src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_VdwLJEw_GeomW3P1_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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_VF_avx_256_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_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     real *           vdwgridioffsetptr0;
  90     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  91     real *           vdwioffsetptr1;
  92     real *           vdwgridioffsetptr1;
  93     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  94     real *           vdwioffsetptr2;
  95     real *           vdwgridioffsetptr2;
  96     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  97     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  98     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  99     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 100     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 101     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 102     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 103     real             *charge;
 104     int              nvdwtype;
 105     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 106     int              *vdwtype;
 107     real             *vdwparam;
 108     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 109     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 110     __m256           c6grid_00;
 111     __m256           c6grid_10;
 112     __m256           c6grid_20;
 113     real             *vdwgridparam;
 114     __m256           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 115     __m256           one_half  = _mm256_set1_ps(0.5);
 116     __m256           minus_one = _mm256_set1_ps(-1.0);
 117     __m256i          ewitab;
 118     __m128i          ewitab_lo,ewitab_hi;
 119     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 120     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 121     real             *ewtab;
 122     __m256           dummy_mask,cutoff_mask;
 123     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 124     __m256           one     = _mm256_set1_ps(1.0);
 125     __m256           two     = _mm256_set1_ps(2.0);
 126     x                = xx[0];
 127     f                = ff[0];
 128
 129     nri              = nlist->nri;
 130     iinr             = nlist->iinr;
 131     jindex           = nlist->jindex;
 132     jjnr             = nlist->jjnr;
 133     shiftidx         = nlist->shift;
 134     gid              = nlist->gid;
 135     shiftvec         = fr->shift_vec[0];
 136     fshift           = fr->fshift[0];
 137     facel            = _mm256_set1_ps(fr->epsfac);
 138     charge           = mdatoms->chargeA;
 139     nvdwtype         = fr->ntype;
 140     vdwparam         = fr->nbfp;
 141     vdwtype          = mdatoms->typeA;
 142     vdwgridparam     = fr->ljpme_c6grid;
 143     sh_lj_ewald      = _mm256_set1_ps(fr->ic->sh_lj_ewald);
 144     ewclj            = _mm256_set1_ps(fr->ewaldcoeff_lj);
 145     ewclj2           = _mm256_mul_ps(minus_one,_mm256_mul_ps(ewclj,ewclj));
 146
 147     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 148     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 149     beta2            = _mm256_mul_ps(beta,beta);
 150     beta3            = _mm256_mul_ps(beta,beta2);
 151
 152     ewtab            = fr->ic->tabq_coul_FDV0;
 153     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 154     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 155
 156     /* Setup water-specific parameters */
 157     inr              = nlist->iinr[0];
 158     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 159     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 160     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 161     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 162     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 163
 164     /* Avoid stupid compiler warnings */
 165     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 166     j_coord_offsetA = 0;
 167     j_coord_offsetB = 0;
 168     j_coord_offsetC = 0;
 169     j_coord_offsetD = 0;
 170     j_coord_offsetE = 0;
 171     j_coord_offsetF = 0;
 172     j_coord_offsetG = 0;
 173     j_coord_offsetH = 0;
 174
 175     outeriter        = 0;
 176     inneriter        = 0;
 177
 178     for(iidx=0;iidx<4*DIM;iidx++)
 179     {
 180         scratch[iidx] = 0.0;
 181     }
 182
 183     /* Start outer loop over neighborlists */
 184     for(iidx=0; iidx<nri; iidx++)
 185     {
 186         /* Load shift vector for this list */
 187         i_shift_offset   = DIM*shiftidx[iidx];
 188
 189         /* Load limits for loop over neighbors */
 190         j_index_start    = jindex[iidx];
 191         j_index_end      = jindex[iidx+1];
 192
 193         /* Get outer coordinate index */
 194         inr              = iinr[iidx];
 195         i_coord_offset   = DIM*inr;
 196
 197         /* Load i particle coords and add shift vector */
 198         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 199                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 200
 201         fix0             = _mm256_setzero_ps();
 202         fiy0             = _mm256_setzero_ps();
 203         fiz0             = _mm256_setzero_ps();
 204         fix1             = _mm256_setzero_ps();
 205         fiy1             = _mm256_setzero_ps();
 206         fiz1             = _mm256_setzero_ps();
 207         fix2             = _mm256_setzero_ps();
 208         fiy2             = _mm256_setzero_ps();
 209         fiz2             = _mm256_setzero_ps();
 210
 211         /* Reset potential sums */
 212         velecsum         = _mm256_setzero_ps();
 213         vvdwsum          = _mm256_setzero_ps();
 214
 215         /* Start inner kernel loop */
 216         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 217         {
 218
 219             /* Get j neighbor index, and coordinate index */
 220             jnrA             = jjnr[jidx];
 221             jnrB             = jjnr[jidx+1];
 222             jnrC             = jjnr[jidx+2];
 223             jnrD             = jjnr[jidx+3];
 224             jnrE             = jjnr[jidx+4];
 225             jnrF             = jjnr[jidx+5];
 226             jnrG             = jjnr[jidx+6];
 227             jnrH             = jjnr[jidx+7];
 228             j_coord_offsetA  = DIM*jnrA;
 229             j_coord_offsetB  = DIM*jnrB;
 230             j_coord_offsetC  = DIM*jnrC;
 231             j_coord_offsetD  = DIM*jnrD;
 232             j_coord_offsetE  = DIM*jnrE;
 233             j_coord_offsetF  = DIM*jnrF;
 234             j_coord_offsetG  = DIM*jnrG;
 235             j_coord_offsetH  = DIM*jnrH;
 236
 237             /* load j atom coordinates */
 238             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 239                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 240                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 241                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 242                                                  &jx0,&jy0,&jz0);
 243
 244             /* Calculate displacement vector */
 245             dx00             = _mm256_sub_ps(ix0,jx0);
 246             dy00             = _mm256_sub_ps(iy0,jy0);
 247             dz00             = _mm256_sub_ps(iz0,jz0);
 248             dx10             = _mm256_sub_ps(ix1,jx0);
 249             dy10             = _mm256_sub_ps(iy1,jy0);
 250             dz10             = _mm256_sub_ps(iz1,jz0);
 251             dx20             = _mm256_sub_ps(ix2,jx0);
 252             dy20             = _mm256_sub_ps(iy2,jy0);
 253             dz20             = _mm256_sub_ps(iz2,jz0);
 254
 255             /* Calculate squared distance and things based on it */
 256             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 257             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 258             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 259
 260             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 261             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 262             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 263
 264             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 265             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 266             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 267
 268             /* Load parameters for j particles */
 269             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 270                                                                  charge+jnrC+0,charge+jnrD+0,
 271                                                                  charge+jnrE+0,charge+jnrF+0,
 272                                                                  charge+jnrG+0,charge+jnrH+0);
 273             vdwjidx0A        = 2*vdwtype[jnrA+0];
 274             vdwjidx0B        = 2*vdwtype[jnrB+0];
 275             vdwjidx0C        = 2*vdwtype[jnrC+0];
 276             vdwjidx0D        = 2*vdwtype[jnrD+0];
 277             vdwjidx0E        = 2*vdwtype[jnrE+0];
 278             vdwjidx0F        = 2*vdwtype[jnrF+0];
 279             vdwjidx0G        = 2*vdwtype[jnrG+0];
 280             vdwjidx0H        = 2*vdwtype[jnrH+0];
 281
 282             fjx0             = _mm256_setzero_ps();
 283             fjy0             = _mm256_setzero_ps();
 284             fjz0             = _mm256_setzero_ps();
 285
 286             /**************************
 287              * CALCULATE INTERACTIONS *
 288              **************************/
 289
 290             r00              = _mm256_mul_ps(rsq00,rinv00);
 291
 292             /* Compute parameters for interactions between i and j atoms */
 293             qq00             = _mm256_mul_ps(iq0,jq0);
 294             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 295                                             vdwioffsetptr0+vdwjidx0B,
 296                                             vdwioffsetptr0+vdwjidx0C,
 297                                             vdwioffsetptr0+vdwjidx0D,
 298                                             vdwioffsetptr0+vdwjidx0E,
 299                                             vdwioffsetptr0+vdwjidx0F,
 300                                             vdwioffsetptr0+vdwjidx0G,
 301                                             vdwioffsetptr0+vdwjidx0H,
 302                                             &c6_00,&c12_00);
 303
 304             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 305                                                                   vdwgridioffsetptr0+vdwjidx0B,
 306                                                                   vdwgridioffsetptr0+vdwjidx0C,
 307                                                                   vdwgridioffsetptr0+vdwjidx0D,
 308                                                                   vdwgridioffsetptr0+vdwjidx0E,
 309                                                                   vdwgridioffsetptr0+vdwjidx0F,
 310                                                                   vdwgridioffsetptr0+vdwjidx0G,
 311                                                                   vdwgridioffsetptr0+vdwjidx0H);
 312
 313             /* EWALD ELECTROSTATICS */
 314
 315             /* Analytical PME correction */
 316             zeta2            = _mm256_mul_ps(beta2,rsq00);
 317             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 318             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 319             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 320             felec            = _mm256_mul_ps(qq00,felec);
 321             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 322             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 323             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 324             velec            = _mm256_mul_ps(qq00,velec);
 325
 326             /* Analytical LJ-PME */
 327             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 328             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 329             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 330             exponent         = gmx_simd_exp_r(ewcljrsq);
 331             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 332             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 333             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 334             vvdw6            = _mm256_mul_ps(_mm256_sub_ps(c6_00,_mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly))),rinvsix);
 335             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 336             vvdw             = _mm256_sub_ps(_mm256_mul_ps(vvdw12,one_twelfth),_mm256_mul_ps(vvdw6,one_sixth));
 337             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 338             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,_mm256_sub_ps(vvdw6,_mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6)))),rinvsq00);
 339
 340             /* Update potential sum for this i atom from the interaction with this j atom. */
 341             velecsum         = _mm256_add_ps(velecsum,velec);
 342             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 343
 344             fscal            = _mm256_add_ps(felec,fvdw);
 345
 346             /* Calculate temporary vectorial force */
 347             tx               = _mm256_mul_ps(fscal,dx00);
 348             ty               = _mm256_mul_ps(fscal,dy00);
 349             tz               = _mm256_mul_ps(fscal,dz00);
 350
 351             /* Update vectorial force */
 352             fix0             = _mm256_add_ps(fix0,tx);
 353             fiy0             = _mm256_add_ps(fiy0,ty);
 354             fiz0             = _mm256_add_ps(fiz0,tz);
 355
 356             fjx0             = _mm256_add_ps(fjx0,tx);
 357             fjy0             = _mm256_add_ps(fjy0,ty);
 358             fjz0             = _mm256_add_ps(fjz0,tz);
 359
 360             /**************************
 361              * CALCULATE INTERACTIONS *
 362              **************************/
 363
 364             r10              = _mm256_mul_ps(rsq10,rinv10);
 365
 366             /* Compute parameters for interactions between i and j atoms */
 367             qq10             = _mm256_mul_ps(iq1,jq0);
 368
 369             /* EWALD ELECTROSTATICS */
 370
 371             /* Analytical PME correction */
 372             zeta2            = _mm256_mul_ps(beta2,rsq10);
 373             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 374             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 375             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 376             felec            = _mm256_mul_ps(qq10,felec);
 377             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 378             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 379             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 380             velec            = _mm256_mul_ps(qq10,velec);
 381
 382             /* Update potential sum for this i atom from the interaction with this j atom. */
 383             velecsum         = _mm256_add_ps(velecsum,velec);
 384
 385             fscal            = felec;
 386
 387             /* Calculate temporary vectorial force */
 388             tx               = _mm256_mul_ps(fscal,dx10);
 389             ty               = _mm256_mul_ps(fscal,dy10);
 390             tz               = _mm256_mul_ps(fscal,dz10);
 391
 392             /* Update vectorial force */
 393             fix1             = _mm256_add_ps(fix1,tx);
 394             fiy1             = _mm256_add_ps(fiy1,ty);
 395             fiz1             = _mm256_add_ps(fiz1,tz);
 396
 397             fjx0             = _mm256_add_ps(fjx0,tx);
 398             fjy0             = _mm256_add_ps(fjy0,ty);
 399             fjz0             = _mm256_add_ps(fjz0,tz);
 400
 401             /**************************
 402              * CALCULATE INTERACTIONS *
 403              **************************/
 404
 405             r20              = _mm256_mul_ps(rsq20,rinv20);
 406
 407             /* Compute parameters for interactions between i and j atoms */
 408             qq20             = _mm256_mul_ps(iq2,jq0);
 409
 410             /* EWALD ELECTROSTATICS */
 411
 412             /* Analytical PME correction */
 413             zeta2            = _mm256_mul_ps(beta2,rsq20);
 414             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 415             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 416             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 417             felec            = _mm256_mul_ps(qq20,felec);
 418             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 419             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 420             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 421             velec            = _mm256_mul_ps(qq20,velec);
 422
 423             /* Update potential sum for this i atom from the interaction with this j atom. */
 424             velecsum         = _mm256_add_ps(velecsum,velec);
 425
 426             fscal            = felec;
 427
 428             /* Calculate temporary vectorial force */
 429             tx               = _mm256_mul_ps(fscal,dx20);
 430             ty               = _mm256_mul_ps(fscal,dy20);
 431             tz               = _mm256_mul_ps(fscal,dz20);
 432
 433             /* Update vectorial force */
 434             fix2             = _mm256_add_ps(fix2,tx);
 435             fiy2             = _mm256_add_ps(fiy2,ty);
 436             fiz2             = _mm256_add_ps(fiz2,tz);
 437
 438             fjx0             = _mm256_add_ps(fjx0,tx);
 439             fjy0             = _mm256_add_ps(fjy0,ty);
 440             fjz0             = _mm256_add_ps(fjz0,tz);
 441
 442             fjptrA             = f+j_coord_offsetA;
 443             fjptrB             = f+j_coord_offsetB;
 444             fjptrC             = f+j_coord_offsetC;
 445             fjptrD             = f+j_coord_offsetD;
 446             fjptrE             = f+j_coord_offsetE;
 447             fjptrF             = f+j_coord_offsetF;
 448             fjptrG             = f+j_coord_offsetG;
 449             fjptrH             = f+j_coord_offsetH;
 450
 451             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 452
 453             /* Inner loop uses 283 flops */
 454         }
 455
 456         if(jidx<j_index_end)
 457         {
 458
 459             /* Get j neighbor index, and coordinate index */
 460             jnrlistA         = jjnr[jidx];
 461             jnrlistB         = jjnr[jidx+1];
 462             jnrlistC         = jjnr[jidx+2];
 463             jnrlistD         = jjnr[jidx+3];
 464             jnrlistE         = jjnr[jidx+4];
 465             jnrlistF         = jjnr[jidx+5];
 466             jnrlistG         = jjnr[jidx+6];
 467             jnrlistH         = jjnr[jidx+7];
 468             /* Sign of each element will be negative for non-real atoms.
 469              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 470              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 471              */
 472             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 473                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 474
 475             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 476             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 477             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 478             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 479             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 480             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 481             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 482             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 483             j_coord_offsetA  = DIM*jnrA;
 484             j_coord_offsetB  = DIM*jnrB;
 485             j_coord_offsetC  = DIM*jnrC;
 486             j_coord_offsetD  = DIM*jnrD;
 487             j_coord_offsetE  = DIM*jnrE;
 488             j_coord_offsetF  = DIM*jnrF;
 489             j_coord_offsetG  = DIM*jnrG;
 490             j_coord_offsetH  = DIM*jnrH;
 491
 492             /* load j atom coordinates */
 493             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 494                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 495                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 496                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 497                                                  &jx0,&jy0,&jz0);
 498
 499             /* Calculate displacement vector */
 500             dx00             = _mm256_sub_ps(ix0,jx0);
 501             dy00             = _mm256_sub_ps(iy0,jy0);
 502             dz00             = _mm256_sub_ps(iz0,jz0);
 503             dx10             = _mm256_sub_ps(ix1,jx0);
 504             dy10             = _mm256_sub_ps(iy1,jy0);
 505             dz10             = _mm256_sub_ps(iz1,jz0);
 506             dx20             = _mm256_sub_ps(ix2,jx0);
 507             dy20             = _mm256_sub_ps(iy2,jy0);
 508             dz20             = _mm256_sub_ps(iz2,jz0);
 509
 510             /* Calculate squared distance and things based on it */
 511             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 512             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 513             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 514
 515             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 516             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 517             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 518
 519             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 520             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 521             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 522
 523             /* Load parameters for j particles */
 524             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 525                                                                  charge+jnrC+0,charge+jnrD+0,
 526                                                                  charge+jnrE+0,charge+jnrF+0,
 527                                                                  charge+jnrG+0,charge+jnrH+0);
 528             vdwjidx0A        = 2*vdwtype[jnrA+0];
 529             vdwjidx0B        = 2*vdwtype[jnrB+0];
 530             vdwjidx0C        = 2*vdwtype[jnrC+0];
 531             vdwjidx0D        = 2*vdwtype[jnrD+0];
 532             vdwjidx0E        = 2*vdwtype[jnrE+0];
 533             vdwjidx0F        = 2*vdwtype[jnrF+0];
 534             vdwjidx0G        = 2*vdwtype[jnrG+0];
 535             vdwjidx0H        = 2*vdwtype[jnrH+0];
 536
 537             fjx0             = _mm256_setzero_ps();
 538             fjy0             = _mm256_setzero_ps();
 539             fjz0             = _mm256_setzero_ps();
 540
 541             /**************************
 542              * CALCULATE INTERACTIONS *
 543              **************************/
 544
 545             r00              = _mm256_mul_ps(rsq00,rinv00);
 546             r00              = _mm256_andnot_ps(dummy_mask,r00);
 547
 548             /* Compute parameters for interactions between i and j atoms */
 549             qq00             = _mm256_mul_ps(iq0,jq0);
 550             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 551                                             vdwioffsetptr0+vdwjidx0B,
 552                                             vdwioffsetptr0+vdwjidx0C,
 553                                             vdwioffsetptr0+vdwjidx0D,
 554                                             vdwioffsetptr0+vdwjidx0E,
 555                                             vdwioffsetptr0+vdwjidx0F,
 556                                             vdwioffsetptr0+vdwjidx0G,
 557                                             vdwioffsetptr0+vdwjidx0H,
 558                                             &c6_00,&c12_00);
 559
 560             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 561                                                                   vdwgridioffsetptr0+vdwjidx0B,
 562                                                                   vdwgridioffsetptr0+vdwjidx0C,
 563                                                                   vdwgridioffsetptr0+vdwjidx0D,
 564                                                                   vdwgridioffsetptr0+vdwjidx0E,
 565                                                                   vdwgridioffsetptr0+vdwjidx0F,
 566                                                                   vdwgridioffsetptr0+vdwjidx0G,
 567                                                                   vdwgridioffsetptr0+vdwjidx0H);
 568
 569             /* EWALD ELECTROSTATICS */
 570
 571             /* Analytical PME correction */
 572             zeta2            = _mm256_mul_ps(beta2,rsq00);
 573             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 574             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 575             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 576             felec            = _mm256_mul_ps(qq00,felec);
 577             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 578             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 579             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 580             velec            = _mm256_mul_ps(qq00,velec);
 581
 582             /* Analytical LJ-PME */
 583             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 584             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 585             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 586             exponent         = gmx_simd_exp_r(ewcljrsq);
 587             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 588             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 589             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 590             vvdw6            = _mm256_mul_ps(_mm256_sub_ps(c6_00,_mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly))),rinvsix);
 591             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 592             vvdw             = _mm256_sub_ps(_mm256_mul_ps(vvdw12,one_twelfth),_mm256_mul_ps(vvdw6,one_sixth));
 593             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 594             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,_mm256_sub_ps(vvdw6,_mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6)))),rinvsq00);
 595
 596             /* Update potential sum for this i atom from the interaction with this j atom. */
 597             velec            = _mm256_andnot_ps(dummy_mask,velec);
 598             velecsum         = _mm256_add_ps(velecsum,velec);
 599             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 600             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 601
 602             fscal            = _mm256_add_ps(felec,fvdw);
 603
 604             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 605
 606             /* Calculate temporary vectorial force */
 607             tx               = _mm256_mul_ps(fscal,dx00);
 608             ty               = _mm256_mul_ps(fscal,dy00);
 609             tz               = _mm256_mul_ps(fscal,dz00);
 610
 611             /* Update vectorial force */
 612             fix0             = _mm256_add_ps(fix0,tx);
 613             fiy0             = _mm256_add_ps(fiy0,ty);
 614             fiz0             = _mm256_add_ps(fiz0,tz);
 615
 616             fjx0             = _mm256_add_ps(fjx0,tx);
 617             fjy0             = _mm256_add_ps(fjy0,ty);
 618             fjz0             = _mm256_add_ps(fjz0,tz);
 619
 620             /**************************
 621              * CALCULATE INTERACTIONS *
 622              **************************/
 623
 624             r10              = _mm256_mul_ps(rsq10,rinv10);
 625             r10              = _mm256_andnot_ps(dummy_mask,r10);
 626
 627             /* Compute parameters for interactions between i and j atoms */
 628             qq10             = _mm256_mul_ps(iq1,jq0);
 629
 630             /* EWALD ELECTROSTATICS */
 631
 632             /* Analytical PME correction */
 633             zeta2            = _mm256_mul_ps(beta2,rsq10);
 634             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 635             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 636             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 637             felec            = _mm256_mul_ps(qq10,felec);
 638             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 639             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 640             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 641             velec            = _mm256_mul_ps(qq10,velec);
 642
 643             /* Update potential sum for this i atom from the interaction with this j atom. */
 644             velec            = _mm256_andnot_ps(dummy_mask,velec);
 645             velecsum         = _mm256_add_ps(velecsum,velec);
 646
 647             fscal            = felec;
 648
 649             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 650
 651             /* Calculate temporary vectorial force */
 652             tx               = _mm256_mul_ps(fscal,dx10);
 653             ty               = _mm256_mul_ps(fscal,dy10);
 654             tz               = _mm256_mul_ps(fscal,dz10);
 655
 656             /* Update vectorial force */
 657             fix1             = _mm256_add_ps(fix1,tx);
 658             fiy1             = _mm256_add_ps(fiy1,ty);
 659             fiz1             = _mm256_add_ps(fiz1,tz);
 660
 661             fjx0             = _mm256_add_ps(fjx0,tx);
 662             fjy0             = _mm256_add_ps(fjy0,ty);
 663             fjz0             = _mm256_add_ps(fjz0,tz);
 664
 665             /**************************
 666              * CALCULATE INTERACTIONS *
 667              **************************/
 668
 669             r20              = _mm256_mul_ps(rsq20,rinv20);
 670             r20              = _mm256_andnot_ps(dummy_mask,r20);
 671
 672             /* Compute parameters for interactions between i and j atoms */
 673             qq20             = _mm256_mul_ps(iq2,jq0);
 674
 675             /* EWALD ELECTROSTATICS */
 676
 677             /* Analytical PME correction */
 678             zeta2            = _mm256_mul_ps(beta2,rsq20);
 679             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 680             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 681             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 682             felec            = _mm256_mul_ps(qq20,felec);
 683             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 684             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 685             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 686             velec            = _mm256_mul_ps(qq20,velec);
 687
 688             /* Update potential sum for this i atom from the interaction with this j atom. */
 689             velec            = _mm256_andnot_ps(dummy_mask,velec);
 690             velecsum         = _mm256_add_ps(velecsum,velec);
 691
 692             fscal            = felec;
 693
 694             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 695
 696             /* Calculate temporary vectorial force */
 697             tx               = _mm256_mul_ps(fscal,dx20);
 698             ty               = _mm256_mul_ps(fscal,dy20);
 699             tz               = _mm256_mul_ps(fscal,dz20);
 700
 701             /* Update vectorial force */
 702             fix2             = _mm256_add_ps(fix2,tx);
 703             fiy2             = _mm256_add_ps(fiy2,ty);
 704             fiz2             = _mm256_add_ps(fiz2,tz);
 705
 706             fjx0             = _mm256_add_ps(fjx0,tx);
 707             fjy0             = _mm256_add_ps(fjy0,ty);
 708             fjz0             = _mm256_add_ps(fjz0,tz);
 709
 710             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 711             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 712             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 713             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 714             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 715             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 716             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 717             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 718
 719             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 720
 721             /* Inner loop uses 286 flops */
 722         }
 723
 724         /* End of innermost loop */
 725
 726         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 727                                                  f+i_coord_offset,fshift+i_shift_offset);
 728
 729         ggid                        = gid[iidx];
 730         /* Update potential energies */
 731         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 732         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 733
 734         /* Increment number of inner iterations */
 735         inneriter                  += j_index_end - j_index_start;
 736
 737         /* Outer loop uses 20 flops */
 738     }
 739
 740     /* Increment number of outer iterations */
 741     outeriter        += nri;
 742
 743     /* Update outer/inner flops */
 744
 745     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*286);
 746 }
 747 /*
 748  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_avx_256_single
 749  * Electrostatics interaction: Ewald
 750  * VdW interaction:            LJEwald
 751  * Geometry:                   Water3-Particle
 752  * Calculate force/pot:        Force
 753  */
 754 void
 755 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_avx_256_single
 756                     (t_nblist                    * gmx_restrict       nlist,
 757                      rvec                        * gmx_restrict          xx,
 758                      rvec                        * gmx_restrict          ff,
 759                      t_forcerec                  * gmx_restrict          fr,
 760                      t_mdatoms                   * gmx_restrict     mdatoms,
 761                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 762                      t_nrnb                      * gmx_restrict        nrnb)
 763 {
 764     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 765      * just 0 for non-waters.
 766      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 767      * jnr indices corresponding to data put in the four positions in the SIMD register.
 768      */
 769     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 770     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 771     int              jnrA,jnrB,jnrC,jnrD;
 772     int              jnrE,jnrF,jnrG,jnrH;
 773     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 774     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 775     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 776     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 777     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 778     real             rcutoff_scalar;
 779     real             *shiftvec,*fshift,*x,*f;
 780     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 781     real             scratch[4*DIM];
 782     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 783     real *           vdwioffsetptr0;
 784     real *           vdwgridioffsetptr0;
 785     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 786     real *           vdwioffsetptr1;
 787     real *           vdwgridioffsetptr1;
 788     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 789     real *           vdwioffsetptr2;
 790     real *           vdwgridioffsetptr2;
 791     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 792     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 793     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 794     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 795     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 796     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 797     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 798     real             *charge;
 799     int              nvdwtype;
 800     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 801     int              *vdwtype;
 802     real             *vdwparam;
 803     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 804     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 805     __m256           c6grid_00;
 806     __m256           c6grid_10;
 807     __m256           c6grid_20;
 808     real             *vdwgridparam;
 809     __m256           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 810     __m256           one_half  = _mm256_set1_ps(0.5);
 811     __m256           minus_one = _mm256_set1_ps(-1.0);
 812     __m256i          ewitab;
 813     __m128i          ewitab_lo,ewitab_hi;
 814     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 815     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 816     real             *ewtab;
 817     __m256           dummy_mask,cutoff_mask;
 818     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 819     __m256           one     = _mm256_set1_ps(1.0);
 820     __m256           two     = _mm256_set1_ps(2.0);
 821     x                = xx[0];
 822     f                = ff[0];
 823
 824     nri              = nlist->nri;
 825     iinr             = nlist->iinr;
 826     jindex           = nlist->jindex;
 827     jjnr             = nlist->jjnr;
 828     shiftidx         = nlist->shift;
 829     gid              = nlist->gid;
 830     shiftvec         = fr->shift_vec[0];
 831     fshift           = fr->fshift[0];
 832     facel            = _mm256_set1_ps(fr->epsfac);
 833     charge           = mdatoms->chargeA;
 834     nvdwtype         = fr->ntype;
 835     vdwparam         = fr->nbfp;
 836     vdwtype          = mdatoms->typeA;
 837     vdwgridparam     = fr->ljpme_c6grid;
 838     sh_lj_ewald      = _mm256_set1_ps(fr->ic->sh_lj_ewald);
 839     ewclj            = _mm256_set1_ps(fr->ewaldcoeff_lj);
 840     ewclj2           = _mm256_mul_ps(minus_one,_mm256_mul_ps(ewclj,ewclj));
 841
 842     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 843     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 844     beta2            = _mm256_mul_ps(beta,beta);
 845     beta3            = _mm256_mul_ps(beta,beta2);
 846
 847     ewtab            = fr->ic->tabq_coul_F;
 848     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 849     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 850
 851     /* Setup water-specific parameters */
 852     inr              = nlist->iinr[0];
 853     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 854     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 855     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 856     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 857     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 858
 859     /* Avoid stupid compiler warnings */
 860     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 861     j_coord_offsetA = 0;
 862     j_coord_offsetB = 0;
 863     j_coord_offsetC = 0;
 864     j_coord_offsetD = 0;
 865     j_coord_offsetE = 0;
 866     j_coord_offsetF = 0;
 867     j_coord_offsetG = 0;
 868     j_coord_offsetH = 0;
 869
 870     outeriter        = 0;
 871     inneriter        = 0;
 872
 873     for(iidx=0;iidx<4*DIM;iidx++)
 874     {
 875         scratch[iidx] = 0.0;
 876     }
 877
 878     /* Start outer loop over neighborlists */
 879     for(iidx=0; iidx<nri; iidx++)
 880     {
 881         /* Load shift vector for this list */
 882         i_shift_offset   = DIM*shiftidx[iidx];
 883
 884         /* Load limits for loop over neighbors */
 885         j_index_start    = jindex[iidx];
 886         j_index_end      = jindex[iidx+1];
 887
 888         /* Get outer coordinate index */
 889         inr              = iinr[iidx];
 890         i_coord_offset   = DIM*inr;
 891
 892         /* Load i particle coords and add shift vector */
 893         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 894                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 895
 896         fix0             = _mm256_setzero_ps();
 897         fiy0             = _mm256_setzero_ps();
 898         fiz0             = _mm256_setzero_ps();
 899         fix1             = _mm256_setzero_ps();
 900         fiy1             = _mm256_setzero_ps();
 901         fiz1             = _mm256_setzero_ps();
 902         fix2             = _mm256_setzero_ps();
 903         fiy2             = _mm256_setzero_ps();
 904         fiz2             = _mm256_setzero_ps();
 905
 906         /* Start inner kernel loop */
 907         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 908         {
 909
 910             /* Get j neighbor index, and coordinate index */
 911             jnrA             = jjnr[jidx];
 912             jnrB             = jjnr[jidx+1];
 913             jnrC             = jjnr[jidx+2];
 914             jnrD             = jjnr[jidx+3];
 915             jnrE             = jjnr[jidx+4];
 916             jnrF             = jjnr[jidx+5];
 917             jnrG             = jjnr[jidx+6];
 918             jnrH             = jjnr[jidx+7];
 919             j_coord_offsetA  = DIM*jnrA;
 920             j_coord_offsetB  = DIM*jnrB;
 921             j_coord_offsetC  = DIM*jnrC;
 922             j_coord_offsetD  = DIM*jnrD;
 923             j_coord_offsetE  = DIM*jnrE;
 924             j_coord_offsetF  = DIM*jnrF;
 925             j_coord_offsetG  = DIM*jnrG;
 926             j_coord_offsetH  = DIM*jnrH;
 927
 928             /* load j atom coordinates */
 929             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 930                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 931                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 932                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 933                                                  &jx0,&jy0,&jz0);
 934
 935             /* Calculate displacement vector */
 936             dx00             = _mm256_sub_ps(ix0,jx0);
 937             dy00             = _mm256_sub_ps(iy0,jy0);
 938             dz00             = _mm256_sub_ps(iz0,jz0);
 939             dx10             = _mm256_sub_ps(ix1,jx0);
 940             dy10             = _mm256_sub_ps(iy1,jy0);
 941             dz10             = _mm256_sub_ps(iz1,jz0);
 942             dx20             = _mm256_sub_ps(ix2,jx0);
 943             dy20             = _mm256_sub_ps(iy2,jy0);
 944             dz20             = _mm256_sub_ps(iz2,jz0);
 945
 946             /* Calculate squared distance and things based on it */
 947             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 948             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 949             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 950
 951             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 952             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 953             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 954
 955             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 956             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 957             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 958
 959             /* Load parameters for j particles */
 960             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 961                                                                  charge+jnrC+0,charge+jnrD+0,
 962                                                                  charge+jnrE+0,charge+jnrF+0,
 963                                                                  charge+jnrG+0,charge+jnrH+0);
 964             vdwjidx0A        = 2*vdwtype[jnrA+0];
 965             vdwjidx0B        = 2*vdwtype[jnrB+0];
 966             vdwjidx0C        = 2*vdwtype[jnrC+0];
 967             vdwjidx0D        = 2*vdwtype[jnrD+0];
 968             vdwjidx0E        = 2*vdwtype[jnrE+0];
 969             vdwjidx0F        = 2*vdwtype[jnrF+0];
 970             vdwjidx0G        = 2*vdwtype[jnrG+0];
 971             vdwjidx0H        = 2*vdwtype[jnrH+0];
 972
 973             fjx0             = _mm256_setzero_ps();
 974             fjy0             = _mm256_setzero_ps();
 975             fjz0             = _mm256_setzero_ps();
 976
 977             /**************************
 978              * CALCULATE INTERACTIONS *
 979              **************************/
 980
 981             r00              = _mm256_mul_ps(rsq00,rinv00);
 982
 983             /* Compute parameters for interactions between i and j atoms */
 984             qq00             = _mm256_mul_ps(iq0,jq0);
 985             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 986                                             vdwioffsetptr0+vdwjidx0B,
 987                                             vdwioffsetptr0+vdwjidx0C,
 988                                             vdwioffsetptr0+vdwjidx0D,
 989                                             vdwioffsetptr0+vdwjidx0E,
 990                                             vdwioffsetptr0+vdwjidx0F,
 991                                             vdwioffsetptr0+vdwjidx0G,
 992                                             vdwioffsetptr0+vdwjidx0H,
 993                                             &c6_00,&c12_00);
 994
 995             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 996                                                                   vdwgridioffsetptr0+vdwjidx0B,
 997                                                                   vdwgridioffsetptr0+vdwjidx0C,
 998                                                                   vdwgridioffsetptr0+vdwjidx0D,
 999                                                                   vdwgridioffsetptr0+vdwjidx0E,
1000                                                                   vdwgridioffsetptr0+vdwjidx0F,
1001                                                                   vdwgridioffsetptr0+vdwjidx0G,
1002                                                                   vdwgridioffsetptr0+vdwjidx0H);
1003
1004             /* EWALD ELECTROSTATICS */
1005
1006             /* Analytical PME correction */
1007             zeta2            = _mm256_mul_ps(beta2,rsq00);
1008             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1009             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1010             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1011             felec            = _mm256_mul_ps(qq00,felec);
1012
1013             /* Analytical LJ-PME */
1014             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1015             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
1016             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
1017             exponent         = gmx_simd_exp_r(ewcljrsq);
1018             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1019             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1020             /* f6A = 6 * C6grid * (1 - poly) */
1021             f6A              = _mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly));
1022             /* f6B = C6grid * exponent * beta^6 */
1023             f6B              = _mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6));
1024             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1025             fvdw              = _mm256_mul_ps(_mm256_add_ps(_mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),_mm256_sub_ps(c6_00,f6A)),rinvsix),f6B),rinvsq00);
1026
1027             fscal            = _mm256_add_ps(felec,fvdw);
1028
1029             /* Calculate temporary vectorial force */
1030             tx               = _mm256_mul_ps(fscal,dx00);
1031             ty               = _mm256_mul_ps(fscal,dy00);
1032             tz               = _mm256_mul_ps(fscal,dz00);
1033
1034             /* Update vectorial force */
1035             fix0             = _mm256_add_ps(fix0,tx);
1036             fiy0             = _mm256_add_ps(fiy0,ty);
1037             fiz0             = _mm256_add_ps(fiz0,tz);
1038
1039             fjx0             = _mm256_add_ps(fjx0,tx);
1040             fjy0             = _mm256_add_ps(fjy0,ty);
1041             fjz0             = _mm256_add_ps(fjz0,tz);
1042
1043             /**************************
1044              * CALCULATE INTERACTIONS *
1045              **************************/
1046
1047             r10              = _mm256_mul_ps(rsq10,rinv10);
1048
1049             /* Compute parameters for interactions between i and j atoms */
1050             qq10             = _mm256_mul_ps(iq1,jq0);
1051
1052             /* EWALD ELECTROSTATICS */
1053
1054             /* Analytical PME correction */
1055             zeta2            = _mm256_mul_ps(beta2,rsq10);
1056             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1057             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1058             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1059             felec            = _mm256_mul_ps(qq10,felec);
1060
1061             fscal            = felec;
1062
1063             /* Calculate temporary vectorial force */
1064             tx               = _mm256_mul_ps(fscal,dx10);
1065             ty               = _mm256_mul_ps(fscal,dy10);
1066             tz               = _mm256_mul_ps(fscal,dz10);
1067
1068             /* Update vectorial force */
1069             fix1             = _mm256_add_ps(fix1,tx);
1070             fiy1             = _mm256_add_ps(fiy1,ty);
1071             fiz1             = _mm256_add_ps(fiz1,tz);
1072
1073             fjx0             = _mm256_add_ps(fjx0,tx);
1074             fjy0             = _mm256_add_ps(fjy0,ty);
1075             fjz0             = _mm256_add_ps(fjz0,tz);
1076
1077             /**************************
1078              * CALCULATE INTERACTIONS *
1079              **************************/
1080
1081             r20              = _mm256_mul_ps(rsq20,rinv20);
1082
1083             /* Compute parameters for interactions between i and j atoms */
1084             qq20             = _mm256_mul_ps(iq2,jq0);
1085
1086             /* EWALD ELECTROSTATICS */
1087
1088             /* Analytical PME correction */
1089             zeta2            = _mm256_mul_ps(beta2,rsq20);
1090             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1091             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1092             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1093             felec            = _mm256_mul_ps(qq20,felec);
1094
1095             fscal            = felec;
1096
1097             /* Calculate temporary vectorial force */
1098             tx               = _mm256_mul_ps(fscal,dx20);
1099             ty               = _mm256_mul_ps(fscal,dy20);
1100             tz               = _mm256_mul_ps(fscal,dz20);
1101
1102             /* Update vectorial force */
1103             fix2             = _mm256_add_ps(fix2,tx);
1104             fiy2             = _mm256_add_ps(fiy2,ty);
1105             fiz2             = _mm256_add_ps(fiz2,tz);
1106
1107             fjx0             = _mm256_add_ps(fjx0,tx);
1108             fjy0             = _mm256_add_ps(fjy0,ty);
1109             fjz0             = _mm256_add_ps(fjz0,tz);
1110
1111             fjptrA             = f+j_coord_offsetA;
1112             fjptrB             = f+j_coord_offsetB;
1113             fjptrC             = f+j_coord_offsetC;
1114             fjptrD             = f+j_coord_offsetD;
1115             fjptrE             = f+j_coord_offsetE;
1116             fjptrF             = f+j_coord_offsetF;
1117             fjptrG             = f+j_coord_offsetG;
1118             fjptrH             = f+j_coord_offsetH;
1119
1120             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1121
1122             /* Inner loop uses 194 flops */
1123         }
1124
1125         if(jidx<j_index_end)
1126         {
1127
1128             /* Get j neighbor index, and coordinate index */
1129             jnrlistA         = jjnr[jidx];
1130             jnrlistB         = jjnr[jidx+1];
1131             jnrlistC         = jjnr[jidx+2];
1132             jnrlistD         = jjnr[jidx+3];
1133             jnrlistE         = jjnr[jidx+4];
1134             jnrlistF         = jjnr[jidx+5];
1135             jnrlistG         = jjnr[jidx+6];
1136             jnrlistH         = jjnr[jidx+7];
1137             /* Sign of each element will be negative for non-real atoms.
1138              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1139              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1140              */
1141             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1142                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1143
1144             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1145             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1146             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1147             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1148             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1149             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1150             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1151             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1152             j_coord_offsetA  = DIM*jnrA;
1153             j_coord_offsetB  = DIM*jnrB;
1154             j_coord_offsetC  = DIM*jnrC;
1155             j_coord_offsetD  = DIM*jnrD;
1156             j_coord_offsetE  = DIM*jnrE;
1157             j_coord_offsetF  = DIM*jnrF;
1158             j_coord_offsetG  = DIM*jnrG;
1159             j_coord_offsetH  = DIM*jnrH;
1160
1161             /* load j atom coordinates */
1162             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1163                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1164                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1165                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1166                                                  &jx0,&jy0,&jz0);
1167
1168             /* Calculate displacement vector */
1169             dx00             = _mm256_sub_ps(ix0,jx0);
1170             dy00             = _mm256_sub_ps(iy0,jy0);
1171             dz00             = _mm256_sub_ps(iz0,jz0);
1172             dx10             = _mm256_sub_ps(ix1,jx0);
1173             dy10             = _mm256_sub_ps(iy1,jy0);
1174             dz10             = _mm256_sub_ps(iz1,jz0);
1175             dx20             = _mm256_sub_ps(ix2,jx0);
1176             dy20             = _mm256_sub_ps(iy2,jy0);
1177             dz20             = _mm256_sub_ps(iz2,jz0);
1178
1179             /* Calculate squared distance and things based on it */
1180             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1181             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1182             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1183
1184             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1185             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1186             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1187
1188             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1189             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1190             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1191
1192             /* Load parameters for j particles */
1193             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1194                                                                  charge+jnrC+0,charge+jnrD+0,
1195                                                                  charge+jnrE+0,charge+jnrF+0,
1196                                                                  charge+jnrG+0,charge+jnrH+0);
1197             vdwjidx0A        = 2*vdwtype[jnrA+0];
1198             vdwjidx0B        = 2*vdwtype[jnrB+0];
1199             vdwjidx0C        = 2*vdwtype[jnrC+0];
1200             vdwjidx0D        = 2*vdwtype[jnrD+0];
1201             vdwjidx0E        = 2*vdwtype[jnrE+0];
1202             vdwjidx0F        = 2*vdwtype[jnrF+0];
1203             vdwjidx0G        = 2*vdwtype[jnrG+0];
1204             vdwjidx0H        = 2*vdwtype[jnrH+0];
1205
1206             fjx0             = _mm256_setzero_ps();
1207             fjy0             = _mm256_setzero_ps();
1208             fjz0             = _mm256_setzero_ps();
1209
1210             /**************************
1211              * CALCULATE INTERACTIONS *
1212              **************************/
1213
1214             r00              = _mm256_mul_ps(rsq00,rinv00);
1215             r00              = _mm256_andnot_ps(dummy_mask,r00);
1216
1217             /* Compute parameters for interactions between i and j atoms */
1218             qq00             = _mm256_mul_ps(iq0,jq0);
1219             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1220                                             vdwioffsetptr0+vdwjidx0B,
1221                                             vdwioffsetptr0+vdwjidx0C,
1222                                             vdwioffsetptr0+vdwjidx0D,
1223                                             vdwioffsetptr0+vdwjidx0E,
1224                                             vdwioffsetptr0+vdwjidx0F,
1225                                             vdwioffsetptr0+vdwjidx0G,
1226                                             vdwioffsetptr0+vdwjidx0H,
1227                                             &c6_00,&c12_00);
1228
1229             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
1230                                                                   vdwgridioffsetptr0+vdwjidx0B,
1231                                                                   vdwgridioffsetptr0+vdwjidx0C,
1232                                                                   vdwgridioffsetptr0+vdwjidx0D,
1233                                                                   vdwgridioffsetptr0+vdwjidx0E,
1234                                                                   vdwgridioffsetptr0+vdwjidx0F,
1235                                                                   vdwgridioffsetptr0+vdwjidx0G,
1236                                                                   vdwgridioffsetptr0+vdwjidx0H);
1237
1238             /* EWALD ELECTROSTATICS */
1239
1240             /* Analytical PME correction */
1241             zeta2            = _mm256_mul_ps(beta2,rsq00);
1242             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1243             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1244             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1245             felec            = _mm256_mul_ps(qq00,felec);
1246
1247             /* Analytical LJ-PME */
1248             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1249             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
1250             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
1251             exponent         = gmx_simd_exp_r(ewcljrsq);
1252             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1253             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1254             /* f6A = 6 * C6grid * (1 - poly) */
1255             f6A              = _mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly));
1256             /* f6B = C6grid * exponent * beta^6 */
1257             f6B              = _mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6));
1258             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1259             fvdw              = _mm256_mul_ps(_mm256_add_ps(_mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),_mm256_sub_ps(c6_00,f6A)),rinvsix),f6B),rinvsq00);
1260
1261             fscal            = _mm256_add_ps(felec,fvdw);
1262
1263             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1264
1265             /* Calculate temporary vectorial force */
1266             tx               = _mm256_mul_ps(fscal,dx00);
1267             ty               = _mm256_mul_ps(fscal,dy00);
1268             tz               = _mm256_mul_ps(fscal,dz00);
1269
1270             /* Update vectorial force */
1271             fix0             = _mm256_add_ps(fix0,tx);
1272             fiy0             = _mm256_add_ps(fiy0,ty);
1273             fiz0             = _mm256_add_ps(fiz0,tz);
1274
1275             fjx0             = _mm256_add_ps(fjx0,tx);
1276             fjy0             = _mm256_add_ps(fjy0,ty);
1277             fjz0             = _mm256_add_ps(fjz0,tz);
1278
1279             /**************************
1280              * CALCULATE INTERACTIONS *
1281              **************************/
1282
1283             r10              = _mm256_mul_ps(rsq10,rinv10);
1284             r10              = _mm256_andnot_ps(dummy_mask,r10);
1285
1286             /* Compute parameters for interactions between i and j atoms */
1287             qq10             = _mm256_mul_ps(iq1,jq0);
1288
1289             /* EWALD ELECTROSTATICS */
1290
1291             /* Analytical PME correction */
1292             zeta2            = _mm256_mul_ps(beta2,rsq10);
1293             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1294             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1295             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1296             felec            = _mm256_mul_ps(qq10,felec);
1297
1298             fscal            = felec;
1299
1300             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1301
1302             /* Calculate temporary vectorial force */
1303             tx               = _mm256_mul_ps(fscal,dx10);
1304             ty               = _mm256_mul_ps(fscal,dy10);
1305             tz               = _mm256_mul_ps(fscal,dz10);
1306
1307             /* Update vectorial force */
1308             fix1             = _mm256_add_ps(fix1,tx);
1309             fiy1             = _mm256_add_ps(fiy1,ty);
1310             fiz1             = _mm256_add_ps(fiz1,tz);
1311
1312             fjx0             = _mm256_add_ps(fjx0,tx);
1313             fjy0             = _mm256_add_ps(fjy0,ty);
1314             fjz0             = _mm256_add_ps(fjz0,tz);
1315
1316             /**************************
1317              * CALCULATE INTERACTIONS *
1318              **************************/
1319
1320             r20              = _mm256_mul_ps(rsq20,rinv20);
1321             r20              = _mm256_andnot_ps(dummy_mask,r20);
1322
1323             /* Compute parameters for interactions between i and j atoms */
1324             qq20             = _mm256_mul_ps(iq2,jq0);
1325
1326             /* EWALD ELECTROSTATICS */
1327
1328             /* Analytical PME correction */
1329             zeta2            = _mm256_mul_ps(beta2,rsq20);
1330             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1331             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1332             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1333             felec            = _mm256_mul_ps(qq20,felec);
1334
1335             fscal            = felec;
1336
1337             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1338
1339             /* Calculate temporary vectorial force */
1340             tx               = _mm256_mul_ps(fscal,dx20);
1341             ty               = _mm256_mul_ps(fscal,dy20);
1342             tz               = _mm256_mul_ps(fscal,dz20);
1343
1344             /* Update vectorial force */
1345             fix2             = _mm256_add_ps(fix2,tx);
1346             fiy2             = _mm256_add_ps(fiy2,ty);
1347             fiz2             = _mm256_add_ps(fiz2,tz);
1348
1349             fjx0             = _mm256_add_ps(fjx0,tx);
1350             fjy0             = _mm256_add_ps(fjy0,ty);
1351             fjz0             = _mm256_add_ps(fjz0,tz);
1352
1353             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1354             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1355             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1356             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1357             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1358             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1359             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1360             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1361
1362             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1363
1364             /* Inner loop uses 197 flops */
1365         }
1366
1367         /* End of innermost loop */
1368
1369         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1370                                                  f+i_coord_offset,fshift+i_shift_offset);
1371
1372         /* Increment number of inner iterations */
1373         inneriter                  += j_index_end - j_index_start;
1374
1375         /* Outer loop uses 18 flops */
1376     }
1377
1378     /* Increment number of outer iterations */
1379     outeriter        += nri;
1380
1381     /* Update outer/inner flops */
1382
1383     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*197);
1384 }