src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_VF_avx_256_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     real *           vdwioffsetptr1;
  91     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  92     real *           vdwioffsetptr2;
  93     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  94     real *           vdwioffsetptr3;
  95     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  96     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  97     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  98     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  99     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 100     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 101     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 102     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 103     real             *charge;
 104     int              nvdwtype;
 105     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 106     int              *vdwtype;
 107     real             *vdwparam;
 108     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 109     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 110     __m256i          ewitab;
 111     __m128i          ewitab_lo,ewitab_hi;
 112     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 113     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 114     real             *ewtab;
 115     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 116     real             rswitch_scalar,d_scalar;
 117     __m256           dummy_mask,cutoff_mask;
 118     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 119     __m256           one     = _mm256_set1_ps(1.0);
 120     __m256           two     = _mm256_set1_ps(2.0);
 121     x                = xx[0];
 122     f                = ff[0];
 123
 124     nri              = nlist->nri;
 125     iinr             = nlist->iinr;
 126     jindex           = nlist->jindex;
 127     jjnr             = nlist->jjnr;
 128     shiftidx         = nlist->shift;
 129     gid              = nlist->gid;
 130     shiftvec         = fr->shift_vec[0];
 131     fshift           = fr->fshift[0];
 132     facel            = _mm256_set1_ps(fr->epsfac);
 133     charge           = mdatoms->chargeA;
 134     nvdwtype         = fr->ntype;
 135     vdwparam         = fr->nbfp;
 136     vdwtype          = mdatoms->typeA;
 137
 138     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 139     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 140     beta2            = _mm256_mul_ps(beta,beta);
 141     beta3            = _mm256_mul_ps(beta,beta2);
 142
 143     ewtab            = fr->ic->tabq_coul_FDV0;
 144     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 145     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 146
 147     /* Setup water-specific parameters */
 148     inr              = nlist->iinr[0];
 149     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 150     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 151     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 152     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 153
 154     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 155     rcutoff_scalar   = fr->rcoulomb;
 156     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 157     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 158
 159     rswitch_scalar   = fr->rcoulomb_switch;
 160     rswitch          = _mm256_set1_ps(rswitch_scalar);
 161     /* Setup switch parameters */
 162     d_scalar         = rcutoff_scalar-rswitch_scalar;
 163     d                = _mm256_set1_ps(d_scalar);
 164     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 165     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 166     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 167     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 168     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 169     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 170
 171     /* Avoid stupid compiler warnings */
 172     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 173     j_coord_offsetA = 0;
 174     j_coord_offsetB = 0;
 175     j_coord_offsetC = 0;
 176     j_coord_offsetD = 0;
 177     j_coord_offsetE = 0;
 178     j_coord_offsetF = 0;
 179     j_coord_offsetG = 0;
 180     j_coord_offsetH = 0;
 181
 182     outeriter        = 0;
 183     inneriter        = 0;
 184
 185     for(iidx=0;iidx<4*DIM;iidx++)
 186     {
 187         scratch[iidx] = 0.0;
 188     }
 189
 190     /* Start outer loop over neighborlists */
 191     for(iidx=0; iidx<nri; iidx++)
 192     {
 193         /* Load shift vector for this list */
 194         i_shift_offset   = DIM*shiftidx[iidx];
 195
 196         /* Load limits for loop over neighbors */
 197         j_index_start    = jindex[iidx];
 198         j_index_end      = jindex[iidx+1];
 199
 200         /* Get outer coordinate index */
 201         inr              = iinr[iidx];
 202         i_coord_offset   = DIM*inr;
 203
 204         /* Load i particle coords and add shift vector */
 205         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 206                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 207
 208         fix0             = _mm256_setzero_ps();
 209         fiy0             = _mm256_setzero_ps();
 210         fiz0             = _mm256_setzero_ps();
 211         fix1             = _mm256_setzero_ps();
 212         fiy1             = _mm256_setzero_ps();
 213         fiz1             = _mm256_setzero_ps();
 214         fix2             = _mm256_setzero_ps();
 215         fiy2             = _mm256_setzero_ps();
 216         fiz2             = _mm256_setzero_ps();
 217         fix3             = _mm256_setzero_ps();
 218         fiy3             = _mm256_setzero_ps();
 219         fiz3             = _mm256_setzero_ps();
 220
 221         /* Reset potential sums */
 222         velecsum         = _mm256_setzero_ps();
 223         vvdwsum          = _mm256_setzero_ps();
 224
 225         /* Start inner kernel loop */
 226         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 227         {
 228
 229             /* Get j neighbor index, and coordinate index */
 230             jnrA             = jjnr[jidx];
 231             jnrB             = jjnr[jidx+1];
 232             jnrC             = jjnr[jidx+2];
 233             jnrD             = jjnr[jidx+3];
 234             jnrE             = jjnr[jidx+4];
 235             jnrF             = jjnr[jidx+5];
 236             jnrG             = jjnr[jidx+6];
 237             jnrH             = jjnr[jidx+7];
 238             j_coord_offsetA  = DIM*jnrA;
 239             j_coord_offsetB  = DIM*jnrB;
 240             j_coord_offsetC  = DIM*jnrC;
 241             j_coord_offsetD  = DIM*jnrD;
 242             j_coord_offsetE  = DIM*jnrE;
 243             j_coord_offsetF  = DIM*jnrF;
 244             j_coord_offsetG  = DIM*jnrG;
 245             j_coord_offsetH  = DIM*jnrH;
 246
 247             /* load j atom coordinates */
 248             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 249                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 250                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 251                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 252                                                  &jx0,&jy0,&jz0);
 253
 254             /* Calculate displacement vector */
 255             dx00             = _mm256_sub_ps(ix0,jx0);
 256             dy00             = _mm256_sub_ps(iy0,jy0);
 257             dz00             = _mm256_sub_ps(iz0,jz0);
 258             dx10             = _mm256_sub_ps(ix1,jx0);
 259             dy10             = _mm256_sub_ps(iy1,jy0);
 260             dz10             = _mm256_sub_ps(iz1,jz0);
 261             dx20             = _mm256_sub_ps(ix2,jx0);
 262             dy20             = _mm256_sub_ps(iy2,jy0);
 263             dz20             = _mm256_sub_ps(iz2,jz0);
 264             dx30             = _mm256_sub_ps(ix3,jx0);
 265             dy30             = _mm256_sub_ps(iy3,jy0);
 266             dz30             = _mm256_sub_ps(iz3,jz0);
 267
 268             /* Calculate squared distance and things based on it */
 269             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 270             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 271             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 272             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 273
 274             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 275             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 276             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 277             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 278
 279             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 280             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 281             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 282             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 283
 284             /* Load parameters for j particles */
 285             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 286                                                                  charge+jnrC+0,charge+jnrD+0,
 287                                                                  charge+jnrE+0,charge+jnrF+0,
 288                                                                  charge+jnrG+0,charge+jnrH+0);
 289             vdwjidx0A        = 2*vdwtype[jnrA+0];
 290             vdwjidx0B        = 2*vdwtype[jnrB+0];
 291             vdwjidx0C        = 2*vdwtype[jnrC+0];
 292             vdwjidx0D        = 2*vdwtype[jnrD+0];
 293             vdwjidx0E        = 2*vdwtype[jnrE+0];
 294             vdwjidx0F        = 2*vdwtype[jnrF+0];
 295             vdwjidx0G        = 2*vdwtype[jnrG+0];
 296             vdwjidx0H        = 2*vdwtype[jnrH+0];
 297
 298             fjx0             = _mm256_setzero_ps();
 299             fjy0             = _mm256_setzero_ps();
 300             fjz0             = _mm256_setzero_ps();
 301
 302             /**************************
 303              * CALCULATE INTERACTIONS *
 304              **************************/
 305
 306             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 307             {
 308
 309             r00              = _mm256_mul_ps(rsq00,rinv00);
 310
 311             /* Compute parameters for interactions between i and j atoms */
 312             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 313                                             vdwioffsetptr0+vdwjidx0B,
 314                                             vdwioffsetptr0+vdwjidx0C,
 315                                             vdwioffsetptr0+vdwjidx0D,
 316                                             vdwioffsetptr0+vdwjidx0E,
 317                                             vdwioffsetptr0+vdwjidx0F,
 318                                             vdwioffsetptr0+vdwjidx0G,
 319                                             vdwioffsetptr0+vdwjidx0H,
 320                                             &c6_00,&c12_00);
 321
 322             /* LENNARD-JONES DISPERSION/REPULSION */
 323
 324             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 325             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 326             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 327             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 328             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 329
 330             d                = _mm256_sub_ps(r00,rswitch);
 331             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 332             d2               = _mm256_mul_ps(d,d);
 333             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 334
 335             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 336
 337             /* Evaluate switch function */
 338             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 339             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 340             vvdw             = _mm256_mul_ps(vvdw,sw);
 341             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 342
 343             /* Update potential sum for this i atom from the interaction with this j atom. */
 344             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 345             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 346
 347             fscal            = fvdw;
 348
 349             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 350
 351             /* Calculate temporary vectorial force */
 352             tx               = _mm256_mul_ps(fscal,dx00);
 353             ty               = _mm256_mul_ps(fscal,dy00);
 354             tz               = _mm256_mul_ps(fscal,dz00);
 355
 356             /* Update vectorial force */
 357             fix0             = _mm256_add_ps(fix0,tx);
 358             fiy0             = _mm256_add_ps(fiy0,ty);
 359             fiz0             = _mm256_add_ps(fiz0,tz);
 360
 361             fjx0             = _mm256_add_ps(fjx0,tx);
 362             fjy0             = _mm256_add_ps(fjy0,ty);
 363             fjz0             = _mm256_add_ps(fjz0,tz);
 364
 365             }
 366
 367             /**************************
 368              * CALCULATE INTERACTIONS *
 369              **************************/
 370
 371             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 372             {
 373
 374             r10              = _mm256_mul_ps(rsq10,rinv10);
 375
 376             /* Compute parameters for interactions between i and j atoms */
 377             qq10             = _mm256_mul_ps(iq1,jq0);
 378
 379             /* EWALD ELECTROSTATICS */
 380
 381             /* Analytical PME correction */
 382             zeta2            = _mm256_mul_ps(beta2,rsq10);
 383             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 384             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 385             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 386             felec            = _mm256_mul_ps(qq10,felec);
 387             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 388             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 389             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 390             velec            = _mm256_mul_ps(qq10,velec);
 391
 392             d                = _mm256_sub_ps(r10,rswitch);
 393             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 394             d2               = _mm256_mul_ps(d,d);
 395             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 396
 397             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 398
 399             /* Evaluate switch function */
 400             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 401             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv10,_mm256_mul_ps(velec,dsw)) );
 402             velec            = _mm256_mul_ps(velec,sw);
 403             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 404
 405             /* Update potential sum for this i atom from the interaction with this j atom. */
 406             velec            = _mm256_and_ps(velec,cutoff_mask);
 407             velecsum         = _mm256_add_ps(velecsum,velec);
 408
 409             fscal            = felec;
 410
 411             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 412
 413             /* Calculate temporary vectorial force */
 414             tx               = _mm256_mul_ps(fscal,dx10);
 415             ty               = _mm256_mul_ps(fscal,dy10);
 416             tz               = _mm256_mul_ps(fscal,dz10);
 417
 418             /* Update vectorial force */
 419             fix1             = _mm256_add_ps(fix1,tx);
 420             fiy1             = _mm256_add_ps(fiy1,ty);
 421             fiz1             = _mm256_add_ps(fiz1,tz);
 422
 423             fjx0             = _mm256_add_ps(fjx0,tx);
 424             fjy0             = _mm256_add_ps(fjy0,ty);
 425             fjz0             = _mm256_add_ps(fjz0,tz);
 426
 427             }
 428
 429             /**************************
 430              * CALCULATE INTERACTIONS *
 431              **************************/
 432
 433             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 434             {
 435
 436             r20              = _mm256_mul_ps(rsq20,rinv20);
 437
 438             /* Compute parameters for interactions between i and j atoms */
 439             qq20             = _mm256_mul_ps(iq2,jq0);
 440
 441             /* EWALD ELECTROSTATICS */
 442
 443             /* Analytical PME correction */
 444             zeta2            = _mm256_mul_ps(beta2,rsq20);
 445             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 446             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 447             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 448             felec            = _mm256_mul_ps(qq20,felec);
 449             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 450             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 451             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 452             velec            = _mm256_mul_ps(qq20,velec);
 453
 454             d                = _mm256_sub_ps(r20,rswitch);
 455             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 456             d2               = _mm256_mul_ps(d,d);
 457             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 458
 459             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 460
 461             /* Evaluate switch function */
 462             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 463             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv20,_mm256_mul_ps(velec,dsw)) );
 464             velec            = _mm256_mul_ps(velec,sw);
 465             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 466
 467             /* Update potential sum for this i atom from the interaction with this j atom. */
 468             velec            = _mm256_and_ps(velec,cutoff_mask);
 469             velecsum         = _mm256_add_ps(velecsum,velec);
 470
 471             fscal            = felec;
 472
 473             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 474
 475             /* Calculate temporary vectorial force */
 476             tx               = _mm256_mul_ps(fscal,dx20);
 477             ty               = _mm256_mul_ps(fscal,dy20);
 478             tz               = _mm256_mul_ps(fscal,dz20);
 479
 480             /* Update vectorial force */
 481             fix2             = _mm256_add_ps(fix2,tx);
 482             fiy2             = _mm256_add_ps(fiy2,ty);
 483             fiz2             = _mm256_add_ps(fiz2,tz);
 484
 485             fjx0             = _mm256_add_ps(fjx0,tx);
 486             fjy0             = _mm256_add_ps(fjy0,ty);
 487             fjz0             = _mm256_add_ps(fjz0,tz);
 488
 489             }
 490
 491             /**************************
 492              * CALCULATE INTERACTIONS *
 493              **************************/
 494
 495             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 496             {
 497
 498             r30              = _mm256_mul_ps(rsq30,rinv30);
 499
 500             /* Compute parameters for interactions between i and j atoms */
 501             qq30             = _mm256_mul_ps(iq3,jq0);
 502
 503             /* EWALD ELECTROSTATICS */
 504
 505             /* Analytical PME correction */
 506             zeta2            = _mm256_mul_ps(beta2,rsq30);
 507             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 508             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 509             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 510             felec            = _mm256_mul_ps(qq30,felec);
 511             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 512             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 513             velec            = _mm256_sub_ps(rinv30,pmecorrV);
 514             velec            = _mm256_mul_ps(qq30,velec);
 515
 516             d                = _mm256_sub_ps(r30,rswitch);
 517             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 518             d2               = _mm256_mul_ps(d,d);
 519             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 520
 521             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 522
 523             /* Evaluate switch function */
 524             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 525             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv30,_mm256_mul_ps(velec,dsw)) );
 526             velec            = _mm256_mul_ps(velec,sw);
 527             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 528
 529             /* Update potential sum for this i atom from the interaction with this j atom. */
 530             velec            = _mm256_and_ps(velec,cutoff_mask);
 531             velecsum         = _mm256_add_ps(velecsum,velec);
 532
 533             fscal            = felec;
 534
 535             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 536
 537             /* Calculate temporary vectorial force */
 538             tx               = _mm256_mul_ps(fscal,dx30);
 539             ty               = _mm256_mul_ps(fscal,dy30);
 540             tz               = _mm256_mul_ps(fscal,dz30);
 541
 542             /* Update vectorial force */
 543             fix3             = _mm256_add_ps(fix3,tx);
 544             fiy3             = _mm256_add_ps(fiy3,ty);
 545             fiz3             = _mm256_add_ps(fiz3,tz);
 546
 547             fjx0             = _mm256_add_ps(fjx0,tx);
 548             fjy0             = _mm256_add_ps(fjy0,ty);
 549             fjz0             = _mm256_add_ps(fjz0,tz);
 550
 551             }
 552
 553             fjptrA             = f+j_coord_offsetA;
 554             fjptrB             = f+j_coord_offsetB;
 555             fjptrC             = f+j_coord_offsetC;
 556             fjptrD             = f+j_coord_offsetD;
 557             fjptrE             = f+j_coord_offsetE;
 558             fjptrF             = f+j_coord_offsetF;
 559             fjptrG             = f+j_coord_offsetG;
 560             fjptrH             = f+j_coord_offsetH;
 561
 562             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 563
 564             /* Inner loop uses 386 flops */
 565         }
 566
 567         if(jidx<j_index_end)
 568         {
 569
 570             /* Get j neighbor index, and coordinate index */
 571             jnrlistA         = jjnr[jidx];
 572             jnrlistB         = jjnr[jidx+1];
 573             jnrlistC         = jjnr[jidx+2];
 574             jnrlistD         = jjnr[jidx+3];
 575             jnrlistE         = jjnr[jidx+4];
 576             jnrlistF         = jjnr[jidx+5];
 577             jnrlistG         = jjnr[jidx+6];
 578             jnrlistH         = jjnr[jidx+7];
 579             /* Sign of each element will be negative for non-real atoms.
 580              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 581              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 582              */
 583             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 584                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 585
 586             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 587             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 588             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 589             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 590             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 591             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 592             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 593             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 594             j_coord_offsetA  = DIM*jnrA;
 595             j_coord_offsetB  = DIM*jnrB;
 596             j_coord_offsetC  = DIM*jnrC;
 597             j_coord_offsetD  = DIM*jnrD;
 598             j_coord_offsetE  = DIM*jnrE;
 599             j_coord_offsetF  = DIM*jnrF;
 600             j_coord_offsetG  = DIM*jnrG;
 601             j_coord_offsetH  = DIM*jnrH;
 602
 603             /* load j atom coordinates */
 604             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 605                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 606                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 607                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 608                                                  &jx0,&jy0,&jz0);
 609
 610             /* Calculate displacement vector */
 611             dx00             = _mm256_sub_ps(ix0,jx0);
 612             dy00             = _mm256_sub_ps(iy0,jy0);
 613             dz00             = _mm256_sub_ps(iz0,jz0);
 614             dx10             = _mm256_sub_ps(ix1,jx0);
 615             dy10             = _mm256_sub_ps(iy1,jy0);
 616             dz10             = _mm256_sub_ps(iz1,jz0);
 617             dx20             = _mm256_sub_ps(ix2,jx0);
 618             dy20             = _mm256_sub_ps(iy2,jy0);
 619             dz20             = _mm256_sub_ps(iz2,jz0);
 620             dx30             = _mm256_sub_ps(ix3,jx0);
 621             dy30             = _mm256_sub_ps(iy3,jy0);
 622             dz30             = _mm256_sub_ps(iz3,jz0);
 623
 624             /* Calculate squared distance and things based on it */
 625             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 626             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 627             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 628             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 629
 630             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 631             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 632             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 633             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 634
 635             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 636             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 637             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 638             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 639
 640             /* Load parameters for j particles */
 641             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 642                                                                  charge+jnrC+0,charge+jnrD+0,
 643                                                                  charge+jnrE+0,charge+jnrF+0,
 644                                                                  charge+jnrG+0,charge+jnrH+0);
 645             vdwjidx0A        = 2*vdwtype[jnrA+0];
 646             vdwjidx0B        = 2*vdwtype[jnrB+0];
 647             vdwjidx0C        = 2*vdwtype[jnrC+0];
 648             vdwjidx0D        = 2*vdwtype[jnrD+0];
 649             vdwjidx0E        = 2*vdwtype[jnrE+0];
 650             vdwjidx0F        = 2*vdwtype[jnrF+0];
 651             vdwjidx0G        = 2*vdwtype[jnrG+0];
 652             vdwjidx0H        = 2*vdwtype[jnrH+0];
 653
 654             fjx0             = _mm256_setzero_ps();
 655             fjy0             = _mm256_setzero_ps();
 656             fjz0             = _mm256_setzero_ps();
 657
 658             /**************************
 659              * CALCULATE INTERACTIONS *
 660              **************************/
 661
 662             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 663             {
 664
 665             r00              = _mm256_mul_ps(rsq00,rinv00);
 666             r00              = _mm256_andnot_ps(dummy_mask,r00);
 667
 668             /* Compute parameters for interactions between i and j atoms */
 669             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 670                                             vdwioffsetptr0+vdwjidx0B,
 671                                             vdwioffsetptr0+vdwjidx0C,
 672                                             vdwioffsetptr0+vdwjidx0D,
 673                                             vdwioffsetptr0+vdwjidx0E,
 674                                             vdwioffsetptr0+vdwjidx0F,
 675                                             vdwioffsetptr0+vdwjidx0G,
 676                                             vdwioffsetptr0+vdwjidx0H,
 677                                             &c6_00,&c12_00);
 678
 679             /* LENNARD-JONES DISPERSION/REPULSION */
 680
 681             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 682             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 683             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 684             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 685             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 686
 687             d                = _mm256_sub_ps(r00,rswitch);
 688             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 689             d2               = _mm256_mul_ps(d,d);
 690             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 691
 692             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 693
 694             /* Evaluate switch function */
 695             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 696             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 697             vvdw             = _mm256_mul_ps(vvdw,sw);
 698             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 699
 700             /* Update potential sum for this i atom from the interaction with this j atom. */
 701             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 702             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 703             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 704
 705             fscal            = fvdw;
 706
 707             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 708
 709             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 710
 711             /* Calculate temporary vectorial force */
 712             tx               = _mm256_mul_ps(fscal,dx00);
 713             ty               = _mm256_mul_ps(fscal,dy00);
 714             tz               = _mm256_mul_ps(fscal,dz00);
 715
 716             /* Update vectorial force */
 717             fix0             = _mm256_add_ps(fix0,tx);
 718             fiy0             = _mm256_add_ps(fiy0,ty);
 719             fiz0             = _mm256_add_ps(fiz0,tz);
 720
 721             fjx0             = _mm256_add_ps(fjx0,tx);
 722             fjy0             = _mm256_add_ps(fjy0,ty);
 723             fjz0             = _mm256_add_ps(fjz0,tz);
 724
 725             }
 726
 727             /**************************
 728              * CALCULATE INTERACTIONS *
 729              **************************/
 730
 731             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 732             {
 733
 734             r10              = _mm256_mul_ps(rsq10,rinv10);
 735             r10              = _mm256_andnot_ps(dummy_mask,r10);
 736
 737             /* Compute parameters for interactions between i and j atoms */
 738             qq10             = _mm256_mul_ps(iq1,jq0);
 739
 740             /* EWALD ELECTROSTATICS */
 741
 742             /* Analytical PME correction */
 743             zeta2            = _mm256_mul_ps(beta2,rsq10);
 744             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 745             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 746             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 747             felec            = _mm256_mul_ps(qq10,felec);
 748             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 749             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 750             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 751             velec            = _mm256_mul_ps(qq10,velec);
 752
 753             d                = _mm256_sub_ps(r10,rswitch);
 754             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 755             d2               = _mm256_mul_ps(d,d);
 756             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 757
 758             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 759
 760             /* Evaluate switch function */
 761             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 762             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv10,_mm256_mul_ps(velec,dsw)) );
 763             velec            = _mm256_mul_ps(velec,sw);
 764             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 765
 766             /* Update potential sum for this i atom from the interaction with this j atom. */
 767             velec            = _mm256_and_ps(velec,cutoff_mask);
 768             velec            = _mm256_andnot_ps(dummy_mask,velec);
 769             velecsum         = _mm256_add_ps(velecsum,velec);
 770
 771             fscal            = felec;
 772
 773             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 774
 775             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 776
 777             /* Calculate temporary vectorial force */
 778             tx               = _mm256_mul_ps(fscal,dx10);
 779             ty               = _mm256_mul_ps(fscal,dy10);
 780             tz               = _mm256_mul_ps(fscal,dz10);
 781
 782             /* Update vectorial force */
 783             fix1             = _mm256_add_ps(fix1,tx);
 784             fiy1             = _mm256_add_ps(fiy1,ty);
 785             fiz1             = _mm256_add_ps(fiz1,tz);
 786
 787             fjx0             = _mm256_add_ps(fjx0,tx);
 788             fjy0             = _mm256_add_ps(fjy0,ty);
 789             fjz0             = _mm256_add_ps(fjz0,tz);
 790
 791             }
 792
 793             /**************************
 794              * CALCULATE INTERACTIONS *
 795              **************************/
 796
 797             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 798             {
 799
 800             r20              = _mm256_mul_ps(rsq20,rinv20);
 801             r20              = _mm256_andnot_ps(dummy_mask,r20);
 802
 803             /* Compute parameters for interactions between i and j atoms */
 804             qq20             = _mm256_mul_ps(iq2,jq0);
 805
 806             /* EWALD ELECTROSTATICS */
 807
 808             /* Analytical PME correction */
 809             zeta2            = _mm256_mul_ps(beta2,rsq20);
 810             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 811             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 812             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 813             felec            = _mm256_mul_ps(qq20,felec);
 814             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 815             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 816             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 817             velec            = _mm256_mul_ps(qq20,velec);
 818
 819             d                = _mm256_sub_ps(r20,rswitch);
 820             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 821             d2               = _mm256_mul_ps(d,d);
 822             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 823
 824             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 825
 826             /* Evaluate switch function */
 827             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 828             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv20,_mm256_mul_ps(velec,dsw)) );
 829             velec            = _mm256_mul_ps(velec,sw);
 830             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 831
 832             /* Update potential sum for this i atom from the interaction with this j atom. */
 833             velec            = _mm256_and_ps(velec,cutoff_mask);
 834             velec            = _mm256_andnot_ps(dummy_mask,velec);
 835             velecsum         = _mm256_add_ps(velecsum,velec);
 836
 837             fscal            = felec;
 838
 839             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 840
 841             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 842
 843             /* Calculate temporary vectorial force */
 844             tx               = _mm256_mul_ps(fscal,dx20);
 845             ty               = _mm256_mul_ps(fscal,dy20);
 846             tz               = _mm256_mul_ps(fscal,dz20);
 847
 848             /* Update vectorial force */
 849             fix2             = _mm256_add_ps(fix2,tx);
 850             fiy2             = _mm256_add_ps(fiy2,ty);
 851             fiz2             = _mm256_add_ps(fiz2,tz);
 852
 853             fjx0             = _mm256_add_ps(fjx0,tx);
 854             fjy0             = _mm256_add_ps(fjy0,ty);
 855             fjz0             = _mm256_add_ps(fjz0,tz);
 856
 857             }
 858
 859             /**************************
 860              * CALCULATE INTERACTIONS *
 861              **************************/
 862
 863             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 864             {
 865
 866             r30              = _mm256_mul_ps(rsq30,rinv30);
 867             r30              = _mm256_andnot_ps(dummy_mask,r30);
 868
 869             /* Compute parameters for interactions between i and j atoms */
 870             qq30             = _mm256_mul_ps(iq3,jq0);
 871
 872             /* EWALD ELECTROSTATICS */
 873
 874             /* Analytical PME correction */
 875             zeta2            = _mm256_mul_ps(beta2,rsq30);
 876             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 877             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 878             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 879             felec            = _mm256_mul_ps(qq30,felec);
 880             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 881             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 882             velec            = _mm256_sub_ps(rinv30,pmecorrV);
 883             velec            = _mm256_mul_ps(qq30,velec);
 884
 885             d                = _mm256_sub_ps(r30,rswitch);
 886             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 887             d2               = _mm256_mul_ps(d,d);
 888             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 889
 890             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 891
 892             /* Evaluate switch function */
 893             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 894             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv30,_mm256_mul_ps(velec,dsw)) );
 895             velec            = _mm256_mul_ps(velec,sw);
 896             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 897
 898             /* Update potential sum for this i atom from the interaction with this j atom. */
 899             velec            = _mm256_and_ps(velec,cutoff_mask);
 900             velec            = _mm256_andnot_ps(dummy_mask,velec);
 901             velecsum         = _mm256_add_ps(velecsum,velec);
 902
 903             fscal            = felec;
 904
 905             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 906
 907             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 908
 909             /* Calculate temporary vectorial force */
 910             tx               = _mm256_mul_ps(fscal,dx30);
 911             ty               = _mm256_mul_ps(fscal,dy30);
 912             tz               = _mm256_mul_ps(fscal,dz30);
 913
 914             /* Update vectorial force */
 915             fix3             = _mm256_add_ps(fix3,tx);
 916             fiy3             = _mm256_add_ps(fiy3,ty);
 917             fiz3             = _mm256_add_ps(fiz3,tz);
 918
 919             fjx0             = _mm256_add_ps(fjx0,tx);
 920             fjy0             = _mm256_add_ps(fjy0,ty);
 921             fjz0             = _mm256_add_ps(fjz0,tz);
 922
 923             }
 924
 925             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 926             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 927             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 928             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 929             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 930             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 931             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 932             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 933
 934             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 935
 936             /* Inner loop uses 390 flops */
 937         }
 938
 939         /* End of innermost loop */
 940
 941         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 942                                                  f+i_coord_offset,fshift+i_shift_offset);
 943
 944         ggid                        = gid[iidx];
 945         /* Update potential energies */
 946         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 947         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 948
 949         /* Increment number of inner iterations */
 950         inneriter                  += j_index_end - j_index_start;
 951
 952         /* Outer loop uses 26 flops */
 953     }
 954
 955     /* Increment number of outer iterations */
 956     outeriter        += nri;
 957
 958     /* Update outer/inner flops */
 959
 960     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*390);
 961 }
 962 /*
 963  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_F_avx_256_single
 964  * Electrostatics interaction: Ewald
 965  * VdW interaction:            LennardJones
 966  * Geometry:                   Water4-Particle
 967  * Calculate force/pot:        Force
 968  */
 969 void
 970 nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_F_avx_256_single
 971                     (t_nblist                    * gmx_restrict       nlist,
 972                      rvec                        * gmx_restrict          xx,
 973                      rvec                        * gmx_restrict          ff,
 974                      t_forcerec                  * gmx_restrict          fr,
 975                      t_mdatoms                   * gmx_restrict     mdatoms,
 976                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 977                      t_nrnb                      * gmx_restrict        nrnb)
 978 {
 979     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 980      * just 0 for non-waters.
 981      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 982      * jnr indices corresponding to data put in the four positions in the SIMD register.
 983      */
 984     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 985     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 986     int              jnrA,jnrB,jnrC,jnrD;
 987     int              jnrE,jnrF,jnrG,jnrH;
 988     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 989     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 990     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 991     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 992     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 993     real             rcutoff_scalar;
 994     real             *shiftvec,*fshift,*x,*f;
 995     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 996     real             scratch[4*DIM];
 997     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 998     real *           vdwioffsetptr0;
 999     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1000     real *           vdwioffsetptr1;
1001     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1002     real *           vdwioffsetptr2;
1003     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1004     real *           vdwioffsetptr3;
1005     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1006     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1007     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1008     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1009     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1010     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1011     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
1012     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1013     real             *charge;
1014     int              nvdwtype;
1015     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1016     int              *vdwtype;
1017     real             *vdwparam;
1018     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
1019     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
1020     __m256i          ewitab;
1021     __m128i          ewitab_lo,ewitab_hi;
1022     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1023     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1024     real             *ewtab;
1025     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
1026     real             rswitch_scalar,d_scalar;
1027     __m256           dummy_mask,cutoff_mask;
1028     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1029     __m256           one     = _mm256_set1_ps(1.0);
1030     __m256           two     = _mm256_set1_ps(2.0);
1031     x                = xx[0];
1032     f                = ff[0];
1033
1034     nri              = nlist->nri;
1035     iinr             = nlist->iinr;
1036     jindex           = nlist->jindex;
1037     jjnr             = nlist->jjnr;
1038     shiftidx         = nlist->shift;
1039     gid              = nlist->gid;
1040     shiftvec         = fr->shift_vec[0];
1041     fshift           = fr->fshift[0];
1042     facel            = _mm256_set1_ps(fr->epsfac);
1043     charge           = mdatoms->chargeA;
1044     nvdwtype         = fr->ntype;
1045     vdwparam         = fr->nbfp;
1046     vdwtype          = mdatoms->typeA;
1047
1048     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
1049     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
1050     beta2            = _mm256_mul_ps(beta,beta);
1051     beta3            = _mm256_mul_ps(beta,beta2);
1052
1053     ewtab            = fr->ic->tabq_coul_FDV0;
1054     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
1055     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
1056
1057     /* Setup water-specific parameters */
1058     inr              = nlist->iinr[0];
1059     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1060     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1061     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
1062     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1063
1064     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1065     rcutoff_scalar   = fr->rcoulomb;
1066     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
1067     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
1068
1069     rswitch_scalar   = fr->rcoulomb_switch;
1070     rswitch          = _mm256_set1_ps(rswitch_scalar);
1071     /* Setup switch parameters */
1072     d_scalar         = rcutoff_scalar-rswitch_scalar;
1073     d                = _mm256_set1_ps(d_scalar);
1074     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
1075     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
1076     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
1077     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
1078     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
1079     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
1080
1081     /* Avoid stupid compiler warnings */
1082     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1083     j_coord_offsetA = 0;
1084     j_coord_offsetB = 0;
1085     j_coord_offsetC = 0;
1086     j_coord_offsetD = 0;
1087     j_coord_offsetE = 0;
1088     j_coord_offsetF = 0;
1089     j_coord_offsetG = 0;
1090     j_coord_offsetH = 0;
1091
1092     outeriter        = 0;
1093     inneriter        = 0;
1094
1095     for(iidx=0;iidx<4*DIM;iidx++)
1096     {
1097         scratch[iidx] = 0.0;
1098     }
1099
1100     /* Start outer loop over neighborlists */
1101     for(iidx=0; iidx<nri; iidx++)
1102     {
1103         /* Load shift vector for this list */
1104         i_shift_offset   = DIM*shiftidx[iidx];
1105
1106         /* Load limits for loop over neighbors */
1107         j_index_start    = jindex[iidx];
1108         j_index_end      = jindex[iidx+1];
1109
1110         /* Get outer coordinate index */
1111         inr              = iinr[iidx];
1112         i_coord_offset   = DIM*inr;
1113
1114         /* Load i particle coords and add shift vector */
1115         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1116                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1117
1118         fix0             = _mm256_setzero_ps();
1119         fiy0             = _mm256_setzero_ps();
1120         fiz0             = _mm256_setzero_ps();
1121         fix1             = _mm256_setzero_ps();
1122         fiy1             = _mm256_setzero_ps();
1123         fiz1             = _mm256_setzero_ps();
1124         fix2             = _mm256_setzero_ps();
1125         fiy2             = _mm256_setzero_ps();
1126         fiz2             = _mm256_setzero_ps();
1127         fix3             = _mm256_setzero_ps();
1128         fiy3             = _mm256_setzero_ps();
1129         fiz3             = _mm256_setzero_ps();
1130
1131         /* Start inner kernel loop */
1132         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1133         {
1134
1135             /* Get j neighbor index, and coordinate index */
1136             jnrA             = jjnr[jidx];
1137             jnrB             = jjnr[jidx+1];
1138             jnrC             = jjnr[jidx+2];
1139             jnrD             = jjnr[jidx+3];
1140             jnrE             = jjnr[jidx+4];
1141             jnrF             = jjnr[jidx+5];
1142             jnrG             = jjnr[jidx+6];
1143             jnrH             = jjnr[jidx+7];
1144             j_coord_offsetA  = DIM*jnrA;
1145             j_coord_offsetB  = DIM*jnrB;
1146             j_coord_offsetC  = DIM*jnrC;
1147             j_coord_offsetD  = DIM*jnrD;
1148             j_coord_offsetE  = DIM*jnrE;
1149             j_coord_offsetF  = DIM*jnrF;
1150             j_coord_offsetG  = DIM*jnrG;
1151             j_coord_offsetH  = DIM*jnrH;
1152
1153             /* load j atom coordinates */
1154             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1155                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1156                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1157                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1158                                                  &jx0,&jy0,&jz0);
1159
1160             /* Calculate displacement vector */
1161             dx00             = _mm256_sub_ps(ix0,jx0);
1162             dy00             = _mm256_sub_ps(iy0,jy0);
1163             dz00             = _mm256_sub_ps(iz0,jz0);
1164             dx10             = _mm256_sub_ps(ix1,jx0);
1165             dy10             = _mm256_sub_ps(iy1,jy0);
1166             dz10             = _mm256_sub_ps(iz1,jz0);
1167             dx20             = _mm256_sub_ps(ix2,jx0);
1168             dy20             = _mm256_sub_ps(iy2,jy0);
1169             dz20             = _mm256_sub_ps(iz2,jz0);
1170             dx30             = _mm256_sub_ps(ix3,jx0);
1171             dy30             = _mm256_sub_ps(iy3,jy0);
1172             dz30             = _mm256_sub_ps(iz3,jz0);
1173
1174             /* Calculate squared distance and things based on it */
1175             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1176             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1177             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1178             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1179
1180             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1181             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1182             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1183             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1184
1185             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1186             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1187             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1188             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1189
1190             /* Load parameters for j particles */
1191             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1192                                                                  charge+jnrC+0,charge+jnrD+0,
1193                                                                  charge+jnrE+0,charge+jnrF+0,
1194                                                                  charge+jnrG+0,charge+jnrH+0);
1195             vdwjidx0A        = 2*vdwtype[jnrA+0];
1196             vdwjidx0B        = 2*vdwtype[jnrB+0];
1197             vdwjidx0C        = 2*vdwtype[jnrC+0];
1198             vdwjidx0D        = 2*vdwtype[jnrD+0];
1199             vdwjidx0E        = 2*vdwtype[jnrE+0];
1200             vdwjidx0F        = 2*vdwtype[jnrF+0];
1201             vdwjidx0G        = 2*vdwtype[jnrG+0];
1202             vdwjidx0H        = 2*vdwtype[jnrH+0];
1203
1204             fjx0             = _mm256_setzero_ps();
1205             fjy0             = _mm256_setzero_ps();
1206             fjz0             = _mm256_setzero_ps();
1207
1208             /**************************
1209              * CALCULATE INTERACTIONS *
1210              **************************/
1211
1212             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1213             {
1214
1215             r00              = _mm256_mul_ps(rsq00,rinv00);
1216
1217             /* Compute parameters for interactions between i and j atoms */
1218             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1219                                             vdwioffsetptr0+vdwjidx0B,
1220                                             vdwioffsetptr0+vdwjidx0C,
1221                                             vdwioffsetptr0+vdwjidx0D,
1222                                             vdwioffsetptr0+vdwjidx0E,
1223                                             vdwioffsetptr0+vdwjidx0F,
1224                                             vdwioffsetptr0+vdwjidx0G,
1225                                             vdwioffsetptr0+vdwjidx0H,
1226                                             &c6_00,&c12_00);
1227
1228             /* LENNARD-JONES DISPERSION/REPULSION */
1229
1230             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1231             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
1232             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
1233             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
1234             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
1235
1236             d                = _mm256_sub_ps(r00,rswitch);
1237             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1238             d2               = _mm256_mul_ps(d,d);
1239             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
1240
1241             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1242
1243             /* Evaluate switch function */
1244             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1245             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
1246             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1247
1248             fscal            = fvdw;
1249
1250             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1251
1252             /* Calculate temporary vectorial force */
1253             tx               = _mm256_mul_ps(fscal,dx00);
1254             ty               = _mm256_mul_ps(fscal,dy00);
1255             tz               = _mm256_mul_ps(fscal,dz00);
1256
1257             /* Update vectorial force */
1258             fix0             = _mm256_add_ps(fix0,tx);
1259             fiy0             = _mm256_add_ps(fiy0,ty);
1260             fiz0             = _mm256_add_ps(fiz0,tz);
1261
1262             fjx0             = _mm256_add_ps(fjx0,tx);
1263             fjy0             = _mm256_add_ps(fjy0,ty);
1264             fjz0             = _mm256_add_ps(fjz0,tz);
1265
1266             }
1267
1268             /**************************
1269              * CALCULATE INTERACTIONS *
1270              **************************/
1271
1272             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1273             {
1274
1275             r10              = _mm256_mul_ps(rsq10,rinv10);
1276
1277             /* Compute parameters for interactions between i and j atoms */
1278             qq10             = _mm256_mul_ps(iq1,jq0);
1279
1280             /* EWALD ELECTROSTATICS */
1281
1282             /* Analytical PME correction */
1283             zeta2            = _mm256_mul_ps(beta2,rsq10);
1284             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1285             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1286             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1287             felec            = _mm256_mul_ps(qq10,felec);
1288             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1289             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1290             velec            = _mm256_sub_ps(rinv10,pmecorrV);
1291             velec            = _mm256_mul_ps(qq10,velec);
1292
1293             d                = _mm256_sub_ps(r10,rswitch);
1294             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1295             d2               = _mm256_mul_ps(d,d);
1296             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
1297
1298             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1299
1300             /* Evaluate switch function */
1301             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1302             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv10,_mm256_mul_ps(velec,dsw)) );
1303             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1304
1305             fscal            = felec;
1306
1307             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1308
1309             /* Calculate temporary vectorial force */
1310             tx               = _mm256_mul_ps(fscal,dx10);
1311             ty               = _mm256_mul_ps(fscal,dy10);
1312             tz               = _mm256_mul_ps(fscal,dz10);
1313
1314             /* Update vectorial force */
1315             fix1             = _mm256_add_ps(fix1,tx);
1316             fiy1             = _mm256_add_ps(fiy1,ty);
1317             fiz1             = _mm256_add_ps(fiz1,tz);
1318
1319             fjx0             = _mm256_add_ps(fjx0,tx);
1320             fjy0             = _mm256_add_ps(fjy0,ty);
1321             fjz0             = _mm256_add_ps(fjz0,tz);
1322
1323             }
1324
1325             /**************************
1326              * CALCULATE INTERACTIONS *
1327              **************************/
1328
1329             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1330             {
1331
1332             r20              = _mm256_mul_ps(rsq20,rinv20);
1333
1334             /* Compute parameters for interactions between i and j atoms */
1335             qq20             = _mm256_mul_ps(iq2,jq0);
1336
1337             /* EWALD ELECTROSTATICS */
1338
1339             /* Analytical PME correction */
1340             zeta2            = _mm256_mul_ps(beta2,rsq20);
1341             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1342             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1343             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1344             felec            = _mm256_mul_ps(qq20,felec);
1345             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1346             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1347             velec            = _mm256_sub_ps(rinv20,pmecorrV);
1348             velec            = _mm256_mul_ps(qq20,velec);
1349
1350             d                = _mm256_sub_ps(r20,rswitch);
1351             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1352             d2               = _mm256_mul_ps(d,d);
1353             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
1354
1355             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1356
1357             /* Evaluate switch function */
1358             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1359             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv20,_mm256_mul_ps(velec,dsw)) );
1360             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1361
1362             fscal            = felec;
1363
1364             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1365
1366             /* Calculate temporary vectorial force */
1367             tx               = _mm256_mul_ps(fscal,dx20);
1368             ty               = _mm256_mul_ps(fscal,dy20);
1369             tz               = _mm256_mul_ps(fscal,dz20);
1370
1371             /* Update vectorial force */
1372             fix2             = _mm256_add_ps(fix2,tx);
1373             fiy2             = _mm256_add_ps(fiy2,ty);
1374             fiz2             = _mm256_add_ps(fiz2,tz);
1375
1376             fjx0             = _mm256_add_ps(fjx0,tx);
1377             fjy0             = _mm256_add_ps(fjy0,ty);
1378             fjz0             = _mm256_add_ps(fjz0,tz);
1379
1380             }
1381
1382             /**************************
1383              * CALCULATE INTERACTIONS *
1384              **************************/
1385
1386             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1387             {
1388
1389             r30              = _mm256_mul_ps(rsq30,rinv30);
1390
1391             /* Compute parameters for interactions between i and j atoms */
1392             qq30             = _mm256_mul_ps(iq3,jq0);
1393
1394             /* EWALD ELECTROSTATICS */
1395
1396             /* Analytical PME correction */
1397             zeta2            = _mm256_mul_ps(beta2,rsq30);
1398             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1399             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1400             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1401             felec            = _mm256_mul_ps(qq30,felec);
1402             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1403             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1404             velec            = _mm256_sub_ps(rinv30,pmecorrV);
1405             velec            = _mm256_mul_ps(qq30,velec);
1406
1407             d                = _mm256_sub_ps(r30,rswitch);
1408             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1409             d2               = _mm256_mul_ps(d,d);
1410             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
1411
1412             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1413
1414             /* Evaluate switch function */
1415             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1416             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv30,_mm256_mul_ps(velec,dsw)) );
1417             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1418
1419             fscal            = felec;
1420
1421             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1422
1423             /* Calculate temporary vectorial force */
1424             tx               = _mm256_mul_ps(fscal,dx30);
1425             ty               = _mm256_mul_ps(fscal,dy30);
1426             tz               = _mm256_mul_ps(fscal,dz30);
1427
1428             /* Update vectorial force */
1429             fix3             = _mm256_add_ps(fix3,tx);
1430             fiy3             = _mm256_add_ps(fiy3,ty);
1431             fiz3             = _mm256_add_ps(fiz3,tz);
1432
1433             fjx0             = _mm256_add_ps(fjx0,tx);
1434             fjy0             = _mm256_add_ps(fjy0,ty);
1435             fjz0             = _mm256_add_ps(fjz0,tz);
1436
1437             }
1438
1439             fjptrA             = f+j_coord_offsetA;
1440             fjptrB             = f+j_coord_offsetB;
1441             fjptrC             = f+j_coord_offsetC;
1442             fjptrD             = f+j_coord_offsetD;
1443             fjptrE             = f+j_coord_offsetE;
1444             fjptrF             = f+j_coord_offsetF;
1445             fjptrG             = f+j_coord_offsetG;
1446             fjptrH             = f+j_coord_offsetH;
1447
1448             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1449
1450             /* Inner loop uses 374 flops */
1451         }
1452
1453         if(jidx<j_index_end)
1454         {
1455
1456             /* Get j neighbor index, and coordinate index */
1457             jnrlistA         = jjnr[jidx];
1458             jnrlistB         = jjnr[jidx+1];
1459             jnrlistC         = jjnr[jidx+2];
1460             jnrlistD         = jjnr[jidx+3];
1461             jnrlistE         = jjnr[jidx+4];
1462             jnrlistF         = jjnr[jidx+5];
1463             jnrlistG         = jjnr[jidx+6];
1464             jnrlistH         = jjnr[jidx+7];
1465             /* Sign of each element will be negative for non-real atoms.
1466              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1467              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1468              */
1469             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1470                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1471
1472             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1473             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1474             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1475             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1476             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1477             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1478             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1479             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1480             j_coord_offsetA  = DIM*jnrA;
1481             j_coord_offsetB  = DIM*jnrB;
1482             j_coord_offsetC  = DIM*jnrC;
1483             j_coord_offsetD  = DIM*jnrD;
1484             j_coord_offsetE  = DIM*jnrE;
1485             j_coord_offsetF  = DIM*jnrF;
1486             j_coord_offsetG  = DIM*jnrG;
1487             j_coord_offsetH  = DIM*jnrH;
1488
1489             /* load j atom coordinates */
1490             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1491                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1492                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1493                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1494                                                  &jx0,&jy0,&jz0);
1495
1496             /* Calculate displacement vector */
1497             dx00             = _mm256_sub_ps(ix0,jx0);
1498             dy00             = _mm256_sub_ps(iy0,jy0);
1499             dz00             = _mm256_sub_ps(iz0,jz0);
1500             dx10             = _mm256_sub_ps(ix1,jx0);
1501             dy10             = _mm256_sub_ps(iy1,jy0);
1502             dz10             = _mm256_sub_ps(iz1,jz0);
1503             dx20             = _mm256_sub_ps(ix2,jx0);
1504             dy20             = _mm256_sub_ps(iy2,jy0);
1505             dz20             = _mm256_sub_ps(iz2,jz0);
1506             dx30             = _mm256_sub_ps(ix3,jx0);
1507             dy30             = _mm256_sub_ps(iy3,jy0);
1508             dz30             = _mm256_sub_ps(iz3,jz0);
1509
1510             /* Calculate squared distance and things based on it */
1511             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1512             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1513             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1514             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1515
1516             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1517             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1518             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1519             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1520
1521             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1522             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1523             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1524             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1525
1526             /* Load parameters for j particles */
1527             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1528                                                                  charge+jnrC+0,charge+jnrD+0,
1529                                                                  charge+jnrE+0,charge+jnrF+0,
1530                                                                  charge+jnrG+0,charge+jnrH+0);
1531             vdwjidx0A        = 2*vdwtype[jnrA+0];
1532             vdwjidx0B        = 2*vdwtype[jnrB+0];
1533             vdwjidx0C        = 2*vdwtype[jnrC+0];
1534             vdwjidx0D        = 2*vdwtype[jnrD+0];
1535             vdwjidx0E        = 2*vdwtype[jnrE+0];
1536             vdwjidx0F        = 2*vdwtype[jnrF+0];
1537             vdwjidx0G        = 2*vdwtype[jnrG+0];
1538             vdwjidx0H        = 2*vdwtype[jnrH+0];
1539
1540             fjx0             = _mm256_setzero_ps();
1541             fjy0             = _mm256_setzero_ps();
1542             fjz0             = _mm256_setzero_ps();
1543
1544             /**************************
1545              * CALCULATE INTERACTIONS *
1546              **************************/
1547
1548             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1549             {
1550
1551             r00              = _mm256_mul_ps(rsq00,rinv00);
1552             r00              = _mm256_andnot_ps(dummy_mask,r00);
1553
1554             /* Compute parameters for interactions between i and j atoms */
1555             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1556                                             vdwioffsetptr0+vdwjidx0B,
1557                                             vdwioffsetptr0+vdwjidx0C,
1558                                             vdwioffsetptr0+vdwjidx0D,
1559                                             vdwioffsetptr0+vdwjidx0E,
1560                                             vdwioffsetptr0+vdwjidx0F,
1561                                             vdwioffsetptr0+vdwjidx0G,
1562                                             vdwioffsetptr0+vdwjidx0H,
1563                                             &c6_00,&c12_00);
1564
1565             /* LENNARD-JONES DISPERSION/REPULSION */
1566
1567             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1568             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
1569             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
1570             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
1571             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
1572
1573             d                = _mm256_sub_ps(r00,rswitch);
1574             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1575             d2               = _mm256_mul_ps(d,d);
1576             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
1577
1578             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1579
1580             /* Evaluate switch function */
1581             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1582             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
1583             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1584
1585             fscal            = fvdw;
1586
1587             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1588
1589             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1590
1591             /* Calculate temporary vectorial force */
1592             tx               = _mm256_mul_ps(fscal,dx00);
1593             ty               = _mm256_mul_ps(fscal,dy00);
1594             tz               = _mm256_mul_ps(fscal,dz00);
1595
1596             /* Update vectorial force */
1597             fix0             = _mm256_add_ps(fix0,tx);
1598             fiy0             = _mm256_add_ps(fiy0,ty);
1599             fiz0             = _mm256_add_ps(fiz0,tz);
1600
1601             fjx0             = _mm256_add_ps(fjx0,tx);
1602             fjy0             = _mm256_add_ps(fjy0,ty);
1603             fjz0             = _mm256_add_ps(fjz0,tz);
1604
1605             }
1606
1607             /**************************
1608              * CALCULATE INTERACTIONS *
1609              **************************/
1610
1611             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1612             {
1613
1614             r10              = _mm256_mul_ps(rsq10,rinv10);
1615             r10              = _mm256_andnot_ps(dummy_mask,r10);
1616
1617             /* Compute parameters for interactions between i and j atoms */
1618             qq10             = _mm256_mul_ps(iq1,jq0);
1619
1620             /* EWALD ELECTROSTATICS */
1621
1622             /* Analytical PME correction */
1623             zeta2            = _mm256_mul_ps(beta2,rsq10);
1624             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1625             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1626             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1627             felec            = _mm256_mul_ps(qq10,felec);
1628             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1629             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1630             velec            = _mm256_sub_ps(rinv10,pmecorrV);
1631             velec            = _mm256_mul_ps(qq10,velec);
1632
1633             d                = _mm256_sub_ps(r10,rswitch);
1634             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1635             d2               = _mm256_mul_ps(d,d);
1636             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
1637
1638             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1639
1640             /* Evaluate switch function */
1641             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1642             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv10,_mm256_mul_ps(velec,dsw)) );
1643             cutoff_mask      = _mm256_cmp_ps(rsq10,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,dx10);
1653             ty               = _mm256_mul_ps(fscal,dy10);
1654             tz               = _mm256_mul_ps(fscal,dz10);
1655
1656             /* Update vectorial force */
1657             fix1             = _mm256_add_ps(fix1,tx);
1658             fiy1             = _mm256_add_ps(fiy1,ty);
1659             fiz1             = _mm256_add_ps(fiz1,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             /**************************
1668              * CALCULATE INTERACTIONS *
1669              **************************/
1670
1671             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1672             {
1673
1674             r20              = _mm256_mul_ps(rsq20,rinv20);
1675             r20              = _mm256_andnot_ps(dummy_mask,r20);
1676
1677             /* Compute parameters for interactions between i and j atoms */
1678             qq20             = _mm256_mul_ps(iq2,jq0);
1679
1680             /* EWALD ELECTROSTATICS */
1681
1682             /* Analytical PME correction */
1683             zeta2            = _mm256_mul_ps(beta2,rsq20);
1684             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1685             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1686             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1687             felec            = _mm256_mul_ps(qq20,felec);
1688             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1689             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1690             velec            = _mm256_sub_ps(rinv20,pmecorrV);
1691             velec            = _mm256_mul_ps(qq20,velec);
1692
1693             d                = _mm256_sub_ps(r20,rswitch);
1694             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1695             d2               = _mm256_mul_ps(d,d);
1696             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
1697
1698             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1699
1700             /* Evaluate switch function */
1701             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1702             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv20,_mm256_mul_ps(velec,dsw)) );
1703             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1704
1705             fscal            = felec;
1706
1707             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1708
1709             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1710
1711             /* Calculate temporary vectorial force */
1712             tx               = _mm256_mul_ps(fscal,dx20);
1713             ty               = _mm256_mul_ps(fscal,dy20);
1714             tz               = _mm256_mul_ps(fscal,dz20);
1715
1716             /* Update vectorial force */
1717             fix2             = _mm256_add_ps(fix2,tx);
1718             fiy2             = _mm256_add_ps(fiy2,ty);
1719             fiz2             = _mm256_add_ps(fiz2,tz);
1720
1721             fjx0             = _mm256_add_ps(fjx0,tx);
1722             fjy0             = _mm256_add_ps(fjy0,ty);
1723             fjz0             = _mm256_add_ps(fjz0,tz);
1724
1725             }
1726
1727             /**************************
1728              * CALCULATE INTERACTIONS *
1729              **************************/
1730
1731             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1732             {
1733
1734             r30              = _mm256_mul_ps(rsq30,rinv30);
1735             r30              = _mm256_andnot_ps(dummy_mask,r30);
1736
1737             /* Compute parameters for interactions between i and j atoms */
1738             qq30             = _mm256_mul_ps(iq3,jq0);
1739
1740             /* EWALD ELECTROSTATICS */
1741
1742             /* Analytical PME correction */
1743             zeta2            = _mm256_mul_ps(beta2,rsq30);
1744             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1745             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1746             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1747             felec            = _mm256_mul_ps(qq30,felec);
1748             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1749             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1750             velec            = _mm256_sub_ps(rinv30,pmecorrV);
1751             velec            = _mm256_mul_ps(qq30,velec);
1752
1753             d                = _mm256_sub_ps(r30,rswitch);
1754             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1755             d2               = _mm256_mul_ps(d,d);
1756             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
1757
1758             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1759
1760             /* Evaluate switch function */
1761             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1762             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv30,_mm256_mul_ps(velec,dsw)) );
1763             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1764
1765             fscal            = felec;
1766
1767             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1768
1769             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1770
1771             /* Calculate temporary vectorial force */
1772             tx               = _mm256_mul_ps(fscal,dx30);
1773             ty               = _mm256_mul_ps(fscal,dy30);
1774             tz               = _mm256_mul_ps(fscal,dz30);
1775
1776             /* Update vectorial force */
1777             fix3             = _mm256_add_ps(fix3,tx);
1778             fiy3             = _mm256_add_ps(fiy3,ty);
1779             fiz3             = _mm256_add_ps(fiz3,tz);
1780
1781             fjx0             = _mm256_add_ps(fjx0,tx);
1782             fjy0             = _mm256_add_ps(fjy0,ty);
1783             fjz0             = _mm256_add_ps(fjz0,tz);
1784
1785             }
1786
1787             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1788             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1789             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1790             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1791             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1792             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1793             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1794             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1795
1796             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1797
1798             /* Inner loop uses 378 flops */
1799         }
1800
1801         /* End of innermost loop */
1802
1803         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1804                                                  f+i_coord_offset,fshift+i_shift_offset);
1805
1806         /* Increment number of inner iterations */
1807         inneriter                  += j_index_end - j_index_start;
1808
1809         /* Outer loop uses 24 flops */
1810     }
1811
1812     /* Increment number of outer iterations */
1813     outeriter        += nri;
1814
1815     /* Update outer/inner flops */
1816
1817     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*378);
1818 }