 644             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 645                                                                   vdwgridioffsetptr0+vdwjidx0B,
 646                                                                   vdwgridioffsetptr0+vdwjidx0C,
 647                                                                   vdwgridioffsetptr0+vdwjidx0D);
 648
 649             /* EWALD ELECTROSTATICS */
 650
 651             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 652             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 653             ewitab           = _mm256_cvttpd_epi32(ewrt);
 654             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 655             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 656                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 657                                             &ewtabF,&ewtabFn);
 658             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 659             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 660
 661             /* Analytical LJ-PME */
 662             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 663             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 664             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 665             exponent         = gmx_simd_exp_d(ewcljrsq);
 666             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 667             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 668             /* f6A = 6 * C6grid * (1 - poly) */
 669             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
 670             /* f6B = C6grid * exponent * beta^6 */
 671             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
 672             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 673             fvdw              = _mm256_mul_pd(_mm256_add_pd(_mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),_mm256_sub_pd(c6_00,f6A)),rinvsix),f6B),rinvsq00);
 674
 675             fscal            = _mm256_add_pd(felec,fvdw);
 676
 677             /* Calculate temporary vectorial force */
 678             tx               = _mm256_mul_pd(fscal,dx00);
 679             ty               = _mm256_mul_pd(fscal,dy00);
 680             tz               = _mm256_mul_pd(fscal,dz00);
 681
 682             /* Update vectorial force */
 683             fix0             = _mm256_add_pd(fix0,tx);
 684             fiy0             = _mm256_add_pd(fiy0,ty);
 685             fiz0             = _mm256_add_pd(fiz0,tz);
 686
 687             fjptrA             = f+j_coord_offsetA;
 688             fjptrB             = f+j_coord_offsetB;
 689             fjptrC             = f+j_coord_offsetC;
 690             fjptrD             = f+j_coord_offsetD;
 691             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 692
 693             /* Inner loop uses 59 flops */
 694         }
 695
 696         if(jidx<j_index_end)
 697         {
 698
 699             /* Get j neighbor index, and coordinate index */
 700             jnrlistA         = jjnr[jidx];
 701             jnrlistB         = jjnr[jidx+1];
 702             jnrlistC         = jjnr[jidx+2];
 703             jnrlistD         = jjnr[jidx+3];
 704             /* Sign of each element will be negative for non-real atoms.
 705              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 706              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 707              */
 708             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 709
 710             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 711             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 712             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 713
 714             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 715             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 716             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 717             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 718             j_coord_offsetA  = DIM*jnrA;
 719             j_coord_offsetB  = DIM*jnrB;
 720             j_coord_offsetC  = DIM*jnrC;
 721             j_coord_offsetD  = DIM*jnrD;
 722
 723             /* load j atom coordinates */
 724             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 725                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 726                                                  &jx0,&jy0,&jz0);
 727
 728             /* Calculate displacement vector */
 729             dx00             = _mm256_sub_pd(ix0,jx0);
 730             dy00             = _mm256_sub_pd(iy0,jy0);
 731             dz00             = _mm256_sub_pd(iz0,jz0);
 732
 733             /* Calculate squared distance and things based on it */
 734             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 735
 736             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 737
 738             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 739
 740             /* Load parameters for j particles */
 741             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 742                                                                  charge+jnrC+0,charge+jnrD+0);
 743             vdwjidx0A        = 2*vdwtype[jnrA+0];
 744             vdwjidx0B        = 2*vdwtype[jnrB+0];
 745             vdwjidx0C        = 2*vdwtype[jnrC+0];
 746             vdwjidx0D        = 2*vdwtype[jnrD+0];
 747
 748             /**************************
 749              * CALCULATE INTERACTIONS *
 750              **************************/
 751
 752             r00              = _mm256_mul_pd(rsq00,rinv00);
 753             r00              = _mm256_andnot_pd(dummy_mask,r00);
 754
 755             /* Compute parameters for interactions between i and j atoms */
 756             qq00             = _mm256_mul_pd(iq0,jq0);
 757             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 758                                             vdwioffsetptr0+vdwjidx0B,
 759                                             vdwioffsetptr0+vdwjidx0C,
 760                                             vdwioffsetptr0+vdwjidx0D,
 761                                             &c6_00,&c12_00);
 762
 763             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 764                                                                   vdwgridioffsetptr0+vdwjidx0B,
 765                                                                   vdwgridioffsetptr0+vdwjidx0C,
 766                                                                   vdwgridioffsetptr0+vdwjidx0D);
 767
 768             /* EWALD ELECTROSTATICS */
 769
 770             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 771             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 772             ewitab           = _mm256_cvttpd_epi32(ewrt);
 773             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 774             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 775                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 776                                             &ewtabF,&ewtabFn);
 777             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 778             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 779
 780             /* Analytical LJ-PME */
 781             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 782             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 783             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 784             exponent         = gmx_simd_exp_d(ewcljrsq);
 785             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 786             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 787             /* f6A = 6 * C6grid * (1 - poly) */
 788             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
 789             /* f6B = C6grid * exponent * beta^6 */
 790             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
 791             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 792             fvdw              = _mm256_mul_pd(_mm256_add_pd(_mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),_mm256_sub_pd(c6_00,f6A)),rinvsix),f6B),rinvsq00);
 793
 794             fscal            = _mm256_add_pd(felec,fvdw);
 795
 796             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 797
 798             /* Calculate temporary vectorial force */
 799             tx               = _mm256_mul_pd(fscal,dx00);
 800             ty               = _mm256_mul_pd(fscal,dy00);
 801             tz               = _mm256_mul_pd(fscal,dz00);
 802
 803             /* Update vectorial force */
 804             fix0             = _mm256_add_pd(fix0,tx);
 805             fiy0             = _mm256_add_pd(fiy0,ty);
 806             fiz0             = _mm256_add_pd(fiz0,tz);
 807
 808             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 809             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 810             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 811             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 812             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 813
 814             /* Inner loop uses 60 flops */
 815         }
 816
 817         /* End of innermost loop */
 818
 819         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 820                                                  f+i_coord_offset,fshift+i_shift_offset);
 821
 822         /* Increment number of inner iterations */
 823         inneriter                  += j_index_end - j_index_start;
 824
 825         /* Outer loop uses 7 flops */
 826     }
 827
 828     /* Increment number of outer iterations */
 829     outeriter        += nri;
 830
 831     /* Update outer/inner flops */
 832
 833     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*60);
 834 }