src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_avx_256_single.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_single 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_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_VF_avx_256_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_VF_avx_256_single
  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,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight 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              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  91     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 100     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 101     __m256i          ewitab;
 102     __m128i          ewitab_lo,ewitab_hi;
 103     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 104     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 105     real             *ewtab;
 106     __m256           dummy_mask,cutoff_mask;
 107     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 108     __m256           one     = _mm256_set1_ps(1.0);
 109     __m256           two     = _mm256_set1_ps(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm256_set1_ps(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     nvdwtype         = fr->ntype;
 124     vdwparam         = fr->nbfp;
 125     vdwtype          = mdatoms->typeA;
 126
 127     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 128     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 129     beta2            = _mm256_mul_ps(beta,beta);
 130     beta3            = _mm256_mul_ps(beta,beta2);
 131
 132     ewtab            = fr->ic->tabq_coul_FDV0;
 133     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 134     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 135
 136     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 137     rcutoff_scalar   = fr->rcoulomb;
 138     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 139     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 140
 141     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 142     rvdw             = _mm256_set1_ps(fr->rvdw);
 143
 144     /* Avoid stupid compiler warnings */
 145     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 146     j_coord_offsetA = 0;
 147     j_coord_offsetB = 0;
 148     j_coord_offsetC = 0;
 149     j_coord_offsetD = 0;
 150     j_coord_offsetE = 0;
 151     j_coord_offsetF = 0;
 152     j_coord_offsetG = 0;
 153     j_coord_offsetH = 0;
 154
 155     outeriter        = 0;
 156     inneriter        = 0;
 157
 158     for(iidx=0;iidx<4*DIM;iidx++)
 159     {
 160         scratch[iidx] = 0.0;
 161     }
 162
 163     /* Start outer loop over neighborlists */
 164     for(iidx=0; iidx<nri; iidx++)
 165     {
 166         /* Load shift vector for this list */
 167         i_shift_offset   = DIM*shiftidx[iidx];
 168
 169         /* Load limits for loop over neighbors */
 170         j_index_start    = jindex[iidx];
 171         j_index_end      = jindex[iidx+1];
 172
 173         /* Get outer coordinate index */
 174         inr              = iinr[iidx];
 175         i_coord_offset   = DIM*inr;
 176
 177         /* Load i particle coords and add shift vector */
 178         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 179
 180         fix0             = _mm256_setzero_ps();
 181         fiy0             = _mm256_setzero_ps();
 182         fiz0             = _mm256_setzero_ps();
 183
 184         /* Load parameters for i particles */
 185         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 186         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 187
 188         /* Reset potential sums */
 189         velecsum         = _mm256_setzero_ps();
 190         vvdwsum          = _mm256_setzero_ps();
 191
 192         /* Start inner kernel loop */
 193         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 194         {
 195
 196             /* Get j neighbor index, and coordinate index */
 197             jnrA             = jjnr[jidx];
 198             jnrB             = jjnr[jidx+1];
 199             jnrC             = jjnr[jidx+2];
 200             jnrD             = jjnr[jidx+3];
 201             jnrE             = jjnr[jidx+4];
 202             jnrF             = jjnr[jidx+5];
 203             jnrG             = jjnr[jidx+6];
 204             jnrH             = jjnr[jidx+7];
 205             j_coord_offsetA  = DIM*jnrA;
 206             j_coord_offsetB  = DIM*jnrB;
 207             j_coord_offsetC  = DIM*jnrC;
 208             j_coord_offsetD  = DIM*jnrD;
 209             j_coord_offsetE  = DIM*jnrE;
 210             j_coord_offsetF  = DIM*jnrF;
 211             j_coord_offsetG  = DIM*jnrG;
 212             j_coord_offsetH  = DIM*jnrH;
 213
 214             /* load j atom coordinates */
 215             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 216                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 217                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 218                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 219                                                  &jx0,&jy0,&jz0);
 220
 221             /* Calculate displacement vector */
 222             dx00             = _mm256_sub_ps(ix0,jx0);
 223             dy00             = _mm256_sub_ps(iy0,jy0);
 224             dz00             = _mm256_sub_ps(iz0,jz0);
 225
 226             /* Calculate squared distance and things based on it */
 227             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 228
 229             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 230
 231             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 232
 233             /* Load parameters for j particles */
 234             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 235                                                                  charge+jnrC+0,charge+jnrD+0,
 236                                                                  charge+jnrE+0,charge+jnrF+0,
 237                                                                  charge+jnrG+0,charge+jnrH+0);
 238             vdwjidx0A        = 2*vdwtype[jnrA+0];
 239             vdwjidx0B        = 2*vdwtype[jnrB+0];
 240             vdwjidx0C        = 2*vdwtype[jnrC+0];
 241             vdwjidx0D        = 2*vdwtype[jnrD+0];
 242             vdwjidx0E        = 2*vdwtype[jnrE+0];
 243             vdwjidx0F        = 2*vdwtype[jnrF+0];
 244             vdwjidx0G        = 2*vdwtype[jnrG+0];
 245             vdwjidx0H        = 2*vdwtype[jnrH+0];
 246
 247             /**************************
 248              * CALCULATE INTERACTIONS *
 249              **************************/
 250
 251             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 252             {
 253
 254             r00              = _mm256_mul_ps(rsq00,rinv00);
 255
 256             /* Compute parameters for interactions between i and j atoms */
 257             qq00             = _mm256_mul_ps(iq0,jq0);
 258             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 259                                             vdwioffsetptr0+vdwjidx0B,
 260                                             vdwioffsetptr0+vdwjidx0C,
 261                                             vdwioffsetptr0+vdwjidx0D,
 262                                             vdwioffsetptr0+vdwjidx0E,
 263                                             vdwioffsetptr0+vdwjidx0F,
 264                                             vdwioffsetptr0+vdwjidx0G,
 265                                             vdwioffsetptr0+vdwjidx0H,
 266                                             &c6_00,&c12_00);
 267
 268             /* EWALD ELECTROSTATICS */
 269
 270             /* Analytical PME correction */
 271             zeta2            = _mm256_mul_ps(beta2,rsq00);
 272             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 273             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 274             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 275             felec            = _mm256_mul_ps(qq00,felec);
 276             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 277             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 278             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 279             velec            = _mm256_mul_ps(qq00,velec);
 280
 281             /* LENNARD-JONES DISPERSION/REPULSION */
 282
 283             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 284             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 285             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 286             vvdw             = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_00,_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 287                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 288             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 289
 290             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 291
 292             /* Update potential sum for this i atom from the interaction with this j atom. */
 293             velec            = _mm256_and_ps(velec,cutoff_mask);
 294             velecsum         = _mm256_add_ps(velecsum,velec);
 295             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 296             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 297
 298             fscal            = _mm256_add_ps(felec,fvdw);
 299
 300             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 301
 302             /* Calculate temporary vectorial force */
 303             tx               = _mm256_mul_ps(fscal,dx00);
 304             ty               = _mm256_mul_ps(fscal,dy00);
 305             tz               = _mm256_mul_ps(fscal,dz00);
 306
 307             /* Update vectorial force */
 308             fix0             = _mm256_add_ps(fix0,tx);
 309             fiy0             = _mm256_add_ps(fiy0,ty);
 310             fiz0             = _mm256_add_ps(fiz0,tz);
 311
 312             fjptrA             = f+j_coord_offsetA;
 313             fjptrB             = f+j_coord_offsetB;
 314             fjptrC             = f+j_coord_offsetC;
 315             fjptrD             = f+j_coord_offsetD;
 316             fjptrE             = f+j_coord_offsetE;
 317             fjptrF             = f+j_coord_offsetF;
 318             fjptrG             = f+j_coord_offsetG;
 319             fjptrH             = f+j_coord_offsetH;
 320             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 321
 322             }
 323
 324             /* Inner loop uses 127 flops */
 325         }
 326
 327         if(jidx<j_index_end)
 328         {
 329
 330             /* Get j neighbor index, and coordinate index */
 331             jnrlistA         = jjnr[jidx];
 332             jnrlistB         = jjnr[jidx+1];
 333             jnrlistC         = jjnr[jidx+2];
 334             jnrlistD         = jjnr[jidx+3];
 335             jnrlistE         = jjnr[jidx+4];
 336             jnrlistF         = jjnr[jidx+5];
 337             jnrlistG         = jjnr[jidx+6];
 338             jnrlistH         = jjnr[jidx+7];
 339             /* Sign of each element will be negative for non-real atoms.
 340              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 341              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 342              */
 343             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 344                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 345
 346             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 347             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 348             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 349             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 350             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 351             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 352             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 353             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 354             j_coord_offsetA  = DIM*jnrA;
 355             j_coord_offsetB  = DIM*jnrB;
 356             j_coord_offsetC  = DIM*jnrC;
 357             j_coord_offsetD  = DIM*jnrD;
 358             j_coord_offsetE  = DIM*jnrE;
 359             j_coord_offsetF  = DIM*jnrF;
 360             j_coord_offsetG  = DIM*jnrG;
 361             j_coord_offsetH  = DIM*jnrH;
 362
 363             /* load j atom coordinates */
 364             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 365                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 366                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 367                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 368                                                  &jx0,&jy0,&jz0);
 369
 370             /* Calculate displacement vector */
 371             dx00             = _mm256_sub_ps(ix0,jx0);
 372             dy00             = _mm256_sub_ps(iy0,jy0);
 373             dz00             = _mm256_sub_ps(iz0,jz0);
 374
 375             /* Calculate squared distance and things based on it */
 376             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 377
 378             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 379
 380             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 381
 382             /* Load parameters for j particles */
 383             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 384                                                                  charge+jnrC+0,charge+jnrD+0,
 385                                                                  charge+jnrE+0,charge+jnrF+0,
 386                                                                  charge+jnrG+0,charge+jnrH+0);
 387             vdwjidx0A        = 2*vdwtype[jnrA+0];
 388             vdwjidx0B        = 2*vdwtype[jnrB+0];
 389             vdwjidx0C        = 2*vdwtype[jnrC+0];
 390             vdwjidx0D        = 2*vdwtype[jnrD+0];
 391             vdwjidx0E        = 2*vdwtype[jnrE+0];
 392             vdwjidx0F        = 2*vdwtype[jnrF+0];
 393             vdwjidx0G        = 2*vdwtype[jnrG+0];
 394             vdwjidx0H        = 2*vdwtype[jnrH+0];
 395
 396             /**************************
 397              * CALCULATE INTERACTIONS *
 398              **************************/
 399
 400             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 401             {
 402
 403             r00              = _mm256_mul_ps(rsq00,rinv00);
 404             r00              = _mm256_andnot_ps(dummy_mask,r00);
 405
 406             /* Compute parameters for interactions between i and j atoms */
 407             qq00             = _mm256_mul_ps(iq0,jq0);
 408             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 409                                             vdwioffsetptr0+vdwjidx0B,
 410                                             vdwioffsetptr0+vdwjidx0C,
 411                                             vdwioffsetptr0+vdwjidx0D,
 412                                             vdwioffsetptr0+vdwjidx0E,
 413                                             vdwioffsetptr0+vdwjidx0F,
 414                                             vdwioffsetptr0+vdwjidx0G,
 415                                             vdwioffsetptr0+vdwjidx0H,
 416                                             &c6_00,&c12_00);
 417
 418             /* EWALD ELECTROSTATICS */
 419
 420             /* Analytical PME correction */
 421             zeta2            = _mm256_mul_ps(beta2,rsq00);
 422             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 423             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 424             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 425             felec            = _mm256_mul_ps(qq00,felec);
 426             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 427             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 428             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 429             velec            = _mm256_mul_ps(qq00,velec);
 430
 431             /* LENNARD-JONES DISPERSION/REPULSION */
 432
 433             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 434             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 435             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 436             vvdw             = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_00,_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 437                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 438             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 439
 440             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 441
 442             /* Update potential sum for this i atom from the interaction with this j atom. */
 443             velec            = _mm256_and_ps(velec,cutoff_mask);
 444             velec            = _mm256_andnot_ps(dummy_mask,velec);
 445             velecsum         = _mm256_add_ps(velecsum,velec);
 446             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 447             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 448             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 449
 450             fscal            = _mm256_add_ps(felec,fvdw);
 451
 452             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 453
 454             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 455
 456             /* Calculate temporary vectorial force */
 457             tx               = _mm256_mul_ps(fscal,dx00);
 458             ty               = _mm256_mul_ps(fscal,dy00);
 459             tz               = _mm256_mul_ps(fscal,dz00);
 460
 461             /* Update vectorial force */
 462             fix0             = _mm256_add_ps(fix0,tx);
 463             fiy0             = _mm256_add_ps(fiy0,ty);
 464             fiz0             = _mm256_add_ps(fiz0,tz);
 465
 466             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 467             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 468             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 469             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 470             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 471             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 472             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 473             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 474             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 475
 476             }
 477
 478             /* Inner loop uses 128 flops */
 479         }
 480
 481         /* End of innermost loop */
 482
 483         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 484                                                  f+i_coord_offset,fshift+i_shift_offset);
 485
 486         ggid                        = gid[iidx];
 487         /* Update potential energies */
 488         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 489         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 490
 491         /* Increment number of inner iterations */
 492         inneriter                  += j_index_end - j_index_start;
 493
 494         /* Outer loop uses 9 flops */
 495     }
 496
 497     /* Increment number of outer iterations */
 498     outeriter        += nri;
 499
 500     /* Update outer/inner flops */
 501
 502     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*128);
 503 }
 504 /*
 505  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_avx_256_single
 506  * Electrostatics interaction: Ewald
 507  * VdW interaction:            LennardJones
 508  * Geometry:                   Particle-Particle
 509  * Calculate force/pot:        Force
 510  */
 511 void
 512 nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_avx_256_single
 513                     (t_nblist                    * gmx_restrict       nlist,
 514                      rvec                        * gmx_restrict          xx,
 515                      rvec                        * gmx_restrict          ff,
 516                      t_forcerec                  * gmx_restrict          fr,
 517                      t_mdatoms                   * gmx_restrict     mdatoms,
 518                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 519                      t_nrnb                      * gmx_restrict        nrnb)
 520 {
 521     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 522      * just 0 for non-waters.
 523      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 524      * jnr indices corresponding to data put in the four positions in the SIMD register.
 525      */
 526     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 527     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 528     int              jnrA,jnrB,jnrC,jnrD;
 529     int              jnrE,jnrF,jnrG,jnrH;
 530     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 531     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 532     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 533     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 534     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 535     real             rcutoff_scalar;
 536     real             *shiftvec,*fshift,*x,*f;
 537     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 538     real             scratch[4*DIM];
 539     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 540     real *           vdwioffsetptr0;
 541     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 542     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 543     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 544     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 545     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 546     real             *charge;
 547     int              nvdwtype;
 548     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 549     int              *vdwtype;
 550     real             *vdwparam;
 551     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 552     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 553     __m256i          ewitab;
 554     __m128i          ewitab_lo,ewitab_hi;
 555     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 556     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 557     real             *ewtab;
 558     __m256           dummy_mask,cutoff_mask;
 559     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 560     __m256           one     = _mm256_set1_ps(1.0);
 561     __m256           two     = _mm256_set1_ps(2.0);
 562     x                = xx[0];
 563     f                = ff[0];
 564
 565     nri              = nlist->nri;
 566     iinr             = nlist->iinr;
 567     jindex           = nlist->jindex;
 568     jjnr             = nlist->jjnr;
 569     shiftidx         = nlist->shift;
 570     gid              = nlist->gid;
 571     shiftvec         = fr->shift_vec[0];
 572     fshift           = fr->fshift[0];
 573     facel            = _mm256_set1_ps(fr->epsfac);
 574     charge           = mdatoms->chargeA;
 575     nvdwtype         = fr->ntype;
 576     vdwparam         = fr->nbfp;
 577     vdwtype          = mdatoms->typeA;
 578
 579     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 580     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 581     beta2            = _mm256_mul_ps(beta,beta);
 582     beta3            = _mm256_mul_ps(beta,beta2);
 583
 584     ewtab            = fr->ic->tabq_coul_F;
 585     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 586     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 587
 588     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 589     rcutoff_scalar   = fr->rcoulomb;
 590     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 591     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 592
 593     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 594     rvdw             = _mm256_set1_ps(fr->rvdw);
 595
 596     /* Avoid stupid compiler warnings */
 597     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 598     j_coord_offsetA = 0;
 599     j_coord_offsetB = 0;
 600     j_coord_offsetC = 0;
 601     j_coord_offsetD = 0;
 602     j_coord_offsetE = 0;
 603     j_coord_offsetF = 0;
 604     j_coord_offsetG = 0;
 605     j_coord_offsetH = 0;
 606
 607     outeriter        = 0;
 608     inneriter        = 0;
 609
 610     for(iidx=0;iidx<4*DIM;iidx++)
 611     {
 612         scratch[iidx] = 0.0;
 613     }
 614
 615     /* Start outer loop over neighborlists */
 616     for(iidx=0; iidx<nri; iidx++)
 617     {
 618         /* Load shift vector for this list */
 619         i_shift_offset   = DIM*shiftidx[iidx];
 620
 621         /* Load limits for loop over neighbors */
 622         j_index_start    = jindex[iidx];
 623         j_index_end      = jindex[iidx+1];
 624
 625         /* Get outer coordinate index */
 626         inr              = iinr[iidx];
 627         i_coord_offset   = DIM*inr;
 628
 629         /* Load i particle coords and add shift vector */
 630         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 631
 632         fix0             = _mm256_setzero_ps();
 633         fiy0             = _mm256_setzero_ps();
 634         fiz0             = _mm256_setzero_ps();
 635
 636         /* Load parameters for i particles */
 637         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 638         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 639
 640         /* Start inner kernel loop */
 641         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 642         {
 643
 644             /* Get j neighbor index, and coordinate index */
 645             jnrA             = jjnr[jidx];
 646             jnrB             = jjnr[jidx+1];
 647             jnrC             = jjnr[jidx+2];
 648             jnrD             = jjnr[jidx+3];
 649             jnrE             = jjnr[jidx+4];
 650             jnrF             = jjnr[jidx+5];
 651             jnrG             = jjnr[jidx+6];
 652             jnrH             = jjnr[jidx+7];
 653             j_coord_offsetA  = DIM*jnrA;
 654             j_coord_offsetB  = DIM*jnrB;
 655             j_coord_offsetC  = DIM*jnrC;
 656             j_coord_offsetD  = DIM*jnrD;
 657             j_coord_offsetE  = DIM*jnrE;
 658             j_coord_offsetF  = DIM*jnrF;
 659             j_coord_offsetG  = DIM*jnrG;
 660             j_coord_offsetH  = DIM*jnrH;
 661
 662             /* load j atom coordinates */
 663             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 664                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 665                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 666                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 667                                                  &jx0,&jy0,&jz0);
 668
 669             /* Calculate displacement vector */
 670             dx00             = _mm256_sub_ps(ix0,jx0);
 671             dy00             = _mm256_sub_ps(iy0,jy0);
 672             dz00             = _mm256_sub_ps(iz0,jz0);
 673
 674             /* Calculate squared distance and things based on it */
 675             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 676
 677             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 678
 679             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 680
 681             /* Load parameters for j particles */
 682             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 683                                                                  charge+jnrC+0,charge+jnrD+0,
 684                                                                  charge+jnrE+0,charge+jnrF+0,
 685                                                                  charge+jnrG+0,charge+jnrH+0);
 686             vdwjidx0A        = 2*vdwtype[jnrA+0];
 687             vdwjidx0B        = 2*vdwtype[jnrB+0];
 688             vdwjidx0C        = 2*vdwtype[jnrC+0];
 689             vdwjidx0D        = 2*vdwtype[jnrD+0];
 690             vdwjidx0E        = 2*vdwtype[jnrE+0];
 691             vdwjidx0F        = 2*vdwtype[jnrF+0];
 692             vdwjidx0G        = 2*vdwtype[jnrG+0];
 693             vdwjidx0H        = 2*vdwtype[jnrH+0];
 694
 695             /**************************
 696              * CALCULATE INTERACTIONS *
 697              **************************/
 698
 699             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 700             {
 701
 702             r00              = _mm256_mul_ps(rsq00,rinv00);
 703
 704             /* Compute parameters for interactions between i and j atoms */
 705             qq00             = _mm256_mul_ps(iq0,jq0);
 706             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 707                                             vdwioffsetptr0+vdwjidx0B,
 708                                             vdwioffsetptr0+vdwjidx0C,
 709                                             vdwioffsetptr0+vdwjidx0D,
 710                                             vdwioffsetptr0+vdwjidx0E,
 711                                             vdwioffsetptr0+vdwjidx0F,
 712                                             vdwioffsetptr0+vdwjidx0G,
 713                                             vdwioffsetptr0+vdwjidx0H,
 714                                             &c6_00,&c12_00);
 715
 716             /* EWALD ELECTROSTATICS */
 717
 718             /* Analytical PME correction */
 719             zeta2            = _mm256_mul_ps(beta2,rsq00);
 720             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 721             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 722             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 723             felec            = _mm256_mul_ps(qq00,felec);
 724
 725             /* LENNARD-JONES DISPERSION/REPULSION */
 726
 727             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 728             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 729
 730             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 731
 732             fscal            = _mm256_add_ps(felec,fvdw);
 733
 734             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 735
 736             /* Calculate temporary vectorial force */
 737             tx               = _mm256_mul_ps(fscal,dx00);
 738             ty               = _mm256_mul_ps(fscal,dy00);
 739             tz               = _mm256_mul_ps(fscal,dz00);
 740
 741             /* Update vectorial force */
 742             fix0             = _mm256_add_ps(fix0,tx);
 743             fiy0             = _mm256_add_ps(fiy0,ty);
 744             fiz0             = _mm256_add_ps(fiz0,tz);
 745
 746             fjptrA             = f+j_coord_offsetA;
 747             fjptrB             = f+j_coord_offsetB;
 748             fjptrC             = f+j_coord_offsetC;
 749             fjptrD             = f+j_coord_offsetD;
 750             fjptrE             = f+j_coord_offsetE;
 751             fjptrF             = f+j_coord_offsetF;
 752             fjptrG             = f+j_coord_offsetG;
 753             fjptrH             = f+j_coord_offsetH;
 754             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 755
 756             }
 757
 758             /* Inner loop uses 66 flops */
 759         }
 760
 761         if(jidx<j_index_end)
 762         {
 763
 764             /* Get j neighbor index, and coordinate index */
 765             jnrlistA         = jjnr[jidx];
 766             jnrlistB         = jjnr[jidx+1];
 767             jnrlistC         = jjnr[jidx+2];
 768             jnrlistD         = jjnr[jidx+3];
 769             jnrlistE         = jjnr[jidx+4];
 770             jnrlistF         = jjnr[jidx+5];
 771             jnrlistG         = jjnr[jidx+6];
 772             jnrlistH         = jjnr[jidx+7];
 773             /* Sign of each element will be negative for non-real atoms.
 774              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 775              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 776              */
 777             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 778                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 779
 780             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 781             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 782             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 783             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 784             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 785             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 786             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 787             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 788             j_coord_offsetA  = DIM*jnrA;
 789             j_coord_offsetB  = DIM*jnrB;
 790             j_coord_offsetC  = DIM*jnrC;
 791             j_coord_offsetD  = DIM*jnrD;
 792             j_coord_offsetE  = DIM*jnrE;
 793             j_coord_offsetF  = DIM*jnrF;
 794             j_coord_offsetG  = DIM*jnrG;
 795             j_coord_offsetH  = DIM*jnrH;
 796
 797             /* load j atom coordinates */
 798             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 799                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 800                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 801                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 802                                                  &jx0,&jy0,&jz0);
 803
 804             /* Calculate displacement vector */
 805             dx00             = _mm256_sub_ps(ix0,jx0);
 806             dy00             = _mm256_sub_ps(iy0,jy0);
 807             dz00             = _mm256_sub_ps(iz0,jz0);
 808
 809             /* Calculate squared distance and things based on it */
 810             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 811
 812             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 813
 814             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 815
 816             /* Load parameters for j particles */
 817             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 818                                                                  charge+jnrC+0,charge+jnrD+0,
 819                                                                  charge+jnrE+0,charge+jnrF+0,
 820                                                                  charge+jnrG+0,charge+jnrH+0);
 821             vdwjidx0A        = 2*vdwtype[jnrA+0];
 822             vdwjidx0B        = 2*vdwtype[jnrB+0];
 823             vdwjidx0C        = 2*vdwtype[jnrC+0];
 824             vdwjidx0D        = 2*vdwtype[jnrD+0];
 825             vdwjidx0E        = 2*vdwtype[jnrE+0];
 826             vdwjidx0F        = 2*vdwtype[jnrF+0];
 827             vdwjidx0G        = 2*vdwtype[jnrG+0];
 828             vdwjidx0H        = 2*vdwtype[jnrH+0];
 829
 830             /**************************
 831              * CALCULATE INTERACTIONS *
 832              **************************/
 833
 834             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 835             {
 836
 837             r00              = _mm256_mul_ps(rsq00,rinv00);
 838             r00              = _mm256_andnot_ps(dummy_mask,r00);
 839
 840             /* Compute parameters for interactions between i and j atoms */
 841             qq00             = _mm256_mul_ps(iq0,jq0);
 842             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 843                                             vdwioffsetptr0+vdwjidx0B,
 844                                             vdwioffsetptr0+vdwjidx0C,
 845                                             vdwioffsetptr0+vdwjidx0D,
 846                                             vdwioffsetptr0+vdwjidx0E,
 847                                             vdwioffsetptr0+vdwjidx0F,
 848                                             vdwioffsetptr0+vdwjidx0G,
 849                                             vdwioffsetptr0+vdwjidx0H,
 850                                             &c6_00,&c12_00);
 851
 852             /* EWALD ELECTROSTATICS */
 853
 854             /* Analytical PME correction */
 855             zeta2            = _mm256_mul_ps(beta2,rsq00);
 856             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 857             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 858             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 859             felec            = _mm256_mul_ps(qq00,felec);
 860
 861             /* LENNARD-JONES DISPERSION/REPULSION */
 862
 863             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 864             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 865
 866             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 867
 868             fscal            = _mm256_add_ps(felec,fvdw);
 869
 870             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 871
 872             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 873
 874             /* Calculate temporary vectorial force */
 875             tx               = _mm256_mul_ps(fscal,dx00);
 876             ty               = _mm256_mul_ps(fscal,dy00);
 877             tz               = _mm256_mul_ps(fscal,dz00);
 878
 879             /* Update vectorial force */
 880             fix0             = _mm256_add_ps(fix0,tx);
 881             fiy0             = _mm256_add_ps(fiy0,ty);
 882             fiz0             = _mm256_add_ps(fiz0,tz);
 883
 884             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 885             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 886             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 887             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 888             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 889             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 890             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 891             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 892             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 893
 894             }
 895
 896             /* Inner loop uses 67 flops */
 897         }
 898
 899         /* End of innermost loop */
 900
 901         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 902                                                  f+i_coord_offset,fshift+i_shift_offset);
 903
 904         /* Increment number of inner iterations */
 905         inneriter                  += j_index_end - j_index_start;
 906
 907         /* Outer loop uses 7 flops */
 908     }
 909
 910     /* Increment number of outer iterations */
 911     outeriter        += nri;
 912
 913     /* Update outer/inner flops */
 914
 915     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*67);
 916 }