src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_VdwLJEw_GeomW4P1_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_ElecEw_VdwLJEw_GeomW4P1_VF_avx_256_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_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     real *           vdwgridioffsetptr0;
  88     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  89     real *           vdwioffsetptr1;
  90     real *           vdwgridioffsetptr1;
  91     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  92     real *           vdwioffsetptr2;
  93     real *           vdwgridioffsetptr2;
  94     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  95     real *           vdwioffsetptr3;
  96     real *           vdwgridioffsetptr3;
  97     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  98     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  99     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 100     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 101     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 102     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 103     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 104     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 105     real             *charge;
 106     int              nvdwtype;
 107     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 108     int              *vdwtype;
 109     real             *vdwparam;
 110     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 111     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 112     __m256           c6grid_00;
 113     __m256           c6grid_10;
 114     __m256           c6grid_20;
 115     __m256           c6grid_30;
 116     real             *vdwgridparam;
 117     __m256           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 118     __m256           one_half  = _mm256_set1_ps(0.5);
 119     __m256           minus_one = _mm256_set1_ps(-1.0);
 120     __m256i          ewitab;
 121     __m128i          ewitab_lo,ewitab_hi;
 122     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 123     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 124     real             *ewtab;
 125     __m256           dummy_mask,cutoff_mask;
 126     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 127     __m256           one     = _mm256_set1_ps(1.0);
 128     __m256           two     = _mm256_set1_ps(2.0);
 129     x                = xx[0];
 130     f                = ff[0];
 131
 132     nri              = nlist->nri;
 133     iinr             = nlist->iinr;
 134     jindex           = nlist->jindex;
 135     jjnr             = nlist->jjnr;
 136     shiftidx         = nlist->shift;
 137     gid              = nlist->gid;
 138     shiftvec         = fr->shift_vec[0];
 139     fshift           = fr->fshift[0];
 140     facel            = _mm256_set1_ps(fr->epsfac);
 141     charge           = mdatoms->chargeA;
 142     nvdwtype         = fr->ntype;
 143     vdwparam         = fr->nbfp;
 144     vdwtype          = mdatoms->typeA;
 145     vdwgridparam     = fr->ljpme_c6grid;
 146     sh_lj_ewald      = _mm256_set1_ps(fr->ic->sh_lj_ewald);
 147     ewclj            = _mm256_set1_ps(fr->ewaldcoeff_lj);
 148     ewclj2           = _mm256_mul_ps(minus_one,_mm256_mul_ps(ewclj,ewclj));
 149
 150     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 151     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 152     beta2            = _mm256_mul_ps(beta,beta);
 153     beta3            = _mm256_mul_ps(beta,beta2);
 154
 155     ewtab            = fr->ic->tabq_coul_FDV0;
 156     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 157     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 158
 159     /* Setup water-specific parameters */
 160     inr              = nlist->iinr[0];
 161     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 162     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 163     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 164     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 165     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 166
 167     /* Avoid stupid compiler warnings */
 168     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 169     j_coord_offsetA = 0;
 170     j_coord_offsetB = 0;
 171     j_coord_offsetC = 0;
 172     j_coord_offsetD = 0;
 173     j_coord_offsetE = 0;
 174     j_coord_offsetF = 0;
 175     j_coord_offsetG = 0;
 176     j_coord_offsetH = 0;
 177
 178     outeriter        = 0;
 179     inneriter        = 0;
 180
 181     for(iidx=0;iidx<4*DIM;iidx++)
 182     {
 183         scratch[iidx] = 0.0;
 184     }
 185
 186     /* Start outer loop over neighborlists */
 187     for(iidx=0; iidx<nri; iidx++)
 188     {
 189         /* Load shift vector for this list */
 190         i_shift_offset   = DIM*shiftidx[iidx];
 191
 192         /* Load limits for loop over neighbors */
 193         j_index_start    = jindex[iidx];
 194         j_index_end      = jindex[iidx+1];
 195
 196         /* Get outer coordinate index */
 197         inr              = iinr[iidx];
 198         i_coord_offset   = DIM*inr;
 199
 200         /* Load i particle coords and add shift vector */
 201         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 202                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 203
 204         fix0             = _mm256_setzero_ps();
 205         fiy0             = _mm256_setzero_ps();
 206         fiz0             = _mm256_setzero_ps();
 207         fix1             = _mm256_setzero_ps();
 208         fiy1             = _mm256_setzero_ps();
 209         fiz1             = _mm256_setzero_ps();
 210         fix2             = _mm256_setzero_ps();
 211         fiy2             = _mm256_setzero_ps();
 212         fiz2             = _mm256_setzero_ps();
 213         fix3             = _mm256_setzero_ps();
 214         fiy3             = _mm256_setzero_ps();
 215         fiz3             = _mm256_setzero_ps();
 216
 217         /* Reset potential sums */
 218         velecsum         = _mm256_setzero_ps();
 219         vvdwsum          = _mm256_setzero_ps();
 220
 221         /* Start inner kernel loop */
 222         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 223         {
 224
 225             /* Get j neighbor index, and coordinate index */
 226             jnrA             = jjnr[jidx];
 227             jnrB             = jjnr[jidx+1];
 228             jnrC             = jjnr[jidx+2];
 229             jnrD             = jjnr[jidx+3];
 230             jnrE             = jjnr[jidx+4];
 231             jnrF             = jjnr[jidx+5];
 232             jnrG             = jjnr[jidx+6];
 233             jnrH             = jjnr[jidx+7];
 234             j_coord_offsetA  = DIM*jnrA;
 235             j_coord_offsetB  = DIM*jnrB;
 236             j_coord_offsetC  = DIM*jnrC;
 237             j_coord_offsetD  = DIM*jnrD;
 238             j_coord_offsetE  = DIM*jnrE;
 239             j_coord_offsetF  = DIM*jnrF;
 240             j_coord_offsetG  = DIM*jnrG;
 241             j_coord_offsetH  = DIM*jnrH;
 242
 243             /* load j atom coordinates */
 244             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 245                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 246                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 247                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 248                                                  &jx0,&jy0,&jz0);
 249
 250             /* Calculate displacement vector */
 251             dx00             = _mm256_sub_ps(ix0,jx0);
 252             dy00             = _mm256_sub_ps(iy0,jy0);
 253             dz00             = _mm256_sub_ps(iz0,jz0);
 254             dx10             = _mm256_sub_ps(ix1,jx0);
 255             dy10             = _mm256_sub_ps(iy1,jy0);
 256             dz10             = _mm256_sub_ps(iz1,jz0);
 257             dx20             = _mm256_sub_ps(ix2,jx0);
 258             dy20             = _mm256_sub_ps(iy2,jy0);
 259             dz20             = _mm256_sub_ps(iz2,jz0);
 260             dx30             = _mm256_sub_ps(ix3,jx0);
 261             dy30             = _mm256_sub_ps(iy3,jy0);
 262             dz30             = _mm256_sub_ps(iz3,jz0);
 263
 264             /* Calculate squared distance and things based on it */
 265             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 266             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 267             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 268             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 269
 270             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 271             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 272             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 273             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 274
 275             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 276             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 277             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 278             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 279
 280             /* Load parameters for j particles */
 281             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 282                                                                  charge+jnrC+0,charge+jnrD+0,
 283                                                                  charge+jnrE+0,charge+jnrF+0,
 284                                                                  charge+jnrG+0,charge+jnrH+0);
 285             vdwjidx0A        = 2*vdwtype[jnrA+0];
 286             vdwjidx0B        = 2*vdwtype[jnrB+0];
 287             vdwjidx0C        = 2*vdwtype[jnrC+0];
 288             vdwjidx0D        = 2*vdwtype[jnrD+0];
 289             vdwjidx0E        = 2*vdwtype[jnrE+0];
 290             vdwjidx0F        = 2*vdwtype[jnrF+0];
 291             vdwjidx0G        = 2*vdwtype[jnrG+0];
 292             vdwjidx0H        = 2*vdwtype[jnrH+0];
 293
 294             fjx0             = _mm256_setzero_ps();
 295             fjy0             = _mm256_setzero_ps();
 296             fjz0             = _mm256_setzero_ps();
 297
 298             /**************************
 299              * CALCULATE INTERACTIONS *
 300              **************************/
 301
 302             r00              = _mm256_mul_ps(rsq00,rinv00);
 303
 304             /* Compute parameters for interactions between i and j atoms */
 305             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 306                                             vdwioffsetptr0+vdwjidx0B,
 307                                             vdwioffsetptr0+vdwjidx0C,
 308                                             vdwioffsetptr0+vdwjidx0D,
 309                                             vdwioffsetptr0+vdwjidx0E,
 310                                             vdwioffsetptr0+vdwjidx0F,
 311                                             vdwioffsetptr0+vdwjidx0G,
 312                                             vdwioffsetptr0+vdwjidx0H,
 313                                             &c6_00,&c12_00);
 314
 315             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 316                                                                   vdwgridioffsetptr0+vdwjidx0B,
 317                                                                   vdwgridioffsetptr0+vdwjidx0C,
 318                                                                   vdwgridioffsetptr0+vdwjidx0D,
 319                                                                   vdwgridioffsetptr0+vdwjidx0E,
 320                                                                   vdwgridioffsetptr0+vdwjidx0F,
 321                                                                   vdwgridioffsetptr0+vdwjidx0G,
 322                                                                   vdwgridioffsetptr0+vdwjidx0H);
 323
 324             /* Analytical LJ-PME */
 325             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 326             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 327             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 328             exponent         = gmx_simd_exp_r(ewcljrsq);
 329             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 330             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 331             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 332             vvdw6            = _mm256_mul_ps(_mm256_sub_ps(c6_00,_mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly))),rinvsix);
 333             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 334             vvdw             = _mm256_sub_ps(_mm256_mul_ps(vvdw12,one_twelfth),_mm256_mul_ps(vvdw6,one_sixth));
 335             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 336             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);
 337
 338             /* Update potential sum for this i atom from the interaction with this j atom. */
 339             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 340
 341             fscal            = fvdw;
 342
 343             /* Calculate temporary vectorial force */
 344             tx               = _mm256_mul_ps(fscal,dx00);
 345             ty               = _mm256_mul_ps(fscal,dy00);
 346             tz               = _mm256_mul_ps(fscal,dz00);
 347
 348             /* Update vectorial force */
 349             fix0             = _mm256_add_ps(fix0,tx);
 350             fiy0             = _mm256_add_ps(fiy0,ty);
 351             fiz0             = _mm256_add_ps(fiz0,tz);
 352
 353             fjx0             = _mm256_add_ps(fjx0,tx);
 354             fjy0             = _mm256_add_ps(fjy0,ty);
 355             fjz0             = _mm256_add_ps(fjz0,tz);
 356
 357             /**************************
 358              * CALCULATE INTERACTIONS *
 359              **************************/
 360
 361             r10              = _mm256_mul_ps(rsq10,rinv10);
 362
 363             /* Compute parameters for interactions between i and j atoms */
 364             qq10             = _mm256_mul_ps(iq1,jq0);
 365
 366             /* EWALD ELECTROSTATICS */
 367
 368             /* Analytical PME correction */
 369             zeta2            = _mm256_mul_ps(beta2,rsq10);
 370             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 371             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 372             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 373             felec            = _mm256_mul_ps(qq10,felec);
 374             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 375             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 376             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 377             velec            = _mm256_mul_ps(qq10,velec);
 378
 379             /* Update potential sum for this i atom from the interaction with this j atom. */
 380             velecsum         = _mm256_add_ps(velecsum,velec);
 381
 382             fscal            = felec;
 383
 384             /* Calculate temporary vectorial force */
 385             tx               = _mm256_mul_ps(fscal,dx10);
 386             ty               = _mm256_mul_ps(fscal,dy10);
 387             tz               = _mm256_mul_ps(fscal,dz10);
 388
 389             /* Update vectorial force */
 390             fix1             = _mm256_add_ps(fix1,tx);
 391             fiy1             = _mm256_add_ps(fiy1,ty);
 392             fiz1             = _mm256_add_ps(fiz1,tz);
 393
 394             fjx0             = _mm256_add_ps(fjx0,tx);
 395             fjy0             = _mm256_add_ps(fjy0,ty);
 396             fjz0             = _mm256_add_ps(fjz0,tz);
 397
 398             /**************************
 399              * CALCULATE INTERACTIONS *
 400              **************************/
 401
 402             r20              = _mm256_mul_ps(rsq20,rinv20);
 403
 404             /* Compute parameters for interactions between i and j atoms */
 405             qq20             = _mm256_mul_ps(iq2,jq0);
 406
 407             /* EWALD ELECTROSTATICS */
 408
 409             /* Analytical PME correction */
 410             zeta2            = _mm256_mul_ps(beta2,rsq20);
 411             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 412             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 413             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 414             felec            = _mm256_mul_ps(qq20,felec);
 415             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 416             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 417             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 418             velec            = _mm256_mul_ps(qq20,velec);
 419
 420             /* Update potential sum for this i atom from the interaction with this j atom. */
 421             velecsum         = _mm256_add_ps(velecsum,velec);
 422
 423             fscal            = felec;
 424
 425             /* Calculate temporary vectorial force */
 426             tx               = _mm256_mul_ps(fscal,dx20);
 427             ty               = _mm256_mul_ps(fscal,dy20);
 428             tz               = _mm256_mul_ps(fscal,dz20);
 429
 430             /* Update vectorial force */
 431             fix2             = _mm256_add_ps(fix2,tx);
 432             fiy2             = _mm256_add_ps(fiy2,ty);
 433             fiz2             = _mm256_add_ps(fiz2,tz);
 434
 435             fjx0             = _mm256_add_ps(fjx0,tx);
 436             fjy0             = _mm256_add_ps(fjy0,ty);
 437             fjz0             = _mm256_add_ps(fjz0,tz);
 438
 439             /**************************
 440              * CALCULATE INTERACTIONS *
 441              **************************/
 442
 443             r30              = _mm256_mul_ps(rsq30,rinv30);
 444
 445             /* Compute parameters for interactions between i and j atoms */
 446             qq30             = _mm256_mul_ps(iq3,jq0);
 447
 448             /* EWALD ELECTROSTATICS */
 449
 450             /* Analytical PME correction */
 451             zeta2            = _mm256_mul_ps(beta2,rsq30);
 452             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 453             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 454             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 455             felec            = _mm256_mul_ps(qq30,felec);
 456             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 457             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 458             velec            = _mm256_sub_ps(rinv30,pmecorrV);
 459             velec            = _mm256_mul_ps(qq30,velec);
 460
 461             /* Update potential sum for this i atom from the interaction with this j atom. */
 462             velecsum         = _mm256_add_ps(velecsum,velec);
 463
 464             fscal            = felec;
 465
 466             /* Calculate temporary vectorial force */
 467             tx               = _mm256_mul_ps(fscal,dx30);
 468             ty               = _mm256_mul_ps(fscal,dy30);
 469             tz               = _mm256_mul_ps(fscal,dz30);
 470
 471             /* Update vectorial force */
 472             fix3             = _mm256_add_ps(fix3,tx);
 473             fiy3             = _mm256_add_ps(fiy3,ty);
 474             fiz3             = _mm256_add_ps(fiz3,tz);
 475
 476             fjx0             = _mm256_add_ps(fjx0,tx);
 477             fjy0             = _mm256_add_ps(fjy0,ty);
 478             fjz0             = _mm256_add_ps(fjz0,tz);
 479
 480             fjptrA             = f+j_coord_offsetA;
 481             fjptrB             = f+j_coord_offsetB;
 482             fjptrC             = f+j_coord_offsetC;
 483             fjptrD             = f+j_coord_offsetD;
 484             fjptrE             = f+j_coord_offsetE;
 485             fjptrF             = f+j_coord_offsetF;
 486             fjptrG             = f+j_coord_offsetG;
 487             fjptrH             = f+j_coord_offsetH;
 488
 489             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 490
 491             /* Inner loop uses 306 flops */
 492         }
 493
 494         if(jidx<j_index_end)
 495         {
 496
 497             /* Get j neighbor index, and coordinate index */
 498             jnrlistA         = jjnr[jidx];
 499             jnrlistB         = jjnr[jidx+1];
 500             jnrlistC         = jjnr[jidx+2];
 501             jnrlistD         = jjnr[jidx+3];
 502             jnrlistE         = jjnr[jidx+4];
 503             jnrlistF         = jjnr[jidx+5];
 504             jnrlistG         = jjnr[jidx+6];
 505             jnrlistH         = jjnr[jidx+7];
 506             /* Sign of each element will be negative for non-real atoms.
 507              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 508              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 509              */
 510             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 511                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 512
 513             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 514             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 515             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 516             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 517             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 518             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 519             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 520             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 521             j_coord_offsetA  = DIM*jnrA;
 522             j_coord_offsetB  = DIM*jnrB;
 523             j_coord_offsetC  = DIM*jnrC;
 524             j_coord_offsetD  = DIM*jnrD;
 525             j_coord_offsetE  = DIM*jnrE;
 526             j_coord_offsetF  = DIM*jnrF;
 527             j_coord_offsetG  = DIM*jnrG;
 528             j_coord_offsetH  = DIM*jnrH;
 529
 530             /* load j atom coordinates */
 531             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 532                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 533                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 534                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 535                                                  &jx0,&jy0,&jz0);
 536
 537             /* Calculate displacement vector */
 538             dx00             = _mm256_sub_ps(ix0,jx0);
 539             dy00             = _mm256_sub_ps(iy0,jy0);
 540             dz00             = _mm256_sub_ps(iz0,jz0);
 541             dx10             = _mm256_sub_ps(ix1,jx0);
 542             dy10             = _mm256_sub_ps(iy1,jy0);
 543             dz10             = _mm256_sub_ps(iz1,jz0);
 544             dx20             = _mm256_sub_ps(ix2,jx0);
 545             dy20             = _mm256_sub_ps(iy2,jy0);
 546             dz20             = _mm256_sub_ps(iz2,jz0);
 547             dx30             = _mm256_sub_ps(ix3,jx0);
 548             dy30             = _mm256_sub_ps(iy3,jy0);
 549             dz30             = _mm256_sub_ps(iz3,jz0);
 550
 551             /* Calculate squared distance and things based on it */
 552             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 553             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 554             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 555             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 556
 557             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 558             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 559             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 560             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 561
 562             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 563             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 564             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 565             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 566
 567             /* Load parameters for j particles */
 568             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 569                                                                  charge+jnrC+0,charge+jnrD+0,
 570                                                                  charge+jnrE+0,charge+jnrF+0,
 571                                                                  charge+jnrG+0,charge+jnrH+0);
 572             vdwjidx0A        = 2*vdwtype[jnrA+0];
 573             vdwjidx0B        = 2*vdwtype[jnrB+0];
 574             vdwjidx0C        = 2*vdwtype[jnrC+0];
 575             vdwjidx0D        = 2*vdwtype[jnrD+0];
 576             vdwjidx0E        = 2*vdwtype[jnrE+0];
 577             vdwjidx0F        = 2*vdwtype[jnrF+0];
 578             vdwjidx0G        = 2*vdwtype[jnrG+0];
 579             vdwjidx0H        = 2*vdwtype[jnrH+0];
 580
 581             fjx0             = _mm256_setzero_ps();
 582             fjy0             = _mm256_setzero_ps();
 583             fjz0             = _mm256_setzero_ps();
 584
 585             /**************************
 586              * CALCULATE INTERACTIONS *
 587              **************************/
 588
 589             r00              = _mm256_mul_ps(rsq00,rinv00);
 590             r00              = _mm256_andnot_ps(dummy_mask,r00);
 591
 592             /* Compute parameters for interactions between i and j atoms */
 593             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 594                                             vdwioffsetptr0+vdwjidx0B,
 595                                             vdwioffsetptr0+vdwjidx0C,
 596                                             vdwioffsetptr0+vdwjidx0D,
 597                                             vdwioffsetptr0+vdwjidx0E,
 598                                             vdwioffsetptr0+vdwjidx0F,
 599                                             vdwioffsetptr0+vdwjidx0G,
 600                                             vdwioffsetptr0+vdwjidx0H,
 601                                             &c6_00,&c12_00);
 602
 603             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 604                                                                   vdwgridioffsetptr0+vdwjidx0B,
 605                                                                   vdwgridioffsetptr0+vdwjidx0C,
 606                                                                   vdwgridioffsetptr0+vdwjidx0D,
 607                                                                   vdwgridioffsetptr0+vdwjidx0E,
 608                                                                   vdwgridioffsetptr0+vdwjidx0F,
 609                                                                   vdwgridioffsetptr0+vdwjidx0G,
 610                                                                   vdwgridioffsetptr0+vdwjidx0H);
 611
 612             /* Analytical LJ-PME */
 613             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 614             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 615             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 616             exponent         = gmx_simd_exp_r(ewcljrsq);
 617             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 618             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 619             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 620             vvdw6            = _mm256_mul_ps(_mm256_sub_ps(c6_00,_mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly))),rinvsix);
 621             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 622             vvdw             = _mm256_sub_ps(_mm256_mul_ps(vvdw12,one_twelfth),_mm256_mul_ps(vvdw6,one_sixth));
 623             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 624             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);
 625
 626             /* Update potential sum for this i atom from the interaction with this j atom. */
 627             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 628             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 629
 630             fscal            = fvdw;
 631
 632             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 633
 634             /* Calculate temporary vectorial force */
 635             tx               = _mm256_mul_ps(fscal,dx00);
 636             ty               = _mm256_mul_ps(fscal,dy00);
 637             tz               = _mm256_mul_ps(fscal,dz00);
 638
 639             /* Update vectorial force */
 640             fix0             = _mm256_add_ps(fix0,tx);
 641             fiy0             = _mm256_add_ps(fiy0,ty);
 642             fiz0             = _mm256_add_ps(fiz0,tz);
 643
 644             fjx0             = _mm256_add_ps(fjx0,tx);
 645             fjy0             = _mm256_add_ps(fjy0,ty);
 646             fjz0             = _mm256_add_ps(fjz0,tz);
 647
 648             /**************************
 649              * CALCULATE INTERACTIONS *
 650              **************************/
 651
 652             r10              = _mm256_mul_ps(rsq10,rinv10);
 653             r10              = _mm256_andnot_ps(dummy_mask,r10);
 654
 655             /* Compute parameters for interactions between i and j atoms */
 656             qq10             = _mm256_mul_ps(iq1,jq0);
 657
 658             /* EWALD ELECTROSTATICS */
 659
 660             /* Analytical PME correction */
 661             zeta2            = _mm256_mul_ps(beta2,rsq10);
 662             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 663             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 664             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 665             felec            = _mm256_mul_ps(qq10,felec);
 666             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 667             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 668             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 669             velec            = _mm256_mul_ps(qq10,velec);
 670
 671             /* Update potential sum for this i atom from the interaction with this j atom. */
 672             velec            = _mm256_andnot_ps(dummy_mask,velec);
 673             velecsum         = _mm256_add_ps(velecsum,velec);
 674
 675             fscal            = felec;
 676
 677             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 678
 679             /* Calculate temporary vectorial force */
 680             tx               = _mm256_mul_ps(fscal,dx10);
 681             ty               = _mm256_mul_ps(fscal,dy10);
 682             tz               = _mm256_mul_ps(fscal,dz10);
 683
 684             /* Update vectorial force */
 685             fix1             = _mm256_add_ps(fix1,tx);
 686             fiy1             = _mm256_add_ps(fiy1,ty);
 687             fiz1             = _mm256_add_ps(fiz1,tz);
 688
 689             fjx0             = _mm256_add_ps(fjx0,tx);
 690             fjy0             = _mm256_add_ps(fjy0,ty);
 691             fjz0             = _mm256_add_ps(fjz0,tz);
 692
 693             /**************************
 694              * CALCULATE INTERACTIONS *
 695              **************************/
 696
 697             r20              = _mm256_mul_ps(rsq20,rinv20);
 698             r20              = _mm256_andnot_ps(dummy_mask,r20);
 699
 700             /* Compute parameters for interactions between i and j atoms */
 701             qq20             = _mm256_mul_ps(iq2,jq0);
 702
 703             /* EWALD ELECTROSTATICS */
 704
 705             /* Analytical PME correction */
 706             zeta2            = _mm256_mul_ps(beta2,rsq20);
 707             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 708             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 709             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 710             felec            = _mm256_mul_ps(qq20,felec);
 711             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 712             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 713             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 714             velec            = _mm256_mul_ps(qq20,velec);
 715
 716             /* Update potential sum for this i atom from the interaction with this j atom. */
 717             velec            = _mm256_andnot_ps(dummy_mask,velec);
 718             velecsum         = _mm256_add_ps(velecsum,velec);
 719
 720             fscal            = felec;
 721
 722             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 723
 724             /* Calculate temporary vectorial force */
 725             tx               = _mm256_mul_ps(fscal,dx20);
 726             ty               = _mm256_mul_ps(fscal,dy20);
 727             tz               = _mm256_mul_ps(fscal,dz20);
 728
 729             /* Update vectorial force */
 730             fix2             = _mm256_add_ps(fix2,tx);
 731             fiy2             = _mm256_add_ps(fiy2,ty);
 732             fiz2             = _mm256_add_ps(fiz2,tz);
 733
 734             fjx0             = _mm256_add_ps(fjx0,tx);
 735             fjy0             = _mm256_add_ps(fjy0,ty);
 736             fjz0             = _mm256_add_ps(fjz0,tz);
 737
 738             /**************************
 739              * CALCULATE INTERACTIONS *
 740              **************************/
 741
 742             r30              = _mm256_mul_ps(rsq30,rinv30);
 743             r30              = _mm256_andnot_ps(dummy_mask,r30);
 744
 745             /* Compute parameters for interactions between i and j atoms */
 746             qq30             = _mm256_mul_ps(iq3,jq0);
 747
 748             /* EWALD ELECTROSTATICS */
 749
 750             /* Analytical PME correction */
 751             zeta2            = _mm256_mul_ps(beta2,rsq30);
 752             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 753             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 754             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 755             felec            = _mm256_mul_ps(qq30,felec);
 756             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 757             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 758             velec            = _mm256_sub_ps(rinv30,pmecorrV);
 759             velec            = _mm256_mul_ps(qq30,velec);
 760
 761             /* Update potential sum for this i atom from the interaction with this j atom. */
 762             velec            = _mm256_andnot_ps(dummy_mask,velec);
 763             velecsum         = _mm256_add_ps(velecsum,velec);
 764
 765             fscal            = felec;
 766
 767             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 768
 769             /* Calculate temporary vectorial force */
 770             tx               = _mm256_mul_ps(fscal,dx30);
 771             ty               = _mm256_mul_ps(fscal,dy30);
 772             tz               = _mm256_mul_ps(fscal,dz30);
 773
 774             /* Update vectorial force */
 775             fix3             = _mm256_add_ps(fix3,tx);
 776             fiy3             = _mm256_add_ps(fiy3,ty);
 777             fiz3             = _mm256_add_ps(fiz3,tz);
 778
 779             fjx0             = _mm256_add_ps(fjx0,tx);
 780             fjy0             = _mm256_add_ps(fjy0,ty);
 781             fjz0             = _mm256_add_ps(fjz0,tz);
 782
 783             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 784             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 785             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 786             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 787             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 788             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 789             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 790             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 791
 792             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 793
 794             /* Inner loop uses 310 flops */
 795         }
 796
 797         /* End of innermost loop */
 798
 799         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 800                                                  f+i_coord_offset,fshift+i_shift_offset);
 801
 802         ggid                        = gid[iidx];
 803         /* Update potential energies */
 804         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 805         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 806
 807         /* Increment number of inner iterations */
 808         inneriter                  += j_index_end - j_index_start;
 809
 810         /* Outer loop uses 26 flops */
 811     }
 812
 813     /* Increment number of outer iterations */
 814     outeriter        += nri;
 815
 816     /* Update outer/inner flops */
 817
 818     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*310);
 819 }
 820 /*
 821  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_avx_256_single
 822  * Electrostatics interaction: Ewald
 823  * VdW interaction:            LJEwald
 824  * Geometry:                   Water4-Particle
 825  * Calculate force/pot:        Force
 826  */
 827 void
 828 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_avx_256_single
 829                     (t_nblist                    * gmx_restrict       nlist,
 830                      rvec                        * gmx_restrict          xx,
 831                      rvec                        * gmx_restrict          ff,
 832                      t_forcerec                  * gmx_restrict          fr,
 833                      t_mdatoms                   * gmx_restrict     mdatoms,
 834                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 835                      t_nrnb                      * gmx_restrict        nrnb)
 836 {
 837     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 838      * just 0 for non-waters.
 839      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 840      * jnr indices corresponding to data put in the four positions in the SIMD register.
 841      */
 842     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 843     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 844     int              jnrA,jnrB,jnrC,jnrD;
 845     int              jnrE,jnrF,jnrG,jnrH;
 846     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 847     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 848     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 849     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 850     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 851     real             rcutoff_scalar;
 852     real             *shiftvec,*fshift,*x,*f;
 853     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 854     real             scratch[4*DIM];
 855     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 856     real *           vdwioffsetptr0;
 857     real *           vdwgridioffsetptr0;
 858     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 859     real *           vdwioffsetptr1;
 860     real *           vdwgridioffsetptr1;
 861     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 862     real *           vdwioffsetptr2;
 863     real *           vdwgridioffsetptr2;
 864     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 865     real *           vdwioffsetptr3;
 866     real *           vdwgridioffsetptr3;
 867     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 868     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 869     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 870     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 871     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 872     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 873     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 874     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 875     real             *charge;
 876     int              nvdwtype;
 877     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 878     int              *vdwtype;
 879     real             *vdwparam;
 880     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 881     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 882     __m256           c6grid_00;
 883     __m256           c6grid_10;
 884     __m256           c6grid_20;
 885     __m256           c6grid_30;
 886     real             *vdwgridparam;
 887     __m256           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 888     __m256           one_half  = _mm256_set1_ps(0.5);
 889     __m256           minus_one = _mm256_set1_ps(-1.0);
 890     __m256i          ewitab;
 891     __m128i          ewitab_lo,ewitab_hi;
 892     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 893     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 894     real             *ewtab;
 895     __m256           dummy_mask,cutoff_mask;
 896     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 897     __m256           one     = _mm256_set1_ps(1.0);
 898     __m256           two     = _mm256_set1_ps(2.0);
 899     x                = xx[0];
 900     f                = ff[0];
 901
 902     nri              = nlist->nri;
 903     iinr             = nlist->iinr;
 904     jindex           = nlist->jindex;
 905     jjnr             = nlist->jjnr;
 906     shiftidx         = nlist->shift;
 907     gid              = nlist->gid;
 908     shiftvec         = fr->shift_vec[0];
 909     fshift           = fr->fshift[0];
 910     facel            = _mm256_set1_ps(fr->epsfac);
 911     charge           = mdatoms->chargeA;
 912     nvdwtype         = fr->ntype;
 913     vdwparam         = fr->nbfp;
 914     vdwtype          = mdatoms->typeA;
 915     vdwgridparam     = fr->ljpme_c6grid;
 916     sh_lj_ewald      = _mm256_set1_ps(fr->ic->sh_lj_ewald);
 917     ewclj            = _mm256_set1_ps(fr->ewaldcoeff_lj);
 918     ewclj2           = _mm256_mul_ps(minus_one,_mm256_mul_ps(ewclj,ewclj));
 919
 920     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 921     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 922     beta2            = _mm256_mul_ps(beta,beta);
 923     beta3            = _mm256_mul_ps(beta,beta2);
 924
 925     ewtab            = fr->ic->tabq_coul_F;
 926     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 927     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 928
 929     /* Setup water-specific parameters */
 930     inr              = nlist->iinr[0];
 931     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 932     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 933     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 934     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 935     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 936
 937     /* Avoid stupid compiler warnings */
 938     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 939     j_coord_offsetA = 0;
 940     j_coord_offsetB = 0;
 941     j_coord_offsetC = 0;
 942     j_coord_offsetD = 0;
 943     j_coord_offsetE = 0;
 944     j_coord_offsetF = 0;
 945     j_coord_offsetG = 0;
 946     j_coord_offsetH = 0;
 947
 948     outeriter        = 0;
 949     inneriter        = 0;
 950
 951     for(iidx=0;iidx<4*DIM;iidx++)
 952     {
 953         scratch[iidx] = 0.0;
 954     }
 955
 956     /* Start outer loop over neighborlists */
 957     for(iidx=0; iidx<nri; iidx++)
 958     {
 959         /* Load shift vector for this list */
 960         i_shift_offset   = DIM*shiftidx[iidx];
 961
 962         /* Load limits for loop over neighbors */
 963         j_index_start    = jindex[iidx];
 964         j_index_end      = jindex[iidx+1];
 965
 966         /* Get outer coordinate index */
 967         inr              = iinr[iidx];
 968         i_coord_offset   = DIM*inr;
 969
 970         /* Load i particle coords and add shift vector */
 971         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 972                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 973
 974         fix0             = _mm256_setzero_ps();
 975         fiy0             = _mm256_setzero_ps();
 976         fiz0             = _mm256_setzero_ps();
 977         fix1             = _mm256_setzero_ps();
 978         fiy1             = _mm256_setzero_ps();
 979         fiz1             = _mm256_setzero_ps();
 980         fix2             = _mm256_setzero_ps();
 981         fiy2             = _mm256_setzero_ps();
 982         fiz2             = _mm256_setzero_ps();
 983         fix3             = _mm256_setzero_ps();
 984         fiy3             = _mm256_setzero_ps();
 985         fiz3             = _mm256_setzero_ps();
 986
 987         /* Start inner kernel loop */
 988         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 989         {
 990
 991             /* Get j neighbor index, and coordinate index */
 992             jnrA             = jjnr[jidx];
 993             jnrB             = jjnr[jidx+1];
 994             jnrC             = jjnr[jidx+2];
 995             jnrD             = jjnr[jidx+3];
 996             jnrE             = jjnr[jidx+4];
 997             jnrF             = jjnr[jidx+5];
 998             jnrG             = jjnr[jidx+6];
 999             jnrH             = jjnr[jidx+7];
1000             j_coord_offsetA  = DIM*jnrA;
1001             j_coord_offsetB  = DIM*jnrB;
1002             j_coord_offsetC  = DIM*jnrC;
1003             j_coord_offsetD  = DIM*jnrD;
1004             j_coord_offsetE  = DIM*jnrE;
1005             j_coord_offsetF  = DIM*jnrF;
1006             j_coord_offsetG  = DIM*jnrG;
1007             j_coord_offsetH  = DIM*jnrH;
1008
1009             /* load j atom coordinates */
1010             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1011                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1012                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1013                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1014                                                  &jx0,&jy0,&jz0);
1015
1016             /* Calculate displacement vector */
1017             dx00             = _mm256_sub_ps(ix0,jx0);
1018             dy00             = _mm256_sub_ps(iy0,jy0);
1019             dz00             = _mm256_sub_ps(iz0,jz0);
1020             dx10             = _mm256_sub_ps(ix1,jx0);
1021             dy10             = _mm256_sub_ps(iy1,jy0);
1022             dz10             = _mm256_sub_ps(iz1,jz0);
1023             dx20             = _mm256_sub_ps(ix2,jx0);
1024             dy20             = _mm256_sub_ps(iy2,jy0);
1025             dz20             = _mm256_sub_ps(iz2,jz0);
1026             dx30             = _mm256_sub_ps(ix3,jx0);
1027             dy30             = _mm256_sub_ps(iy3,jy0);
1028             dz30             = _mm256_sub_ps(iz3,jz0);
1029
1030             /* Calculate squared distance and things based on it */
1031             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1032             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1033             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1034             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1035
1036             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1037             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1038             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1039             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1040
1041             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1042             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1043             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1044             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1045
1046             /* Load parameters for j particles */
1047             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1048                                                                  charge+jnrC+0,charge+jnrD+0,
1049                                                                  charge+jnrE+0,charge+jnrF+0,
1050                                                                  charge+jnrG+0,charge+jnrH+0);
1051             vdwjidx0A        = 2*vdwtype[jnrA+0];
1052             vdwjidx0B        = 2*vdwtype[jnrB+0];
1053             vdwjidx0C        = 2*vdwtype[jnrC+0];
1054             vdwjidx0D        = 2*vdwtype[jnrD+0];
1055             vdwjidx0E        = 2*vdwtype[jnrE+0];
1056             vdwjidx0F        = 2*vdwtype[jnrF+0];
1057             vdwjidx0G        = 2*vdwtype[jnrG+0];
1058             vdwjidx0H        = 2*vdwtype[jnrH+0];
1059
1060             fjx0             = _mm256_setzero_ps();
1061             fjy0             = _mm256_setzero_ps();
1062             fjz0             = _mm256_setzero_ps();
1063
1064             /**************************
1065              * CALCULATE INTERACTIONS *
1066              **************************/
1067
1068             r00              = _mm256_mul_ps(rsq00,rinv00);
1069
1070             /* Compute parameters for interactions between i and j atoms */
1071             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1072                                             vdwioffsetptr0+vdwjidx0B,
1073                                             vdwioffsetptr0+vdwjidx0C,
1074                                             vdwioffsetptr0+vdwjidx0D,
1075                                             vdwioffsetptr0+vdwjidx0E,
1076                                             vdwioffsetptr0+vdwjidx0F,
1077                                             vdwioffsetptr0+vdwjidx0G,
1078                                             vdwioffsetptr0+vdwjidx0H,
1079                                             &c6_00,&c12_00);
1080
1081             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
1082                                                                   vdwgridioffsetptr0+vdwjidx0B,
1083                                                                   vdwgridioffsetptr0+vdwjidx0C,
1084                                                                   vdwgridioffsetptr0+vdwjidx0D,
1085                                                                   vdwgridioffsetptr0+vdwjidx0E,
1086                                                                   vdwgridioffsetptr0+vdwjidx0F,
1087                                                                   vdwgridioffsetptr0+vdwjidx0G,
1088                                                                   vdwgridioffsetptr0+vdwjidx0H);
1089
1090             /* Analytical LJ-PME */
1091             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1092             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
1093             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
1094             exponent         = gmx_simd_exp_r(ewcljrsq);
1095             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1096             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1097             /* f6A = 6 * C6grid * (1 - poly) */
1098             f6A              = _mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly));
1099             /* f6B = C6grid * exponent * beta^6 */
1100             f6B              = _mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6));
1101             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1102             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);
1103
1104             fscal            = fvdw;
1105
1106             /* Calculate temporary vectorial force */
1107             tx               = _mm256_mul_ps(fscal,dx00);
1108             ty               = _mm256_mul_ps(fscal,dy00);
1109             tz               = _mm256_mul_ps(fscal,dz00);
1110
1111             /* Update vectorial force */
1112             fix0             = _mm256_add_ps(fix0,tx);
1113             fiy0             = _mm256_add_ps(fiy0,ty);
1114             fiz0             = _mm256_add_ps(fiz0,tz);
1115
1116             fjx0             = _mm256_add_ps(fjx0,tx);
1117             fjy0             = _mm256_add_ps(fjy0,ty);
1118             fjz0             = _mm256_add_ps(fjz0,tz);
1119
1120             /**************************
1121              * CALCULATE INTERACTIONS *
1122              **************************/
1123
1124             r10              = _mm256_mul_ps(rsq10,rinv10);
1125
1126             /* Compute parameters for interactions between i and j atoms */
1127             qq10             = _mm256_mul_ps(iq1,jq0);
1128
1129             /* EWALD ELECTROSTATICS */
1130
1131             /* Analytical PME correction */
1132             zeta2            = _mm256_mul_ps(beta2,rsq10);
1133             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1134             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1135             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1136             felec            = _mm256_mul_ps(qq10,felec);
1137
1138             fscal            = felec;
1139
1140             /* Calculate temporary vectorial force */
1141             tx               = _mm256_mul_ps(fscal,dx10);
1142             ty               = _mm256_mul_ps(fscal,dy10);
1143             tz               = _mm256_mul_ps(fscal,dz10);
1144
1145             /* Update vectorial force */
1146             fix1             = _mm256_add_ps(fix1,tx);
1147             fiy1             = _mm256_add_ps(fiy1,ty);
1148             fiz1             = _mm256_add_ps(fiz1,tz);
1149
1150             fjx0             = _mm256_add_ps(fjx0,tx);
1151             fjy0             = _mm256_add_ps(fjy0,ty);
1152             fjz0             = _mm256_add_ps(fjz0,tz);
1153
1154             /**************************
1155              * CALCULATE INTERACTIONS *
1156              **************************/
1157
1158             r20              = _mm256_mul_ps(rsq20,rinv20);
1159
1160             /* Compute parameters for interactions between i and j atoms */
1161             qq20             = _mm256_mul_ps(iq2,jq0);
1162
1163             /* EWALD ELECTROSTATICS */
1164
1165             /* Analytical PME correction */
1166             zeta2            = _mm256_mul_ps(beta2,rsq20);
1167             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1168             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1169             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1170             felec            = _mm256_mul_ps(qq20,felec);
1171
1172             fscal            = felec;
1173
1174             /* Calculate temporary vectorial force */
1175             tx               = _mm256_mul_ps(fscal,dx20);
1176             ty               = _mm256_mul_ps(fscal,dy20);
1177             tz               = _mm256_mul_ps(fscal,dz20);
1178
1179             /* Update vectorial force */
1180             fix2             = _mm256_add_ps(fix2,tx);
1181             fiy2             = _mm256_add_ps(fiy2,ty);
1182             fiz2             = _mm256_add_ps(fiz2,tz);
1183
1184             fjx0             = _mm256_add_ps(fjx0,tx);
1185             fjy0             = _mm256_add_ps(fjy0,ty);
1186             fjz0             = _mm256_add_ps(fjz0,tz);
1187
1188             /**************************
1189              * CALCULATE INTERACTIONS *
1190              **************************/
1191
1192             r30              = _mm256_mul_ps(rsq30,rinv30);
1193
1194             /* Compute parameters for interactions between i and j atoms */
1195             qq30             = _mm256_mul_ps(iq3,jq0);
1196
1197             /* EWALD ELECTROSTATICS */
1198
1199             /* Analytical PME correction */
1200             zeta2            = _mm256_mul_ps(beta2,rsq30);
1201             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1202             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1203             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1204             felec            = _mm256_mul_ps(qq30,felec);
1205
1206             fscal            = felec;
1207
1208             /* Calculate temporary vectorial force */
1209             tx               = _mm256_mul_ps(fscal,dx30);
1210             ty               = _mm256_mul_ps(fscal,dy30);
1211             tz               = _mm256_mul_ps(fscal,dz30);
1212
1213             /* Update vectorial force */
1214             fix3             = _mm256_add_ps(fix3,tx);
1215             fiy3             = _mm256_add_ps(fiy3,ty);
1216             fiz3             = _mm256_add_ps(fiz3,tz);
1217
1218             fjx0             = _mm256_add_ps(fjx0,tx);
1219             fjy0             = _mm256_add_ps(fjy0,ty);
1220             fjz0             = _mm256_add_ps(fjz0,tz);
1221
1222             fjptrA             = f+j_coord_offsetA;
1223             fjptrB             = f+j_coord_offsetB;
1224             fjptrC             = f+j_coord_offsetC;
1225             fjptrD             = f+j_coord_offsetD;
1226             fjptrE             = f+j_coord_offsetE;
1227             fjptrF             = f+j_coord_offsetF;
1228             fjptrG             = f+j_coord_offsetG;
1229             fjptrH             = f+j_coord_offsetH;
1230
1231             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1232
1233             /* Inner loop uses 217 flops */
1234         }
1235
1236         if(jidx<j_index_end)
1237         {
1238
1239             /* Get j neighbor index, and coordinate index */
1240             jnrlistA         = jjnr[jidx];
1241             jnrlistB         = jjnr[jidx+1];
1242             jnrlistC         = jjnr[jidx+2];
1243             jnrlistD         = jjnr[jidx+3];
1244             jnrlistE         = jjnr[jidx+4];
1245             jnrlistF         = jjnr[jidx+5];
1246             jnrlistG         = jjnr[jidx+6];
1247             jnrlistH         = jjnr[jidx+7];
1248             /* Sign of each element will be negative for non-real atoms.
1249              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1250              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1251              */
1252             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1253                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1254
1255             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1256             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1257             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1258             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1259             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1260             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1261             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1262             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1263             j_coord_offsetA  = DIM*jnrA;
1264             j_coord_offsetB  = DIM*jnrB;
1265             j_coord_offsetC  = DIM*jnrC;
1266             j_coord_offsetD  = DIM*jnrD;
1267             j_coord_offsetE  = DIM*jnrE;
1268             j_coord_offsetF  = DIM*jnrF;
1269             j_coord_offsetG  = DIM*jnrG;
1270             j_coord_offsetH  = DIM*jnrH;
1271
1272             /* load j atom coordinates */
1273             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1274                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1275                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1276                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1277                                                  &jx0,&jy0,&jz0);
1278
1279             /* Calculate displacement vector */
1280             dx00             = _mm256_sub_ps(ix0,jx0);
1281             dy00             = _mm256_sub_ps(iy0,jy0);
1282             dz00             = _mm256_sub_ps(iz0,jz0);
1283             dx10             = _mm256_sub_ps(ix1,jx0);
1284             dy10             = _mm256_sub_ps(iy1,jy0);
1285             dz10             = _mm256_sub_ps(iz1,jz0);
1286             dx20             = _mm256_sub_ps(ix2,jx0);
1287             dy20             = _mm256_sub_ps(iy2,jy0);
1288             dz20             = _mm256_sub_ps(iz2,jz0);
1289             dx30             = _mm256_sub_ps(ix3,jx0);
1290             dy30             = _mm256_sub_ps(iy3,jy0);
1291             dz30             = _mm256_sub_ps(iz3,jz0);
1292
1293             /* Calculate squared distance and things based on it */
1294             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1295             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1296             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1297             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1298
1299             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1300             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1301             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1302             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1303
1304             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1305             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1306             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1307             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1308
1309             /* Load parameters for j particles */
1310             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1311                                                                  charge+jnrC+0,charge+jnrD+0,
1312                                                                  charge+jnrE+0,charge+jnrF+0,
1313                                                                  charge+jnrG+0,charge+jnrH+0);
1314             vdwjidx0A        = 2*vdwtype[jnrA+0];
1315             vdwjidx0B        = 2*vdwtype[jnrB+0];
1316             vdwjidx0C        = 2*vdwtype[jnrC+0];
1317             vdwjidx0D        = 2*vdwtype[jnrD+0];
1318             vdwjidx0E        = 2*vdwtype[jnrE+0];
1319             vdwjidx0F        = 2*vdwtype[jnrF+0];
1320             vdwjidx0G        = 2*vdwtype[jnrG+0];
1321             vdwjidx0H        = 2*vdwtype[jnrH+0];
1322
1323             fjx0             = _mm256_setzero_ps();
1324             fjy0             = _mm256_setzero_ps();
1325             fjz0             = _mm256_setzero_ps();
1326
1327             /**************************
1328              * CALCULATE INTERACTIONS *
1329              **************************/
1330
1331             r00              = _mm256_mul_ps(rsq00,rinv00);
1332             r00              = _mm256_andnot_ps(dummy_mask,r00);
1333
1334             /* Compute parameters for interactions between i and j atoms */
1335             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1336                                             vdwioffsetptr0+vdwjidx0B,
1337                                             vdwioffsetptr0+vdwjidx0C,
1338                                             vdwioffsetptr0+vdwjidx0D,
1339                                             vdwioffsetptr0+vdwjidx0E,
1340                                             vdwioffsetptr0+vdwjidx0F,
1341                                             vdwioffsetptr0+vdwjidx0G,
1342                                             vdwioffsetptr0+vdwjidx0H,
1343                                             &c6_00,&c12_00);
1344
1345             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
1346                                                                   vdwgridioffsetptr0+vdwjidx0B,
1347                                                                   vdwgridioffsetptr0+vdwjidx0C,
1348                                                                   vdwgridioffsetptr0+vdwjidx0D,
1349                                                                   vdwgridioffsetptr0+vdwjidx0E,
1350                                                                   vdwgridioffsetptr0+vdwjidx0F,
1351                                                                   vdwgridioffsetptr0+vdwjidx0G,
1352                                                                   vdwgridioffsetptr0+vdwjidx0H);
1353
1354             /* Analytical LJ-PME */
1355             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1356             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
1357             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
1358             exponent         = gmx_simd_exp_r(ewcljrsq);
1359             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1360             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1361             /* f6A = 6 * C6grid * (1 - poly) */
1362             f6A              = _mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly));
1363             /* f6B = C6grid * exponent * beta^6 */
1364             f6B              = _mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6));
1365             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1366             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);
1367
1368             fscal            = fvdw;
1369
1370             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1371
1372             /* Calculate temporary vectorial force */
1373             tx               = _mm256_mul_ps(fscal,dx00);
1374             ty               = _mm256_mul_ps(fscal,dy00);
1375             tz               = _mm256_mul_ps(fscal,dz00);
1376
1377             /* Update vectorial force */
1378             fix0             = _mm256_add_ps(fix0,tx);
1379             fiy0             = _mm256_add_ps(fiy0,ty);
1380             fiz0             = _mm256_add_ps(fiz0,tz);
1381
1382             fjx0             = _mm256_add_ps(fjx0,tx);
1383             fjy0             = _mm256_add_ps(fjy0,ty);
1384             fjz0             = _mm256_add_ps(fjz0,tz);
1385
1386             /**************************
1387              * CALCULATE INTERACTIONS *
1388              **************************/
1389
1390             r10              = _mm256_mul_ps(rsq10,rinv10);
1391             r10              = _mm256_andnot_ps(dummy_mask,r10);
1392
1393             /* Compute parameters for interactions between i and j atoms */
1394             qq10             = _mm256_mul_ps(iq1,jq0);
1395
1396             /* EWALD ELECTROSTATICS */
1397
1398             /* Analytical PME correction */
1399             zeta2            = _mm256_mul_ps(beta2,rsq10);
1400             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1401             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1402             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1403             felec            = _mm256_mul_ps(qq10,felec);
1404
1405             fscal            = felec;
1406
1407             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1408
1409             /* Calculate temporary vectorial force */
1410             tx               = _mm256_mul_ps(fscal,dx10);
1411             ty               = _mm256_mul_ps(fscal,dy10);
1412             tz               = _mm256_mul_ps(fscal,dz10);
1413
1414             /* Update vectorial force */
1415             fix1             = _mm256_add_ps(fix1,tx);
1416             fiy1             = _mm256_add_ps(fiy1,ty);
1417             fiz1             = _mm256_add_ps(fiz1,tz);
1418
1419             fjx0             = _mm256_add_ps(fjx0,tx);
1420             fjy0             = _mm256_add_ps(fjy0,ty);
1421             fjz0             = _mm256_add_ps(fjz0,tz);
1422
1423             /**************************
1424              * CALCULATE INTERACTIONS *
1425              **************************/
1426
1427             r20              = _mm256_mul_ps(rsq20,rinv20);
1428             r20              = _mm256_andnot_ps(dummy_mask,r20);
1429
1430             /* Compute parameters for interactions between i and j atoms */
1431             qq20             = _mm256_mul_ps(iq2,jq0);
1432
1433             /* EWALD ELECTROSTATICS */
1434
1435             /* Analytical PME correction */
1436             zeta2            = _mm256_mul_ps(beta2,rsq20);
1437             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1438             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1439             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1440             felec            = _mm256_mul_ps(qq20,felec);
1441
1442             fscal            = felec;
1443
1444             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1445
1446             /* Calculate temporary vectorial force */
1447             tx               = _mm256_mul_ps(fscal,dx20);
1448             ty               = _mm256_mul_ps(fscal,dy20);
1449             tz               = _mm256_mul_ps(fscal,dz20);
1450
1451             /* Update vectorial force */
1452             fix2             = _mm256_add_ps(fix2,tx);
1453             fiy2             = _mm256_add_ps(fiy2,ty);
1454             fiz2             = _mm256_add_ps(fiz2,tz);
1455
1456             fjx0             = _mm256_add_ps(fjx0,tx);
1457             fjy0             = _mm256_add_ps(fjy0,ty);
1458             fjz0             = _mm256_add_ps(fjz0,tz);
1459
1460             /**************************
1461              * CALCULATE INTERACTIONS *
1462              **************************/
1463
1464             r30              = _mm256_mul_ps(rsq30,rinv30);
1465             r30              = _mm256_andnot_ps(dummy_mask,r30);
1466
1467             /* Compute parameters for interactions between i and j atoms */
1468             qq30             = _mm256_mul_ps(iq3,jq0);
1469
1470             /* EWALD ELECTROSTATICS */
1471
1472             /* Analytical PME correction */
1473             zeta2            = _mm256_mul_ps(beta2,rsq30);
1474             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1475             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1476             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1477             felec            = _mm256_mul_ps(qq30,felec);
1478
1479             fscal            = felec;
1480
1481             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1482
1483             /* Calculate temporary vectorial force */
1484             tx               = _mm256_mul_ps(fscal,dx30);
1485             ty               = _mm256_mul_ps(fscal,dy30);
1486             tz               = _mm256_mul_ps(fscal,dz30);
1487
1488             /* Update vectorial force */
1489             fix3             = _mm256_add_ps(fix3,tx);
1490             fiy3             = _mm256_add_ps(fiy3,ty);
1491             fiz3             = _mm256_add_ps(fiz3,tz);
1492
1493             fjx0             = _mm256_add_ps(fjx0,tx);
1494             fjy0             = _mm256_add_ps(fjy0,ty);
1495             fjz0             = _mm256_add_ps(fjz0,tz);
1496
1497             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1498             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1499             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1500             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1501             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1502             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1503             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1504             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1505
1506             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1507
1508             /* Inner loop uses 221 flops */
1509         }
1510
1511         /* End of innermost loop */
1512
1513         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1514                                                  f+i_coord_offset,fshift+i_shift_offset);
1515
1516         /* Increment number of inner iterations */
1517         inneriter                  += j_index_end - j_index_start;
1518
1519         /* Outer loop uses 24 flops */
1520     }
1521
1522     /* Increment number of outer iterations */
1523     outeriter        += nri;
1524
1525     /* Update outer/inner flops */
1526
1527     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*221);
1528 }