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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_VF_avx_256_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_VF_avx_256_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  89     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     int              nvdwtype;
  94     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  98     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  99     __m256i          ewitab;
 100     __m128i          ewitab_lo,ewitab_hi;
 101     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 102     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 103     real             *ewtab;
 104     __m256           dummy_mask,cutoff_mask;
 105     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 106     __m256           one     = _mm256_set1_ps(1.0);
 107     __m256           two     = _mm256_set1_ps(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_ps(fr->epsfac);
 120     charge           = mdatoms->chargeA;
 121     nvdwtype         = fr->ntype;
 122     vdwparam         = fr->nbfp;
 123     vdwtype          = mdatoms->typeA;
 124
 125     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 126     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 127     beta2            = _mm256_mul_ps(beta,beta);
 128     beta3            = _mm256_mul_ps(beta,beta2);
 129
 130     ewtab            = fr->ic->tabq_coul_FDV0;
 131     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 132     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 133
 134     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 135     rcutoff_scalar   = fr->rcoulomb;
 136     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 137     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 138
 139     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 140     rvdw             = _mm256_set1_ps(fr->rvdw);
 141
 142     /* Avoid stupid compiler warnings */
 143     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 144     j_coord_offsetA = 0;
 145     j_coord_offsetB = 0;
 146     j_coord_offsetC = 0;
 147     j_coord_offsetD = 0;
 148     j_coord_offsetE = 0;
 149     j_coord_offsetF = 0;
 150     j_coord_offsetG = 0;
 151     j_coord_offsetH = 0;
 152
 153     outeriter        = 0;
 154     inneriter        = 0;
 155
 156     for(iidx=0;iidx<4*DIM;iidx++)
 157     {
 158         scratch[iidx] = 0.0;
 159     }
 160
 161     /* Start outer loop over neighborlists */
 162     for(iidx=0; iidx<nri; iidx++)
 163     {
 164         /* Load shift vector for this list */
 165         i_shift_offset   = DIM*shiftidx[iidx];
 166
 167         /* Load limits for loop over neighbors */
 168         j_index_start    = jindex[iidx];
 169         j_index_end      = jindex[iidx+1];
 170
 171         /* Get outer coordinate index */
 172         inr              = iinr[iidx];
 173         i_coord_offset   = DIM*inr;
 174
 175         /* Load i particle coords and add shift vector */
 176         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 177
 178         fix0             = _mm256_setzero_ps();
 179         fiy0             = _mm256_setzero_ps();
 180         fiz0             = _mm256_setzero_ps();
 181
 182         /* Load parameters for i particles */
 183         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 184         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 185
 186         /* Reset potential sums */
 187         velecsum         = _mm256_setzero_ps();
 188         vvdwsum          = _mm256_setzero_ps();
 189
 190         /* Start inner kernel loop */
 191         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 192         {
 193
 194             /* Get j neighbor index, and coordinate index */
 195             jnrA             = jjnr[jidx];
 196             jnrB             = jjnr[jidx+1];
 197             jnrC             = jjnr[jidx+2];
 198             jnrD             = jjnr[jidx+3];
 199             jnrE             = jjnr[jidx+4];
 200             jnrF             = jjnr[jidx+5];
 201             jnrG             = jjnr[jidx+6];
 202             jnrH             = jjnr[jidx+7];
 203             j_coord_offsetA  = DIM*jnrA;
 204             j_coord_offsetB  = DIM*jnrB;
 205             j_coord_offsetC  = DIM*jnrC;
 206             j_coord_offsetD  = DIM*jnrD;
 207             j_coord_offsetE  = DIM*jnrE;
 208             j_coord_offsetF  = DIM*jnrF;
 209             j_coord_offsetG  = DIM*jnrG;
 210             j_coord_offsetH  = DIM*jnrH;
 211
 212             /* load j atom coordinates */
 213             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 214                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 215                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 216                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 217                                                  &jx0,&jy0,&jz0);
 218
 219             /* Calculate displacement vector */
 220             dx00             = _mm256_sub_ps(ix0,jx0);
 221             dy00             = _mm256_sub_ps(iy0,jy0);
 222             dz00             = _mm256_sub_ps(iz0,jz0);
 223
 224             /* Calculate squared distance and things based on it */
 225             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 226
 227             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 228
 229             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 230
 231             /* Load parameters for j particles */
 232             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 233                                                                  charge+jnrC+0,charge+jnrD+0,
 234                                                                  charge+jnrE+0,charge+jnrF+0,
 235                                                                  charge+jnrG+0,charge+jnrH+0);
 236             vdwjidx0A        = 2*vdwtype[jnrA+0];
 237             vdwjidx0B        = 2*vdwtype[jnrB+0];
 238             vdwjidx0C        = 2*vdwtype[jnrC+0];
 239             vdwjidx0D        = 2*vdwtype[jnrD+0];
 240             vdwjidx0E        = 2*vdwtype[jnrE+0];
 241             vdwjidx0F        = 2*vdwtype[jnrF+0];
 242             vdwjidx0G        = 2*vdwtype[jnrG+0];
 243             vdwjidx0H        = 2*vdwtype[jnrH+0];
 244
 245             /**************************
 246              * CALCULATE INTERACTIONS *
 247              **************************/
 248
 249             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 250             {
 251
 252             r00              = _mm256_mul_ps(rsq00,rinv00);
 253
 254             /* Compute parameters for interactions between i and j atoms */
 255             qq00             = _mm256_mul_ps(iq0,jq0);
 256             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 257                                             vdwioffsetptr0+vdwjidx0B,
 258                                             vdwioffsetptr0+vdwjidx0C,
 259                                             vdwioffsetptr0+vdwjidx0D,
 260                                             vdwioffsetptr0+vdwjidx0E,
 261                                             vdwioffsetptr0+vdwjidx0F,
 262                                             vdwioffsetptr0+vdwjidx0G,
 263                                             vdwioffsetptr0+vdwjidx0H,
 264                                             &c6_00,&c12_00);
 265
 266             /* EWALD ELECTROSTATICS */
 267
 268             /* Analytical PME correction */
 269             zeta2            = _mm256_mul_ps(beta2,rsq00);
 270             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 271             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 272             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 273             felec            = _mm256_mul_ps(qq00,felec);
 274             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 275             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 276             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 277             velec            = _mm256_mul_ps(qq00,velec);
 278
 279             /* LENNARD-JONES DISPERSION/REPULSION */
 280
 281             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 282             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 283             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 284             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) ,
 285                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 286             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 287
 288             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 289
 290             /* Update potential sum for this i atom from the interaction with this j atom. */
 291             velec            = _mm256_and_ps(velec,cutoff_mask);
 292             velecsum         = _mm256_add_ps(velecsum,velec);
 293             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 294             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 295
 296             fscal            = _mm256_add_ps(felec,fvdw);
 297
 298             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 299
 300             /* Calculate temporary vectorial force */
 301             tx               = _mm256_mul_ps(fscal,dx00);
 302             ty               = _mm256_mul_ps(fscal,dy00);
 303             tz               = _mm256_mul_ps(fscal,dz00);
 304
 305             /* Update vectorial force */
 306             fix0             = _mm256_add_ps(fix0,tx);
 307             fiy0             = _mm256_add_ps(fiy0,ty);
 308             fiz0             = _mm256_add_ps(fiz0,tz);
 309
 310             fjptrA             = f+j_coord_offsetA;
 311             fjptrB             = f+j_coord_offsetB;
 312             fjptrC             = f+j_coord_offsetC;
 313             fjptrD             = f+j_coord_offsetD;
 314             fjptrE             = f+j_coord_offsetE;
 315             fjptrF             = f+j_coord_offsetF;
 316             fjptrG             = f+j_coord_offsetG;
 317             fjptrH             = f+j_coord_offsetH;
 318             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 319
 320             }
 321
 322             /* Inner loop uses 127 flops */
 323         }
 324
 325         if(jidx<j_index_end)
 326         {
 327
 328             /* Get j neighbor index, and coordinate index */
 329             jnrlistA         = jjnr[jidx];
 330             jnrlistB         = jjnr[jidx+1];
 331             jnrlistC         = jjnr[jidx+2];
 332             jnrlistD         = jjnr[jidx+3];
 333             jnrlistE         = jjnr[jidx+4];
 334             jnrlistF         = jjnr[jidx+5];
 335             jnrlistG         = jjnr[jidx+6];
 336             jnrlistH         = jjnr[jidx+7];
 337             /* Sign of each element will be negative for non-real atoms.
 338              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 339              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 340              */
 341             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 342                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 343
 344             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 345             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 346             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 347             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 348             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 349             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 350             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 351             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 352             j_coord_offsetA  = DIM*jnrA;
 353             j_coord_offsetB  = DIM*jnrB;
 354             j_coord_offsetC  = DIM*jnrC;
 355             j_coord_offsetD  = DIM*jnrD;
 356             j_coord_offsetE  = DIM*jnrE;
 357             j_coord_offsetF  = DIM*jnrF;
 358             j_coord_offsetG  = DIM*jnrG;
 359             j_coord_offsetH  = DIM*jnrH;
 360
 361             /* load j atom coordinates */
 362             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 363                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 364                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 365                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 366                                                  &jx0,&jy0,&jz0);
 367
 368             /* Calculate displacement vector */
 369             dx00             = _mm256_sub_ps(ix0,jx0);
 370             dy00             = _mm256_sub_ps(iy0,jy0);
 371             dz00             = _mm256_sub_ps(iz0,jz0);
 372
 373             /* Calculate squared distance and things based on it */
 374             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 375
 376             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 377
 378             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 379
 380             /* Load parameters for j particles */
 381             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 382                                                                  charge+jnrC+0,charge+jnrD+0,
 383                                                                  charge+jnrE+0,charge+jnrF+0,
 384                                                                  charge+jnrG+0,charge+jnrH+0);
 385             vdwjidx0A        = 2*vdwtype[jnrA+0];
 386             vdwjidx0B        = 2*vdwtype[jnrB+0];
 387             vdwjidx0C        = 2*vdwtype[jnrC+0];
 388             vdwjidx0D        = 2*vdwtype[jnrD+0];
 389             vdwjidx0E        = 2*vdwtype[jnrE+0];
 390             vdwjidx0F        = 2*vdwtype[jnrF+0];
 391             vdwjidx0G        = 2*vdwtype[jnrG+0];
 392             vdwjidx0H        = 2*vdwtype[jnrH+0];
 393
 394             /**************************
 395              * CALCULATE INTERACTIONS *
 396              **************************/
 397
 398             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 399             {
 400
 401             r00              = _mm256_mul_ps(rsq00,rinv00);
 402             r00              = _mm256_andnot_ps(dummy_mask,r00);
 403
 404             /* Compute parameters for interactions between i and j atoms */
 405             qq00             = _mm256_mul_ps(iq0,jq0);
 406             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 407                                             vdwioffsetptr0+vdwjidx0B,
 408                                             vdwioffsetptr0+vdwjidx0C,
 409                                             vdwioffsetptr0+vdwjidx0D,
 410                                             vdwioffsetptr0+vdwjidx0E,
 411                                             vdwioffsetptr0+vdwjidx0F,
 412                                             vdwioffsetptr0+vdwjidx0G,
 413                                             vdwioffsetptr0+vdwjidx0H,
 414                                             &c6_00,&c12_00);
 415
 416             /* EWALD ELECTROSTATICS */
 417
 418             /* Analytical PME correction */
 419             zeta2            = _mm256_mul_ps(beta2,rsq00);
 420             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 421             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 422             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 423             felec            = _mm256_mul_ps(qq00,felec);
 424             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 425             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 426             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 427             velec            = _mm256_mul_ps(qq00,velec);
 428
 429             /* LENNARD-JONES DISPERSION/REPULSION */
 430
 431             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 432             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 433             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 434             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) ,
 435                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 436             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 437
 438             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 439
 440             /* Update potential sum for this i atom from the interaction with this j atom. */
 441             velec            = _mm256_and_ps(velec,cutoff_mask);
 442             velec            = _mm256_andnot_ps(dummy_mask,velec);
 443             velecsum         = _mm256_add_ps(velecsum,velec);
 444             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 445             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 446             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 447
 448             fscal            = _mm256_add_ps(felec,fvdw);
 449
 450             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 451
 452             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 453
 454             /* Calculate temporary vectorial force */
 455             tx               = _mm256_mul_ps(fscal,dx00);
 456             ty               = _mm256_mul_ps(fscal,dy00);
 457             tz               = _mm256_mul_ps(fscal,dz00);
 458
 459             /* Update vectorial force */
 460             fix0             = _mm256_add_ps(fix0,tx);
 461             fiy0             = _mm256_add_ps(fiy0,ty);
 462             fiz0             = _mm256_add_ps(fiz0,tz);
 463
 464             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 465             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 466             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 467             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 468             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 469             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 470             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 471             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 472             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 473
 474             }
 475
 476             /* Inner loop uses 128 flops */
 477         }
 478
 479         /* End of innermost loop */
 480
 481         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 482                                                  f+i_coord_offset,fshift+i_shift_offset);
 483
 484         ggid                        = gid[iidx];
 485         /* Update potential energies */
 486         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 487         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 488
 489         /* Increment number of inner iterations */
 490         inneriter                  += j_index_end - j_index_start;
 491
 492         /* Outer loop uses 9 flops */
 493     }
 494
 495     /* Increment number of outer iterations */
 496     outeriter        += nri;
 497
 498     /* Update outer/inner flops */
 499
 500     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*128);
 501 }
 502 /*
 503  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_avx_256_single
 504  * Electrostatics interaction: Ewald
 505  * VdW interaction:            LennardJones
 506  * Geometry:                   Particle-Particle
 507  * Calculate force/pot:        Force
 508  */
 509 void
 510 nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_avx_256_single
 511                     (t_nblist                    * gmx_restrict       nlist,
 512                      rvec                        * gmx_restrict          xx,
 513                      rvec                        * gmx_restrict          ff,
 514                      t_forcerec                  * gmx_restrict          fr,
 515                      t_mdatoms                   * gmx_restrict     mdatoms,
 516                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 517                      t_nrnb                      * gmx_restrict        nrnb)
 518 {
 519     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 520      * just 0 for non-waters.
 521      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 522      * jnr indices corresponding to data put in the four positions in the SIMD register.
 523      */
 524     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 525     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 526     int              jnrA,jnrB,jnrC,jnrD;
 527     int              jnrE,jnrF,jnrG,jnrH;
 528     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 529     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 530     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 531     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 532     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 533     real             rcutoff_scalar;
 534     real             *shiftvec,*fshift,*x,*f;
 535     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 536     real             scratch[4*DIM];
 537     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 538     real *           vdwioffsetptr0;
 539     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 540     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 541     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 542     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 543     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 544     real             *charge;
 545     int              nvdwtype;
 546     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 547     int              *vdwtype;
 548     real             *vdwparam;
 549     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 550     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 551     __m256i          ewitab;
 552     __m128i          ewitab_lo,ewitab_hi;
 553     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 554     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 555     real             *ewtab;
 556     __m256           dummy_mask,cutoff_mask;
 557     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 558     __m256           one     = _mm256_set1_ps(1.0);
 559     __m256           two     = _mm256_set1_ps(2.0);
 560     x                = xx[0];
 561     f                = ff[0];
 562
 563     nri              = nlist->nri;
 564     iinr             = nlist->iinr;
 565     jindex           = nlist->jindex;
 566     jjnr             = nlist->jjnr;
 567     shiftidx         = nlist->shift;
 568     gid              = nlist->gid;
 569     shiftvec         = fr->shift_vec[0];
 570     fshift           = fr->fshift[0];
 571     facel            = _mm256_set1_ps(fr->epsfac);
 572     charge           = mdatoms->chargeA;
 573     nvdwtype         = fr->ntype;
 574     vdwparam         = fr->nbfp;
 575     vdwtype          = mdatoms->typeA;
 576
 577     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 578     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 579     beta2            = _mm256_mul_ps(beta,beta);
 580     beta3            = _mm256_mul_ps(beta,beta2);
 581
 582     ewtab            = fr->ic->tabq_coul_F;
 583     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 584     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 585
 586     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 587     rcutoff_scalar   = fr->rcoulomb;
 588     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 589     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 590
 591     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 592     rvdw             = _mm256_set1_ps(fr->rvdw);
 593
 594     /* Avoid stupid compiler warnings */
 595     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 596     j_coord_offsetA = 0;
 597     j_coord_offsetB = 0;
 598     j_coord_offsetC = 0;
 599     j_coord_offsetD = 0;
 600     j_coord_offsetE = 0;
 601     j_coord_offsetF = 0;
 602     j_coord_offsetG = 0;
 603     j_coord_offsetH = 0;
 604
 605     outeriter        = 0;
 606     inneriter        = 0;
 607
 608     for(iidx=0;iidx<4*DIM;iidx++)
 609     {
 610         scratch[iidx] = 0.0;
 611     }
 612
 613     /* Start outer loop over neighborlists */
 614     for(iidx=0; iidx<nri; iidx++)
 615     {
 616         /* Load shift vector for this list */
 617         i_shift_offset   = DIM*shiftidx[iidx];
 618
 619         /* Load limits for loop over neighbors */
 620         j_index_start    = jindex[iidx];
 621         j_index_end      = jindex[iidx+1];
 622
 623         /* Get outer coordinate index */
 624         inr              = iinr[iidx];
 625         i_coord_offset   = DIM*inr;
 626
 627         /* Load i particle coords and add shift vector */
 628         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 629
 630         fix0             = _mm256_setzero_ps();
 631         fiy0             = _mm256_setzero_ps();
 632         fiz0             = _mm256_setzero_ps();
 633
 634         /* Load parameters for i particles */
 635         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 636         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 637
 638         /* Start inner kernel loop */
 639         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 640         {
 641
 642             /* Get j neighbor index, and coordinate index */
 643             jnrA             = jjnr[jidx];
 644             jnrB             = jjnr[jidx+1];
 645             jnrC             = jjnr[jidx+2];
 646             jnrD             = jjnr[jidx+3];
 647             jnrE             = jjnr[jidx+4];
 648             jnrF             = jjnr[jidx+5];
 649             jnrG             = jjnr[jidx+6];
 650             jnrH             = jjnr[jidx+7];
 651             j_coord_offsetA  = DIM*jnrA;
 652             j_coord_offsetB  = DIM*jnrB;
 653             j_coord_offsetC  = DIM*jnrC;
 654             j_coord_offsetD  = DIM*jnrD;
 655             j_coord_offsetE  = DIM*jnrE;
 656             j_coord_offsetF  = DIM*jnrF;
 657             j_coord_offsetG  = DIM*jnrG;
 658             j_coord_offsetH  = DIM*jnrH;
 659
 660             /* load j atom coordinates */
 661             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 662                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 663                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 664                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 665                                                  &jx0,&jy0,&jz0);
 666
 667             /* Calculate displacement vector */
 668             dx00             = _mm256_sub_ps(ix0,jx0);
 669             dy00             = _mm256_sub_ps(iy0,jy0);
 670             dz00             = _mm256_sub_ps(iz0,jz0);
 671
 672             /* Calculate squared distance and things based on it */
 673             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 674
 675             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 676
 677             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 678
 679             /* Load parameters for j particles */
 680             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 681                                                                  charge+jnrC+0,charge+jnrD+0,
 682                                                                  charge+jnrE+0,charge+jnrF+0,
 683                                                                  charge+jnrG+0,charge+jnrH+0);
 684             vdwjidx0A        = 2*vdwtype[jnrA+0];
 685             vdwjidx0B        = 2*vdwtype[jnrB+0];
 686             vdwjidx0C        = 2*vdwtype[jnrC+0];
 687             vdwjidx0D        = 2*vdwtype[jnrD+0];
 688             vdwjidx0E        = 2*vdwtype[jnrE+0];
 689             vdwjidx0F        = 2*vdwtype[jnrF+0];
 690             vdwjidx0G        = 2*vdwtype[jnrG+0];
 691             vdwjidx0H        = 2*vdwtype[jnrH+0];
 692
 693             /**************************
 694              * CALCULATE INTERACTIONS *
 695              **************************/
 696
 697             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 698             {
 699
 700             r00              = _mm256_mul_ps(rsq00,rinv00);
 701
 702             /* Compute parameters for interactions between i and j atoms */
 703             qq00             = _mm256_mul_ps(iq0,jq0);
 704             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 705                                             vdwioffsetptr0+vdwjidx0B,
 706                                             vdwioffsetptr0+vdwjidx0C,
 707                                             vdwioffsetptr0+vdwjidx0D,
 708                                             vdwioffsetptr0+vdwjidx0E,
 709                                             vdwioffsetptr0+vdwjidx0F,
 710                                             vdwioffsetptr0+vdwjidx0G,
 711                                             vdwioffsetptr0+vdwjidx0H,
 712                                             &c6_00,&c12_00);
 713
 714             /* EWALD ELECTROSTATICS */
 715
 716             /* Analytical PME correction */
 717             zeta2            = _mm256_mul_ps(beta2,rsq00);
 718             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 719             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 720             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 721             felec            = _mm256_mul_ps(qq00,felec);
 722
 723             /* LENNARD-JONES DISPERSION/REPULSION */
 724
 725             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 726             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 727
 728             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 729
 730             fscal            = _mm256_add_ps(felec,fvdw);
 731
 732             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 733
 734             /* Calculate temporary vectorial force */
 735             tx               = _mm256_mul_ps(fscal,dx00);
 736             ty               = _mm256_mul_ps(fscal,dy00);
 737             tz               = _mm256_mul_ps(fscal,dz00);
 738
 739             /* Update vectorial force */
 740             fix0             = _mm256_add_ps(fix0,tx);
 741             fiy0             = _mm256_add_ps(fiy0,ty);
 742             fiz0             = _mm256_add_ps(fiz0,tz);
 743
 744             fjptrA             = f+j_coord_offsetA;
 745             fjptrB             = f+j_coord_offsetB;
 746             fjptrC             = f+j_coord_offsetC;
 747             fjptrD             = f+j_coord_offsetD;
 748             fjptrE             = f+j_coord_offsetE;
 749             fjptrF             = f+j_coord_offsetF;
 750             fjptrG             = f+j_coord_offsetG;
 751             fjptrH             = f+j_coord_offsetH;
 752             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 753
 754             }
 755
 756             /* Inner loop uses 66 flops */
 757         }
 758
 759         if(jidx<j_index_end)
 760         {
 761
 762             /* Get j neighbor index, and coordinate index */
 763             jnrlistA         = jjnr[jidx];
 764             jnrlistB         = jjnr[jidx+1];
 765             jnrlistC         = jjnr[jidx+2];
 766             jnrlistD         = jjnr[jidx+3];
 767             jnrlistE         = jjnr[jidx+4];
 768             jnrlistF         = jjnr[jidx+5];
 769             jnrlistG         = jjnr[jidx+6];
 770             jnrlistH         = jjnr[jidx+7];
 771             /* Sign of each element will be negative for non-real atoms.
 772              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 773              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 774              */
 775             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 776                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 777
 778             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 779             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 780             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 781             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 782             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 783             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 784             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 785             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 786             j_coord_offsetA  = DIM*jnrA;
 787             j_coord_offsetB  = DIM*jnrB;
 788             j_coord_offsetC  = DIM*jnrC;
 789             j_coord_offsetD  = DIM*jnrD;
 790             j_coord_offsetE  = DIM*jnrE;
 791             j_coord_offsetF  = DIM*jnrF;
 792             j_coord_offsetG  = DIM*jnrG;
 793             j_coord_offsetH  = DIM*jnrH;
 794
 795             /* load j atom coordinates */
 796             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 797                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 798                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 799                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 800                                                  &jx0,&jy0,&jz0);
 801
 802             /* Calculate displacement vector */
 803             dx00             = _mm256_sub_ps(ix0,jx0);
 804             dy00             = _mm256_sub_ps(iy0,jy0);
 805             dz00             = _mm256_sub_ps(iz0,jz0);
 806
 807             /* Calculate squared distance and things based on it */
 808             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 809
 810             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 811
 812             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 813
 814             /* Load parameters for j particles */
 815             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 816                                                                  charge+jnrC+0,charge+jnrD+0,
 817                                                                  charge+jnrE+0,charge+jnrF+0,
 818                                                                  charge+jnrG+0,charge+jnrH+0);
 819             vdwjidx0A        = 2*vdwtype[jnrA+0];
 820             vdwjidx0B        = 2*vdwtype[jnrB+0];
 821             vdwjidx0C        = 2*vdwtype[jnrC+0];
 822             vdwjidx0D        = 2*vdwtype[jnrD+0];
 823             vdwjidx0E        = 2*vdwtype[jnrE+0];
 824             vdwjidx0F        = 2*vdwtype[jnrF+0];
 825             vdwjidx0G        = 2*vdwtype[jnrG+0];
 826             vdwjidx0H        = 2*vdwtype[jnrH+0];
 827
 828             /**************************
 829              * CALCULATE INTERACTIONS *
 830              **************************/
 831
 832             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 833             {
 834
 835             r00              = _mm256_mul_ps(rsq00,rinv00);
 836             r00              = _mm256_andnot_ps(dummy_mask,r00);
 837
 838             /* Compute parameters for interactions between i and j atoms */
 839             qq00             = _mm256_mul_ps(iq0,jq0);
 840             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 841                                             vdwioffsetptr0+vdwjidx0B,
 842                                             vdwioffsetptr0+vdwjidx0C,
 843                                             vdwioffsetptr0+vdwjidx0D,
 844                                             vdwioffsetptr0+vdwjidx0E,
 845                                             vdwioffsetptr0+vdwjidx0F,
 846                                             vdwioffsetptr0+vdwjidx0G,
 847                                             vdwioffsetptr0+vdwjidx0H,
 848                                             &c6_00,&c12_00);
 849
 850             /* EWALD ELECTROSTATICS */
 851
 852             /* Analytical PME correction */
 853             zeta2            = _mm256_mul_ps(beta2,rsq00);
 854             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 855             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 856             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 857             felec            = _mm256_mul_ps(qq00,felec);
 858
 859             /* LENNARD-JONES DISPERSION/REPULSION */
 860
 861             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 862             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 863
 864             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 865
 866             fscal            = _mm256_add_ps(felec,fvdw);
 867
 868             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 869
 870             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 871
 872             /* Calculate temporary vectorial force */
 873             tx               = _mm256_mul_ps(fscal,dx00);
 874             ty               = _mm256_mul_ps(fscal,dy00);
 875             tz               = _mm256_mul_ps(fscal,dz00);
 876
 877             /* Update vectorial force */
 878             fix0             = _mm256_add_ps(fix0,tx);
 879             fiy0             = _mm256_add_ps(fiy0,ty);
 880             fiz0             = _mm256_add_ps(fiz0,tz);
 881
 882             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 883             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 884             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 885             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 886             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 887             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 888             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 889             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 890             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 891
 892             }
 893
 894             /* Inner loop uses 67 flops */
 895         }
 896
 897         /* End of innermost loop */
 898
 899         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 900                                                  f+i_coord_offset,fshift+i_shift_offset);
 901
 902         /* Increment number of inner iterations */
 903         inneriter                  += j_index_end - j_index_start;
 904
 905         /* Outer loop uses 7 flops */
 906     }
 907
 908     /* Increment number of outer iterations */
 909     outeriter        += nri;
 910
 911     /* Update outer/inner flops */
 912
 913     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*67);
 914 }