src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEwSh_VdwLJEwSh_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_ElecEwSh_VdwLJEwSh_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_ElecEwSh_VdwLJEwSh_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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 168     rcutoff_scalar   = fr->rcoulomb;
 169     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 170     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 171
 172     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 173     rvdw             = _mm256_set1_ps(fr->rvdw);
 174
 175     /* Avoid stupid compiler warnings */
 176     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 177     j_coord_offsetA = 0;
 178     j_coord_offsetB = 0;
 179     j_coord_offsetC = 0;
 180     j_coord_offsetD = 0;
 181     j_coord_offsetE = 0;
 182     j_coord_offsetF = 0;
 183     j_coord_offsetG = 0;
 184     j_coord_offsetH = 0;
 185
 186     outeriter        = 0;
 187     inneriter        = 0;
 188
 189     for(iidx=0;iidx<4*DIM;iidx++)
 190     {
 191         scratch[iidx] = 0.0;
 192     }
 193
 194     /* Start outer loop over neighborlists */
 195     for(iidx=0; iidx<nri; iidx++)
 196     {
 197         /* Load shift vector for this list */
 198         i_shift_offset   = DIM*shiftidx[iidx];
 199
 200         /* Load limits for loop over neighbors */
 201         j_index_start    = jindex[iidx];
 202         j_index_end      = jindex[iidx+1];
 203
 204         /* Get outer coordinate index */
 205         inr              = iinr[iidx];
 206         i_coord_offset   = DIM*inr;
 207
 208         /* Load i particle coords and add shift vector */
 209         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 210                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 211
 212         fix0             = _mm256_setzero_ps();
 213         fiy0             = _mm256_setzero_ps();
 214         fiz0             = _mm256_setzero_ps();
 215         fix1             = _mm256_setzero_ps();
 216         fiy1             = _mm256_setzero_ps();
 217         fiz1             = _mm256_setzero_ps();
 218         fix2             = _mm256_setzero_ps();
 219         fiy2             = _mm256_setzero_ps();
 220         fiz2             = _mm256_setzero_ps();
 221         fix3             = _mm256_setzero_ps();
 222         fiy3             = _mm256_setzero_ps();
 223         fiz3             = _mm256_setzero_ps();
 224
 225         /* Reset potential sums */
 226         velecsum         = _mm256_setzero_ps();
 227         vvdwsum          = _mm256_setzero_ps();
 228
 229         /* Start inner kernel loop */
 230         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 231         {
 232
 233             /* Get j neighbor index, and coordinate index */
 234             jnrA             = jjnr[jidx];
 235             jnrB             = jjnr[jidx+1];
 236             jnrC             = jjnr[jidx+2];
 237             jnrD             = jjnr[jidx+3];
 238             jnrE             = jjnr[jidx+4];
 239             jnrF             = jjnr[jidx+5];
 240             jnrG             = jjnr[jidx+6];
 241             jnrH             = jjnr[jidx+7];
 242             j_coord_offsetA  = DIM*jnrA;
 243             j_coord_offsetB  = DIM*jnrB;
 244             j_coord_offsetC  = DIM*jnrC;
 245             j_coord_offsetD  = DIM*jnrD;
 246             j_coord_offsetE  = DIM*jnrE;
 247             j_coord_offsetF  = DIM*jnrF;
 248             j_coord_offsetG  = DIM*jnrG;
 249             j_coord_offsetH  = DIM*jnrH;
 250
 251             /* load j atom coordinates */
 252             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 253                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 254                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 255                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 256                                                  &jx0,&jy0,&jz0);
 257
 258             /* Calculate displacement vector */
 259             dx00             = _mm256_sub_ps(ix0,jx0);
 260             dy00             = _mm256_sub_ps(iy0,jy0);
 261             dz00             = _mm256_sub_ps(iz0,jz0);
 262             dx10             = _mm256_sub_ps(ix1,jx0);
 263             dy10             = _mm256_sub_ps(iy1,jy0);
 264             dz10             = _mm256_sub_ps(iz1,jz0);
 265             dx20             = _mm256_sub_ps(ix2,jx0);
 266             dy20             = _mm256_sub_ps(iy2,jy0);
 267             dz20             = _mm256_sub_ps(iz2,jz0);
 268             dx30             = _mm256_sub_ps(ix3,jx0);
 269             dy30             = _mm256_sub_ps(iy3,jy0);
 270             dz30             = _mm256_sub_ps(iz3,jz0);
 271
 272             /* Calculate squared distance and things based on it */
 273             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 274             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 275             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 276             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 277
 278             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 279             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 280             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 281             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 282
 283             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 284             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 285             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 286             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 287
 288             /* Load parameters for j particles */
 289             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 290                                                                  charge+jnrC+0,charge+jnrD+0,
 291                                                                  charge+jnrE+0,charge+jnrF+0,
 292                                                                  charge+jnrG+0,charge+jnrH+0);
 293             vdwjidx0A        = 2*vdwtype[jnrA+0];
 294             vdwjidx0B        = 2*vdwtype[jnrB+0];
 295             vdwjidx0C        = 2*vdwtype[jnrC+0];
 296             vdwjidx0D        = 2*vdwtype[jnrD+0];
 297             vdwjidx0E        = 2*vdwtype[jnrE+0];
 298             vdwjidx0F        = 2*vdwtype[jnrF+0];
 299             vdwjidx0G        = 2*vdwtype[jnrG+0];
 300             vdwjidx0H        = 2*vdwtype[jnrH+0];
 301
 302             fjx0             = _mm256_setzero_ps();
 303             fjy0             = _mm256_setzero_ps();
 304             fjz0             = _mm256_setzero_ps();
 305
 306             /**************************
 307              * CALCULATE INTERACTIONS *
 308              **************************/
 309
 310             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 311             {
 312
 313             r00              = _mm256_mul_ps(rsq00,rinv00);
 314
 315             /* Compute parameters for interactions between i and j atoms */
 316             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 317                                             vdwioffsetptr0+vdwjidx0B,
 318                                             vdwioffsetptr0+vdwjidx0C,
 319                                             vdwioffsetptr0+vdwjidx0D,
 320                                             vdwioffsetptr0+vdwjidx0E,
 321                                             vdwioffsetptr0+vdwjidx0F,
 322                                             vdwioffsetptr0+vdwjidx0G,
 323                                             vdwioffsetptr0+vdwjidx0H,
 324                                             &c6_00,&c12_00);
 325
 326             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 327                                                                   vdwgridioffsetptr0+vdwjidx0B,
 328                                                                   vdwgridioffsetptr0+vdwjidx0C,
 329                                                                   vdwgridioffsetptr0+vdwjidx0D,
 330                                                                   vdwgridioffsetptr0+vdwjidx0E,
 331                                                                   vdwgridioffsetptr0+vdwjidx0F,
 332                                                                   vdwgridioffsetptr0+vdwjidx0G,
 333                                                                   vdwgridioffsetptr0+vdwjidx0H);
 334
 335             /* Analytical LJ-PME */
 336             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 337             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 338             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 339             exponent         = gmx_simd_exp_r(ewcljrsq);
 340             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 341             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 342             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 343             vvdw6            = _mm256_mul_ps(_mm256_sub_ps(c6_00,_mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly))),rinvsix);
 344             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 345             vvdw             = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_00,_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 346                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_add_ps(_mm256_mul_ps(c6_00,sh_vdw_invrcut6),_mm256_mul_ps(c6grid_00,sh_lj_ewald))),one_sixth));
 347             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 348             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);
 349
 350             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 351
 352             /* Update potential sum for this i atom from the interaction with this j atom. */
 353             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 354             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 355
 356             fscal            = fvdw;
 357
 358             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 359
 360             /* Calculate temporary vectorial force */
 361             tx               = _mm256_mul_ps(fscal,dx00);
 362             ty               = _mm256_mul_ps(fscal,dy00);
 363             tz               = _mm256_mul_ps(fscal,dz00);
 364
 365             /* Update vectorial force */
 366             fix0             = _mm256_add_ps(fix0,tx);
 367             fiy0             = _mm256_add_ps(fiy0,ty);
 368             fiz0             = _mm256_add_ps(fiz0,tz);
 369
 370             fjx0             = _mm256_add_ps(fjx0,tx);
 371             fjy0             = _mm256_add_ps(fjy0,ty);
 372             fjz0             = _mm256_add_ps(fjz0,tz);
 373
 374             }
 375
 376             /**************************
 377              * CALCULATE INTERACTIONS *
 378              **************************/
 379
 380             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 381             {
 382
 383             r10              = _mm256_mul_ps(rsq10,rinv10);
 384
 385             /* Compute parameters for interactions between i and j atoms */
 386             qq10             = _mm256_mul_ps(iq1,jq0);
 387
 388             /* EWALD ELECTROSTATICS */
 389
 390             /* Analytical PME correction */
 391             zeta2            = _mm256_mul_ps(beta2,rsq10);
 392             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 393             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 394             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 395             felec            = _mm256_mul_ps(qq10,felec);
 396             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 397             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 398             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 399             velec            = _mm256_mul_ps(qq10,velec);
 400
 401             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 402
 403             /* Update potential sum for this i atom from the interaction with this j atom. */
 404             velec            = _mm256_and_ps(velec,cutoff_mask);
 405             velecsum         = _mm256_add_ps(velecsum,velec);
 406
 407             fscal            = felec;
 408
 409             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 410
 411             /* Calculate temporary vectorial force */
 412             tx               = _mm256_mul_ps(fscal,dx10);
 413             ty               = _mm256_mul_ps(fscal,dy10);
 414             tz               = _mm256_mul_ps(fscal,dz10);
 415
 416             /* Update vectorial force */
 417             fix1             = _mm256_add_ps(fix1,tx);
 418             fiy1             = _mm256_add_ps(fiy1,ty);
 419             fiz1             = _mm256_add_ps(fiz1,tz);
 420
 421             fjx0             = _mm256_add_ps(fjx0,tx);
 422             fjy0             = _mm256_add_ps(fjy0,ty);
 423             fjz0             = _mm256_add_ps(fjz0,tz);
 424
 425             }
 426
 427             /**************************
 428              * CALCULATE INTERACTIONS *
 429              **************************/
 430
 431             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 432             {
 433
 434             r20              = _mm256_mul_ps(rsq20,rinv20);
 435
 436             /* Compute parameters for interactions between i and j atoms */
 437             qq20             = _mm256_mul_ps(iq2,jq0);
 438
 439             /* EWALD ELECTROSTATICS */
 440
 441             /* Analytical PME correction */
 442             zeta2            = _mm256_mul_ps(beta2,rsq20);
 443             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 444             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 445             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 446             felec            = _mm256_mul_ps(qq20,felec);
 447             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 448             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 449             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 450             velec            = _mm256_mul_ps(qq20,velec);
 451
 452             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 453
 454             /* Update potential sum for this i atom from the interaction with this j atom. */
 455             velec            = _mm256_and_ps(velec,cutoff_mask);
 456             velecsum         = _mm256_add_ps(velecsum,velec);
 457
 458             fscal            = felec;
 459
 460             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 461
 462             /* Calculate temporary vectorial force */
 463             tx               = _mm256_mul_ps(fscal,dx20);
 464             ty               = _mm256_mul_ps(fscal,dy20);
 465             tz               = _mm256_mul_ps(fscal,dz20);
 466
 467             /* Update vectorial force */
 468             fix2             = _mm256_add_ps(fix2,tx);
 469             fiy2             = _mm256_add_ps(fiy2,ty);
 470             fiz2             = _mm256_add_ps(fiz2,tz);
 471
 472             fjx0             = _mm256_add_ps(fjx0,tx);
 473             fjy0             = _mm256_add_ps(fjy0,ty);
 474             fjz0             = _mm256_add_ps(fjz0,tz);
 475
 476             }
 477
 478             /**************************
 479              * CALCULATE INTERACTIONS *
 480              **************************/
 481
 482             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 483             {
 484
 485             r30              = _mm256_mul_ps(rsq30,rinv30);
 486
 487             /* Compute parameters for interactions between i and j atoms */
 488             qq30             = _mm256_mul_ps(iq3,jq0);
 489
 490             /* EWALD ELECTROSTATICS */
 491
 492             /* Analytical PME correction */
 493             zeta2            = _mm256_mul_ps(beta2,rsq30);
 494             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 495             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 496             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 497             felec            = _mm256_mul_ps(qq30,felec);
 498             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 499             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 500             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv30,sh_ewald),pmecorrV);
 501             velec            = _mm256_mul_ps(qq30,velec);
 502
 503             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 504
 505             /* Update potential sum for this i atom from the interaction with this j atom. */
 506             velec            = _mm256_and_ps(velec,cutoff_mask);
 507             velecsum         = _mm256_add_ps(velecsum,velec);
 508
 509             fscal            = felec;
 510
 511             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 512
 513             /* Calculate temporary vectorial force */
 514             tx               = _mm256_mul_ps(fscal,dx30);
 515             ty               = _mm256_mul_ps(fscal,dy30);
 516             tz               = _mm256_mul_ps(fscal,dz30);
 517
 518             /* Update vectorial force */
 519             fix3             = _mm256_add_ps(fix3,tx);
 520             fiy3             = _mm256_add_ps(fiy3,ty);
 521             fiz3             = _mm256_add_ps(fiz3,tz);
 522
 523             fjx0             = _mm256_add_ps(fjx0,tx);
 524             fjy0             = _mm256_add_ps(fjy0,ty);
 525             fjz0             = _mm256_add_ps(fjz0,tz);
 526
 527             }
 528
 529             fjptrA             = f+j_coord_offsetA;
 530             fjptrB             = f+j_coord_offsetB;
 531             fjptrC             = f+j_coord_offsetC;
 532             fjptrD             = f+j_coord_offsetD;
 533             fjptrE             = f+j_coord_offsetE;
 534             fjptrF             = f+j_coord_offsetF;
 535             fjptrG             = f+j_coord_offsetG;
 536             fjptrH             = f+j_coord_offsetH;
 537
 538             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 539
 540             /* Inner loop uses 392 flops */
 541         }
 542
 543         if(jidx<j_index_end)
 544         {
 545
 546             /* Get j neighbor index, and coordinate index */
 547             jnrlistA         = jjnr[jidx];
 548             jnrlistB         = jjnr[jidx+1];
 549             jnrlistC         = jjnr[jidx+2];
 550             jnrlistD         = jjnr[jidx+3];
 551             jnrlistE         = jjnr[jidx+4];
 552             jnrlistF         = jjnr[jidx+5];
 553             jnrlistG         = jjnr[jidx+6];
 554             jnrlistH         = jjnr[jidx+7];
 555             /* Sign of each element will be negative for non-real atoms.
 556              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 557              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 558              */
 559             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 560                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 561
 562             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 563             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 564             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 565             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 566             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 567             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 568             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 569             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 570             j_coord_offsetA  = DIM*jnrA;
 571             j_coord_offsetB  = DIM*jnrB;
 572             j_coord_offsetC  = DIM*jnrC;
 573             j_coord_offsetD  = DIM*jnrD;
 574             j_coord_offsetE  = DIM*jnrE;
 575             j_coord_offsetF  = DIM*jnrF;
 576             j_coord_offsetG  = DIM*jnrG;
 577             j_coord_offsetH  = DIM*jnrH;
 578
 579             /* load j atom coordinates */
 580             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 581                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 582                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 583                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 584                                                  &jx0,&jy0,&jz0);
 585
 586             /* Calculate displacement vector */
 587             dx00             = _mm256_sub_ps(ix0,jx0);
 588             dy00             = _mm256_sub_ps(iy0,jy0);
 589             dz00             = _mm256_sub_ps(iz0,jz0);
 590             dx10             = _mm256_sub_ps(ix1,jx0);
 591             dy10             = _mm256_sub_ps(iy1,jy0);
 592             dz10             = _mm256_sub_ps(iz1,jz0);
 593             dx20             = _mm256_sub_ps(ix2,jx0);
 594             dy20             = _mm256_sub_ps(iy2,jy0);
 595             dz20             = _mm256_sub_ps(iz2,jz0);
 596             dx30             = _mm256_sub_ps(ix3,jx0);
 597             dy30             = _mm256_sub_ps(iy3,jy0);
 598             dz30             = _mm256_sub_ps(iz3,jz0);
 599
 600             /* Calculate squared distance and things based on it */
 601             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 602             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 603             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 604             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 605
 606             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 607             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 608             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 609             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 610
 611             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 612             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 613             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 614             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 615
 616             /* Load parameters for j particles */
 617             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 618                                                                  charge+jnrC+0,charge+jnrD+0,
 619                                                                  charge+jnrE+0,charge+jnrF+0,
 620                                                                  charge+jnrG+0,charge+jnrH+0);
 621             vdwjidx0A        = 2*vdwtype[jnrA+0];
 622             vdwjidx0B        = 2*vdwtype[jnrB+0];
 623             vdwjidx0C        = 2*vdwtype[jnrC+0];
 624             vdwjidx0D        = 2*vdwtype[jnrD+0];
 625             vdwjidx0E        = 2*vdwtype[jnrE+0];
 626             vdwjidx0F        = 2*vdwtype[jnrF+0];
 627             vdwjidx0G        = 2*vdwtype[jnrG+0];
 628             vdwjidx0H        = 2*vdwtype[jnrH+0];
 629
 630             fjx0             = _mm256_setzero_ps();
 631             fjy0             = _mm256_setzero_ps();
 632             fjz0             = _mm256_setzero_ps();
 633
 634             /**************************
 635              * CALCULATE INTERACTIONS *
 636              **************************/
 637
 638             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 639             {
 640
 641             r00              = _mm256_mul_ps(rsq00,rinv00);
 642             r00              = _mm256_andnot_ps(dummy_mask,r00);
 643
 644             /* Compute parameters for interactions between i and j atoms */
 645             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 646                                             vdwioffsetptr0+vdwjidx0B,
 647                                             vdwioffsetptr0+vdwjidx0C,
 648                                             vdwioffsetptr0+vdwjidx0D,
 649                                             vdwioffsetptr0+vdwjidx0E,
 650                                             vdwioffsetptr0+vdwjidx0F,
 651                                             vdwioffsetptr0+vdwjidx0G,
 652                                             vdwioffsetptr0+vdwjidx0H,
 653                                             &c6_00,&c12_00);
 654
 655             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 656                                                                   vdwgridioffsetptr0+vdwjidx0B,
 657                                                                   vdwgridioffsetptr0+vdwjidx0C,
 658                                                                   vdwgridioffsetptr0+vdwjidx0D,
 659                                                                   vdwgridioffsetptr0+vdwjidx0E,
 660                                                                   vdwgridioffsetptr0+vdwjidx0F,
 661                                                                   vdwgridioffsetptr0+vdwjidx0G,
 662                                                                   vdwgridioffsetptr0+vdwjidx0H);
 663
 664             /* Analytical LJ-PME */
 665             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 666             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 667             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 668             exponent         = gmx_simd_exp_r(ewcljrsq);
 669             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 670             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 671             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 672             vvdw6            = _mm256_mul_ps(_mm256_sub_ps(c6_00,_mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly))),rinvsix);
 673             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 674             vvdw             = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_00,_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 675                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_add_ps(_mm256_mul_ps(c6_00,sh_vdw_invrcut6),_mm256_mul_ps(c6grid_00,sh_lj_ewald))),one_sixth));
 676             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 677             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);
 678
 679             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 680
 681             /* Update potential sum for this i atom from the interaction with this j atom. */
 682             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 683             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 684             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 685
 686             fscal            = fvdw;
 687
 688             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 689
 690             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 691
 692             /* Calculate temporary vectorial force */
 693             tx               = _mm256_mul_ps(fscal,dx00);
 694             ty               = _mm256_mul_ps(fscal,dy00);
 695             tz               = _mm256_mul_ps(fscal,dz00);
 696
 697             /* Update vectorial force */
 698             fix0             = _mm256_add_ps(fix0,tx);
 699             fiy0             = _mm256_add_ps(fiy0,ty);
 700             fiz0             = _mm256_add_ps(fiz0,tz);
 701
 702             fjx0             = _mm256_add_ps(fjx0,tx);
 703             fjy0             = _mm256_add_ps(fjy0,ty);
 704             fjz0             = _mm256_add_ps(fjz0,tz);
 705
 706             }
 707
 708             /**************************
 709              * CALCULATE INTERACTIONS *
 710              **************************/
 711
 712             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 713             {
 714
 715             r10              = _mm256_mul_ps(rsq10,rinv10);
 716             r10              = _mm256_andnot_ps(dummy_mask,r10);
 717
 718             /* Compute parameters for interactions between i and j atoms */
 719             qq10             = _mm256_mul_ps(iq1,jq0);
 720
 721             /* EWALD ELECTROSTATICS */
 722
 723             /* Analytical PME correction */
 724             zeta2            = _mm256_mul_ps(beta2,rsq10);
 725             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 726             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 727             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 728             felec            = _mm256_mul_ps(qq10,felec);
 729             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 730             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 731             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 732             velec            = _mm256_mul_ps(qq10,velec);
 733
 734             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 735
 736             /* Update potential sum for this i atom from the interaction with this j atom. */
 737             velec            = _mm256_and_ps(velec,cutoff_mask);
 738             velec            = _mm256_andnot_ps(dummy_mask,velec);
 739             velecsum         = _mm256_add_ps(velecsum,velec);
 740
 741             fscal            = felec;
 742
 743             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 744
 745             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 746
 747             /* Calculate temporary vectorial force */
 748             tx               = _mm256_mul_ps(fscal,dx10);
 749             ty               = _mm256_mul_ps(fscal,dy10);
 750             tz               = _mm256_mul_ps(fscal,dz10);
 751
 752             /* Update vectorial force */
 753             fix1             = _mm256_add_ps(fix1,tx);
 754             fiy1             = _mm256_add_ps(fiy1,ty);
 755             fiz1             = _mm256_add_ps(fiz1,tz);
 756
 757             fjx0             = _mm256_add_ps(fjx0,tx);
 758             fjy0             = _mm256_add_ps(fjy0,ty);
 759             fjz0             = _mm256_add_ps(fjz0,tz);
 760
 761             }
 762
 763             /**************************
 764              * CALCULATE INTERACTIONS *
 765              **************************/
 766
 767             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 768             {
 769
 770             r20              = _mm256_mul_ps(rsq20,rinv20);
 771             r20              = _mm256_andnot_ps(dummy_mask,r20);
 772
 773             /* Compute parameters for interactions between i and j atoms */
 774             qq20             = _mm256_mul_ps(iq2,jq0);
 775
 776             /* EWALD ELECTROSTATICS */
 777
 778             /* Analytical PME correction */
 779             zeta2            = _mm256_mul_ps(beta2,rsq20);
 780             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 781             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 782             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 783             felec            = _mm256_mul_ps(qq20,felec);
 784             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 785             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 786             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 787             velec            = _mm256_mul_ps(qq20,velec);
 788
 789             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 790
 791             /* Update potential sum for this i atom from the interaction with this j atom. */
 792             velec            = _mm256_and_ps(velec,cutoff_mask);
 793             velec            = _mm256_andnot_ps(dummy_mask,velec);
 794             velecsum         = _mm256_add_ps(velecsum,velec);
 795
 796             fscal            = felec;
 797
 798             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 799
 800             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 801
 802             /* Calculate temporary vectorial force */
 803             tx               = _mm256_mul_ps(fscal,dx20);
 804             ty               = _mm256_mul_ps(fscal,dy20);
 805             tz               = _mm256_mul_ps(fscal,dz20);
 806
 807             /* Update vectorial force */
 808             fix2             = _mm256_add_ps(fix2,tx);
 809             fiy2             = _mm256_add_ps(fiy2,ty);
 810             fiz2             = _mm256_add_ps(fiz2,tz);
 811
 812             fjx0             = _mm256_add_ps(fjx0,tx);
 813             fjy0             = _mm256_add_ps(fjy0,ty);
 814             fjz0             = _mm256_add_ps(fjz0,tz);
 815
 816             }
 817
 818             /**************************
 819              * CALCULATE INTERACTIONS *
 820              **************************/
 821
 822             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 823             {
 824
 825             r30              = _mm256_mul_ps(rsq30,rinv30);
 826             r30              = _mm256_andnot_ps(dummy_mask,r30);
 827
 828             /* Compute parameters for interactions between i and j atoms */
 829             qq30             = _mm256_mul_ps(iq3,jq0);
 830
 831             /* EWALD ELECTROSTATICS */
 832
 833             /* Analytical PME correction */
 834             zeta2            = _mm256_mul_ps(beta2,rsq30);
 835             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 836             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 837             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 838             felec            = _mm256_mul_ps(qq30,felec);
 839             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 840             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 841             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv30,sh_ewald),pmecorrV);
 842             velec            = _mm256_mul_ps(qq30,velec);
 843
 844             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 845
 846             /* Update potential sum for this i atom from the interaction with this j atom. */
 847             velec            = _mm256_and_ps(velec,cutoff_mask);
 848             velec            = _mm256_andnot_ps(dummy_mask,velec);
 849             velecsum         = _mm256_add_ps(velecsum,velec);
 850
 851             fscal            = felec;
 852
 853             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 854
 855             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 856
 857             /* Calculate temporary vectorial force */
 858             tx               = _mm256_mul_ps(fscal,dx30);
 859             ty               = _mm256_mul_ps(fscal,dy30);
 860             tz               = _mm256_mul_ps(fscal,dz30);
 861
 862             /* Update vectorial force */
 863             fix3             = _mm256_add_ps(fix3,tx);
 864             fiy3             = _mm256_add_ps(fiy3,ty);
 865             fiz3             = _mm256_add_ps(fiz3,tz);
 866
 867             fjx0             = _mm256_add_ps(fjx0,tx);
 868             fjy0             = _mm256_add_ps(fjy0,ty);
 869             fjz0             = _mm256_add_ps(fjz0,tz);
 870
 871             }
 872
 873             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 874             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 875             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 876             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 877             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 878             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 879             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 880             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 881
 882             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 883
 884             /* Inner loop uses 396 flops */
 885         }
 886
 887         /* End of innermost loop */
 888
 889         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 890                                                  f+i_coord_offset,fshift+i_shift_offset);
 891
 892         ggid                        = gid[iidx];
 893         /* Update potential energies */
 894         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 895         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 896
 897         /* Increment number of inner iterations */
 898         inneriter                  += j_index_end - j_index_start;
 899
 900         /* Outer loop uses 26 flops */
 901     }
 902
 903     /* Increment number of outer iterations */
 904     outeriter        += nri;
 905
 906     /* Update outer/inner flops */
 907
 908     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*396);
 909 }
 910 /*
 911  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_avx_256_single
 912  * Electrostatics interaction: Ewald
 913  * VdW interaction:            LJEwald
 914  * Geometry:                   Water4-Particle
 915  * Calculate force/pot:        Force
 916  */
 917 void
 918 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_avx_256_single
 919                     (t_nblist                    * gmx_restrict       nlist,
 920                      rvec                        * gmx_restrict          xx,
 921                      rvec                        * gmx_restrict          ff,
 922                      t_forcerec                  * gmx_restrict          fr,
 923                      t_mdatoms                   * gmx_restrict     mdatoms,
 924                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 925                      t_nrnb                      * gmx_restrict        nrnb)
 926 {
 927     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 928      * just 0 for non-waters.
 929      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 930      * jnr indices corresponding to data put in the four positions in the SIMD register.
 931      */
 932     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 933     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 934     int              jnrA,jnrB,jnrC,jnrD;
 935     int              jnrE,jnrF,jnrG,jnrH;
 936     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 937     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 938     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 939     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 940     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 941     real             rcutoff_scalar;
 942     real             *shiftvec,*fshift,*x,*f;
 943     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 944     real             scratch[4*DIM];
 945     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 946     real *           vdwioffsetptr0;
 947     real *           vdwgridioffsetptr0;
 948     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 949     real *           vdwioffsetptr1;
 950     real *           vdwgridioffsetptr1;
 951     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 952     real *           vdwioffsetptr2;
 953     real *           vdwgridioffsetptr2;
 954     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 955     real *           vdwioffsetptr3;
 956     real *           vdwgridioffsetptr3;
 957     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 958     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 959     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 960     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 961     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 962     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 963     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 964     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 965     real             *charge;
 966     int              nvdwtype;
 967     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 968     int              *vdwtype;
 969     real             *vdwparam;
 970     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 971     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 972     __m256           c6grid_00;
 973     __m256           c6grid_10;
 974     __m256           c6grid_20;
 975     __m256           c6grid_30;
 976     real             *vdwgridparam;
 977     __m256           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 978     __m256           one_half  = _mm256_set1_ps(0.5);
 979     __m256           minus_one = _mm256_set1_ps(-1.0);
 980     __m256i          ewitab;
 981     __m128i          ewitab_lo,ewitab_hi;
 982     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 983     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 984     real             *ewtab;
 985     __m256           dummy_mask,cutoff_mask;
 986     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 987     __m256           one     = _mm256_set1_ps(1.0);
 988     __m256           two     = _mm256_set1_ps(2.0);
 989     x                = xx[0];
 990     f                = ff[0];
 991
 992     nri              = nlist->nri;
 993     iinr             = nlist->iinr;
 994     jindex           = nlist->jindex;
 995     jjnr             = nlist->jjnr;
 996     shiftidx         = nlist->shift;
 997     gid              = nlist->gid;
 998     shiftvec         = fr->shift_vec[0];
 999     fshift           = fr->fshift[0];
1000     facel            = _mm256_set1_ps(fr->epsfac);
1001     charge           = mdatoms->chargeA;
1002     nvdwtype         = fr->ntype;
1003     vdwparam         = fr->nbfp;
1004     vdwtype          = mdatoms->typeA;
1005     vdwgridparam     = fr->ljpme_c6grid;
1006     sh_lj_ewald      = _mm256_set1_ps(fr->ic->sh_lj_ewald);
1007     ewclj            = _mm256_set1_ps(fr->ewaldcoeff_lj);
1008     ewclj2           = _mm256_mul_ps(minus_one,_mm256_mul_ps(ewclj,ewclj));
1009
1010     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
1011     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
1012     beta2            = _mm256_mul_ps(beta,beta);
1013     beta3            = _mm256_mul_ps(beta,beta2);
1014
1015     ewtab            = fr->ic->tabq_coul_F;
1016     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
1017     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
1018
1019     /* Setup water-specific parameters */
1020     inr              = nlist->iinr[0];
1021     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1022     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1023     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
1024     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1025     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
1026
1027     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1028     rcutoff_scalar   = fr->rcoulomb;
1029     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
1030     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
1031
1032     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
1033     rvdw             = _mm256_set1_ps(fr->rvdw);
1034
1035     /* Avoid stupid compiler warnings */
1036     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1037     j_coord_offsetA = 0;
1038     j_coord_offsetB = 0;
1039     j_coord_offsetC = 0;
1040     j_coord_offsetD = 0;
1041     j_coord_offsetE = 0;
1042     j_coord_offsetF = 0;
1043     j_coord_offsetG = 0;
1044     j_coord_offsetH = 0;
1045
1046     outeriter        = 0;
1047     inneriter        = 0;
1048
1049     for(iidx=0;iidx<4*DIM;iidx++)
1050     {
1051         scratch[iidx] = 0.0;
1052     }
1053
1054     /* Start outer loop over neighborlists */
1055     for(iidx=0; iidx<nri; iidx++)
1056     {
1057         /* Load shift vector for this list */
1058         i_shift_offset   = DIM*shiftidx[iidx];
1059
1060         /* Load limits for loop over neighbors */
1061         j_index_start    = jindex[iidx];
1062         j_index_end      = jindex[iidx+1];
1063
1064         /* Get outer coordinate index */
1065         inr              = iinr[iidx];
1066         i_coord_offset   = DIM*inr;
1067
1068         /* Load i particle coords and add shift vector */
1069         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1070                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1071
1072         fix0             = _mm256_setzero_ps();
1073         fiy0             = _mm256_setzero_ps();
1074         fiz0             = _mm256_setzero_ps();
1075         fix1             = _mm256_setzero_ps();
1076         fiy1             = _mm256_setzero_ps();
1077         fiz1             = _mm256_setzero_ps();
1078         fix2             = _mm256_setzero_ps();
1079         fiy2             = _mm256_setzero_ps();
1080         fiz2             = _mm256_setzero_ps();
1081         fix3             = _mm256_setzero_ps();
1082         fiy3             = _mm256_setzero_ps();
1083         fiz3             = _mm256_setzero_ps();
1084
1085         /* Start inner kernel loop */
1086         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1087         {
1088
1089             /* Get j neighbor index, and coordinate index */
1090             jnrA             = jjnr[jidx];
1091             jnrB             = jjnr[jidx+1];
1092             jnrC             = jjnr[jidx+2];
1093             jnrD             = jjnr[jidx+3];
1094             jnrE             = jjnr[jidx+4];
1095             jnrF             = jjnr[jidx+5];
1096             jnrG             = jjnr[jidx+6];
1097             jnrH             = jjnr[jidx+7];
1098             j_coord_offsetA  = DIM*jnrA;
1099             j_coord_offsetB  = DIM*jnrB;
1100             j_coord_offsetC  = DIM*jnrC;
1101             j_coord_offsetD  = DIM*jnrD;
1102             j_coord_offsetE  = DIM*jnrE;
1103             j_coord_offsetF  = DIM*jnrF;
1104             j_coord_offsetG  = DIM*jnrG;
1105             j_coord_offsetH  = DIM*jnrH;
1106
1107             /* load j atom coordinates */
1108             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1109                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1110                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1111                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1112                                                  &jx0,&jy0,&jz0);
1113
1114             /* Calculate displacement vector */
1115             dx00             = _mm256_sub_ps(ix0,jx0);
1116             dy00             = _mm256_sub_ps(iy0,jy0);
1117             dz00             = _mm256_sub_ps(iz0,jz0);
1118             dx10             = _mm256_sub_ps(ix1,jx0);
1119             dy10             = _mm256_sub_ps(iy1,jy0);
1120             dz10             = _mm256_sub_ps(iz1,jz0);
1121             dx20             = _mm256_sub_ps(ix2,jx0);
1122             dy20             = _mm256_sub_ps(iy2,jy0);
1123             dz20             = _mm256_sub_ps(iz2,jz0);
1124             dx30             = _mm256_sub_ps(ix3,jx0);
1125             dy30             = _mm256_sub_ps(iy3,jy0);
1126             dz30             = _mm256_sub_ps(iz3,jz0);
1127
1128             /* Calculate squared distance and things based on it */
1129             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1130             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1131             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1132             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1133
1134             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1135             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1136             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1137             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1138
1139             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1140             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1141             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1142             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1143
1144             /* Load parameters for j particles */
1145             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1146                                                                  charge+jnrC+0,charge+jnrD+0,
1147                                                                  charge+jnrE+0,charge+jnrF+0,
1148                                                                  charge+jnrG+0,charge+jnrH+0);
1149             vdwjidx0A        = 2*vdwtype[jnrA+0];
1150             vdwjidx0B        = 2*vdwtype[jnrB+0];
1151             vdwjidx0C        = 2*vdwtype[jnrC+0];
1152             vdwjidx0D        = 2*vdwtype[jnrD+0];
1153             vdwjidx0E        = 2*vdwtype[jnrE+0];
1154             vdwjidx0F        = 2*vdwtype[jnrF+0];
1155             vdwjidx0G        = 2*vdwtype[jnrG+0];
1156             vdwjidx0H        = 2*vdwtype[jnrH+0];
1157
1158             fjx0             = _mm256_setzero_ps();
1159             fjy0             = _mm256_setzero_ps();
1160             fjz0             = _mm256_setzero_ps();
1161
1162             /**************************
1163              * CALCULATE INTERACTIONS *
1164              **************************/
1165
1166             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1167             {
1168
1169             r00              = _mm256_mul_ps(rsq00,rinv00);
1170
1171             /* Compute parameters for interactions between i and j atoms */
1172             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1173                                             vdwioffsetptr0+vdwjidx0B,
1174                                             vdwioffsetptr0+vdwjidx0C,
1175                                             vdwioffsetptr0+vdwjidx0D,
1176                                             vdwioffsetptr0+vdwjidx0E,
1177                                             vdwioffsetptr0+vdwjidx0F,
1178                                             vdwioffsetptr0+vdwjidx0G,
1179                                             vdwioffsetptr0+vdwjidx0H,
1180                                             &c6_00,&c12_00);
1181
1182             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
1183                                                                   vdwgridioffsetptr0+vdwjidx0B,
1184                                                                   vdwgridioffsetptr0+vdwjidx0C,
1185                                                                   vdwgridioffsetptr0+vdwjidx0D,
1186                                                                   vdwgridioffsetptr0+vdwjidx0E,
1187                                                                   vdwgridioffsetptr0+vdwjidx0F,
1188                                                                   vdwgridioffsetptr0+vdwjidx0G,
1189                                                                   vdwgridioffsetptr0+vdwjidx0H);
1190
1191             /* Analytical LJ-PME */
1192             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1193             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
1194             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
1195             exponent         = gmx_simd_exp_r(ewcljrsq);
1196             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1197             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1198             /* f6A = 6 * C6grid * (1 - poly) */
1199             f6A              = _mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly));
1200             /* f6B = C6grid * exponent * beta^6 */
1201             f6B              = _mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6));
1202             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1203             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);
1204
1205             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1206
1207             fscal            = fvdw;
1208
1209             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1210
1211             /* Calculate temporary vectorial force */
1212             tx               = _mm256_mul_ps(fscal,dx00);
1213             ty               = _mm256_mul_ps(fscal,dy00);
1214             tz               = _mm256_mul_ps(fscal,dz00);
1215
1216             /* Update vectorial force */
1217             fix0             = _mm256_add_ps(fix0,tx);
1218             fiy0             = _mm256_add_ps(fiy0,ty);
1219             fiz0             = _mm256_add_ps(fiz0,tz);
1220
1221             fjx0             = _mm256_add_ps(fjx0,tx);
1222             fjy0             = _mm256_add_ps(fjy0,ty);
1223             fjz0             = _mm256_add_ps(fjz0,tz);
1224
1225             }
1226
1227             /**************************
1228              * CALCULATE INTERACTIONS *
1229              **************************/
1230
1231             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1232             {
1233
1234             r10              = _mm256_mul_ps(rsq10,rinv10);
1235
1236             /* Compute parameters for interactions between i and j atoms */
1237             qq10             = _mm256_mul_ps(iq1,jq0);
1238
1239             /* EWALD ELECTROSTATICS */
1240
1241             /* Analytical PME correction */
1242             zeta2            = _mm256_mul_ps(beta2,rsq10);
1243             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1244             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1245             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1246             felec            = _mm256_mul_ps(qq10,felec);
1247
1248             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1249
1250             fscal            = felec;
1251
1252             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1253
1254             /* Calculate temporary vectorial force */
1255             tx               = _mm256_mul_ps(fscal,dx10);
1256             ty               = _mm256_mul_ps(fscal,dy10);
1257             tz               = _mm256_mul_ps(fscal,dz10);
1258
1259             /* Update vectorial force */
1260             fix1             = _mm256_add_ps(fix1,tx);
1261             fiy1             = _mm256_add_ps(fiy1,ty);
1262             fiz1             = _mm256_add_ps(fiz1,tz);
1263
1264             fjx0             = _mm256_add_ps(fjx0,tx);
1265             fjy0             = _mm256_add_ps(fjy0,ty);
1266             fjz0             = _mm256_add_ps(fjz0,tz);
1267
1268             }
1269
1270             /**************************
1271              * CALCULATE INTERACTIONS *
1272              **************************/
1273
1274             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1275             {
1276
1277             r20              = _mm256_mul_ps(rsq20,rinv20);
1278
1279             /* Compute parameters for interactions between i and j atoms */
1280             qq20             = _mm256_mul_ps(iq2,jq0);
1281
1282             /* EWALD ELECTROSTATICS */
1283
1284             /* Analytical PME correction */
1285             zeta2            = _mm256_mul_ps(beta2,rsq20);
1286             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1287             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1288             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1289             felec            = _mm256_mul_ps(qq20,felec);
1290
1291             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1292
1293             fscal            = felec;
1294
1295             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1296
1297             /* Calculate temporary vectorial force */
1298             tx               = _mm256_mul_ps(fscal,dx20);
1299             ty               = _mm256_mul_ps(fscal,dy20);
1300             tz               = _mm256_mul_ps(fscal,dz20);
1301
1302             /* Update vectorial force */
1303             fix2             = _mm256_add_ps(fix2,tx);
1304             fiy2             = _mm256_add_ps(fiy2,ty);
1305             fiz2             = _mm256_add_ps(fiz2,tz);
1306
1307             fjx0             = _mm256_add_ps(fjx0,tx);
1308             fjy0             = _mm256_add_ps(fjy0,ty);
1309             fjz0             = _mm256_add_ps(fjz0,tz);
1310
1311             }
1312
1313             /**************************
1314              * CALCULATE INTERACTIONS *
1315              **************************/
1316
1317             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1318             {
1319
1320             r30              = _mm256_mul_ps(rsq30,rinv30);
1321
1322             /* Compute parameters for interactions between i and j atoms */
1323             qq30             = _mm256_mul_ps(iq3,jq0);
1324
1325             /* EWALD ELECTROSTATICS */
1326
1327             /* Analytical PME correction */
1328             zeta2            = _mm256_mul_ps(beta2,rsq30);
1329             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1330             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1331             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1332             felec            = _mm256_mul_ps(qq30,felec);
1333
1334             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1335
1336             fscal            = felec;
1337
1338             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1339
1340             /* Calculate temporary vectorial force */
1341             tx               = _mm256_mul_ps(fscal,dx30);
1342             ty               = _mm256_mul_ps(fscal,dy30);
1343             tz               = _mm256_mul_ps(fscal,dz30);
1344
1345             /* Update vectorial force */
1346             fix3             = _mm256_add_ps(fix3,tx);
1347             fiy3             = _mm256_add_ps(fiy3,ty);
1348             fiz3             = _mm256_add_ps(fiz3,tz);
1349
1350             fjx0             = _mm256_add_ps(fjx0,tx);
1351             fjy0             = _mm256_add_ps(fjy0,ty);
1352             fjz0             = _mm256_add_ps(fjz0,tz);
1353
1354             }
1355
1356             fjptrA             = f+j_coord_offsetA;
1357             fjptrB             = f+j_coord_offsetB;
1358             fjptrC             = f+j_coord_offsetC;
1359             fjptrD             = f+j_coord_offsetD;
1360             fjptrE             = f+j_coord_offsetE;
1361             fjptrF             = f+j_coord_offsetF;
1362             fjptrG             = f+j_coord_offsetG;
1363             fjptrH             = f+j_coord_offsetH;
1364
1365             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1366
1367             /* Inner loop uses 229 flops */
1368         }
1369
1370         if(jidx<j_index_end)
1371         {
1372
1373             /* Get j neighbor index, and coordinate index */
1374             jnrlistA         = jjnr[jidx];
1375             jnrlistB         = jjnr[jidx+1];
1376             jnrlistC         = jjnr[jidx+2];
1377             jnrlistD         = jjnr[jidx+3];
1378             jnrlistE         = jjnr[jidx+4];
1379             jnrlistF         = jjnr[jidx+5];
1380             jnrlistG         = jjnr[jidx+6];
1381             jnrlistH         = jjnr[jidx+7];
1382             /* Sign of each element will be negative for non-real atoms.
1383              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1384              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1385              */
1386             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1387                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1388
1389             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1390             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1391             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1392             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1393             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1394             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1395             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1396             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1397             j_coord_offsetA  = DIM*jnrA;
1398             j_coord_offsetB  = DIM*jnrB;
1399             j_coord_offsetC  = DIM*jnrC;
1400             j_coord_offsetD  = DIM*jnrD;
1401             j_coord_offsetE  = DIM*jnrE;
1402             j_coord_offsetF  = DIM*jnrF;
1403             j_coord_offsetG  = DIM*jnrG;
1404             j_coord_offsetH  = DIM*jnrH;
1405
1406             /* load j atom coordinates */
1407             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1408                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1409                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1410                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1411                                                  &jx0,&jy0,&jz0);
1412
1413             /* Calculate displacement vector */
1414             dx00             = _mm256_sub_ps(ix0,jx0);
1415             dy00             = _mm256_sub_ps(iy0,jy0);
1416             dz00             = _mm256_sub_ps(iz0,jz0);
1417             dx10             = _mm256_sub_ps(ix1,jx0);
1418             dy10             = _mm256_sub_ps(iy1,jy0);
1419             dz10             = _mm256_sub_ps(iz1,jz0);
1420             dx20             = _mm256_sub_ps(ix2,jx0);
1421             dy20             = _mm256_sub_ps(iy2,jy0);
1422             dz20             = _mm256_sub_ps(iz2,jz0);
1423             dx30             = _mm256_sub_ps(ix3,jx0);
1424             dy30             = _mm256_sub_ps(iy3,jy0);
1425             dz30             = _mm256_sub_ps(iz3,jz0);
1426
1427             /* Calculate squared distance and things based on it */
1428             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1429             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1430             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1431             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1432
1433             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1434             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1435             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1436             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1437
1438             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1439             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1440             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1441             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1442
1443             /* Load parameters for j particles */
1444             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1445                                                                  charge+jnrC+0,charge+jnrD+0,
1446                                                                  charge+jnrE+0,charge+jnrF+0,
1447                                                                  charge+jnrG+0,charge+jnrH+0);
1448             vdwjidx0A        = 2*vdwtype[jnrA+0];
1449             vdwjidx0B        = 2*vdwtype[jnrB+0];
1450             vdwjidx0C        = 2*vdwtype[jnrC+0];
1451             vdwjidx0D        = 2*vdwtype[jnrD+0];
1452             vdwjidx0E        = 2*vdwtype[jnrE+0];
1453             vdwjidx0F        = 2*vdwtype[jnrF+0];
1454             vdwjidx0G        = 2*vdwtype[jnrG+0];
1455             vdwjidx0H        = 2*vdwtype[jnrH+0];
1456
1457             fjx0             = _mm256_setzero_ps();
1458             fjy0             = _mm256_setzero_ps();
1459             fjz0             = _mm256_setzero_ps();
1460
1461             /**************************
1462              * CALCULATE INTERACTIONS *
1463              **************************/
1464
1465             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1466             {
1467
1468             r00              = _mm256_mul_ps(rsq00,rinv00);
1469             r00              = _mm256_andnot_ps(dummy_mask,r00);
1470
1471             /* Compute parameters for interactions between i and j atoms */
1472             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1473                                             vdwioffsetptr0+vdwjidx0B,
1474                                             vdwioffsetptr0+vdwjidx0C,
1475                                             vdwioffsetptr0+vdwjidx0D,
1476                                             vdwioffsetptr0+vdwjidx0E,
1477                                             vdwioffsetptr0+vdwjidx0F,
1478                                             vdwioffsetptr0+vdwjidx0G,
1479                                             vdwioffsetptr0+vdwjidx0H,
1480                                             &c6_00,&c12_00);
1481
1482             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
1483                                                                   vdwgridioffsetptr0+vdwjidx0B,
1484                                                                   vdwgridioffsetptr0+vdwjidx0C,
1485                                                                   vdwgridioffsetptr0+vdwjidx0D,
1486                                                                   vdwgridioffsetptr0+vdwjidx0E,
1487                                                                   vdwgridioffsetptr0+vdwjidx0F,
1488                                                                   vdwgridioffsetptr0+vdwjidx0G,
1489                                                                   vdwgridioffsetptr0+vdwjidx0H);
1490
1491             /* Analytical LJ-PME */
1492             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1493             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
1494             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
1495             exponent         = gmx_simd_exp_r(ewcljrsq);
1496             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1497             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1498             /* f6A = 6 * C6grid * (1 - poly) */
1499             f6A              = _mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly));
1500             /* f6B = C6grid * exponent * beta^6 */
1501             f6B              = _mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6));
1502             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1503             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);
1504
1505             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1506
1507             fscal            = fvdw;
1508
1509             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1510
1511             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1512
1513             /* Calculate temporary vectorial force */
1514             tx               = _mm256_mul_ps(fscal,dx00);
1515             ty               = _mm256_mul_ps(fscal,dy00);
1516             tz               = _mm256_mul_ps(fscal,dz00);
1517
1518             /* Update vectorial force */
1519             fix0             = _mm256_add_ps(fix0,tx);
1520             fiy0             = _mm256_add_ps(fiy0,ty);
1521             fiz0             = _mm256_add_ps(fiz0,tz);
1522
1523             fjx0             = _mm256_add_ps(fjx0,tx);
1524             fjy0             = _mm256_add_ps(fjy0,ty);
1525             fjz0             = _mm256_add_ps(fjz0,tz);
1526
1527             }
1528
1529             /**************************
1530              * CALCULATE INTERACTIONS *
1531              **************************/
1532
1533             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1534             {
1535
1536             r10              = _mm256_mul_ps(rsq10,rinv10);
1537             r10              = _mm256_andnot_ps(dummy_mask,r10);
1538
1539             /* Compute parameters for interactions between i and j atoms */
1540             qq10             = _mm256_mul_ps(iq1,jq0);
1541
1542             /* EWALD ELECTROSTATICS */
1543
1544             /* Analytical PME correction */
1545             zeta2            = _mm256_mul_ps(beta2,rsq10);
1546             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1547             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1548             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1549             felec            = _mm256_mul_ps(qq10,felec);
1550
1551             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1552
1553             fscal            = felec;
1554
1555             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1556
1557             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1558
1559             /* Calculate temporary vectorial force */
1560             tx               = _mm256_mul_ps(fscal,dx10);
1561             ty               = _mm256_mul_ps(fscal,dy10);
1562             tz               = _mm256_mul_ps(fscal,dz10);
1563
1564             /* Update vectorial force */
1565             fix1             = _mm256_add_ps(fix1,tx);
1566             fiy1             = _mm256_add_ps(fiy1,ty);
1567             fiz1             = _mm256_add_ps(fiz1,tz);
1568
1569             fjx0             = _mm256_add_ps(fjx0,tx);
1570             fjy0             = _mm256_add_ps(fjy0,ty);
1571             fjz0             = _mm256_add_ps(fjz0,tz);
1572
1573             }
1574
1575             /**************************
1576              * CALCULATE INTERACTIONS *
1577              **************************/
1578
1579             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1580             {
1581
1582             r20              = _mm256_mul_ps(rsq20,rinv20);
1583             r20              = _mm256_andnot_ps(dummy_mask,r20);
1584
1585             /* Compute parameters for interactions between i and j atoms */
1586             qq20             = _mm256_mul_ps(iq2,jq0);
1587
1588             /* EWALD ELECTROSTATICS */
1589
1590             /* Analytical PME correction */
1591             zeta2            = _mm256_mul_ps(beta2,rsq20);
1592             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1593             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1594             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1595             felec            = _mm256_mul_ps(qq20,felec);
1596
1597             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1598
1599             fscal            = felec;
1600
1601             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1602
1603             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1604
1605             /* Calculate temporary vectorial force */
1606             tx               = _mm256_mul_ps(fscal,dx20);
1607             ty               = _mm256_mul_ps(fscal,dy20);
1608             tz               = _mm256_mul_ps(fscal,dz20);
1609
1610             /* Update vectorial force */
1611             fix2             = _mm256_add_ps(fix2,tx);
1612             fiy2             = _mm256_add_ps(fiy2,ty);
1613             fiz2             = _mm256_add_ps(fiz2,tz);
1614
1615             fjx0             = _mm256_add_ps(fjx0,tx);
1616             fjy0             = _mm256_add_ps(fjy0,ty);
1617             fjz0             = _mm256_add_ps(fjz0,tz);
1618
1619             }
1620
1621             /**************************
1622              * CALCULATE INTERACTIONS *
1623              **************************/
1624
1625             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1626             {
1627
1628             r30              = _mm256_mul_ps(rsq30,rinv30);
1629             r30              = _mm256_andnot_ps(dummy_mask,r30);
1630
1631             /* Compute parameters for interactions between i and j atoms */
1632             qq30             = _mm256_mul_ps(iq3,jq0);
1633
1634             /* EWALD ELECTROSTATICS */
1635
1636             /* Analytical PME correction */
1637             zeta2            = _mm256_mul_ps(beta2,rsq30);
1638             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1639             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1640             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1641             felec            = _mm256_mul_ps(qq30,felec);
1642
1643             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1644
1645             fscal            = felec;
1646
1647             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1648
1649             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1650
1651             /* Calculate temporary vectorial force */
1652             tx               = _mm256_mul_ps(fscal,dx30);
1653             ty               = _mm256_mul_ps(fscal,dy30);
1654             tz               = _mm256_mul_ps(fscal,dz30);
1655
1656             /* Update vectorial force */
1657             fix3             = _mm256_add_ps(fix3,tx);
1658             fiy3             = _mm256_add_ps(fiy3,ty);
1659             fiz3             = _mm256_add_ps(fiz3,tz);
1660
1661             fjx0             = _mm256_add_ps(fjx0,tx);
1662             fjy0             = _mm256_add_ps(fjy0,ty);
1663             fjz0             = _mm256_add_ps(fjz0,tz);
1664
1665             }
1666
1667             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1668             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1669             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1670             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1671             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1672             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1673             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1674             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1675
1676             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1677
1678             /* Inner loop uses 233 flops */
1679         }
1680
1681         /* End of innermost loop */
1682
1683         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1684                                                  f+i_coord_offset,fshift+i_shift_offset);
1685
1686         /* Increment number of inner iterations */
1687         inneriter                  += j_index_end - j_index_start;
1688
1689         /* Outer loop uses 24 flops */
1690     }
1691
1692     /* Increment number of outer iterations */
1693     outeriter        += nri;
1694
1695     /* Update outer/inner flops */
1696
1697     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*233);
1698 }