src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_double.h"
  48 #include "kernelutil_x86_avx_256_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_VF_avx_256_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_VF_avx_256_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  84     real *           vdwioffsetptr0;
  85     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  86     real *           vdwioffsetptr1;
  87     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  88     real *           vdwioffsetptr2;
  89     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  90     real *           vdwioffsetptr3;
  91     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  93     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  96     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  97     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  98     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  99     real             *charge;
 100     int              nvdwtype;
 101     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 102     int              *vdwtype;
 103     real             *vdwparam;
 104     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 105     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 106     __m128i          ewitab;
 107     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 108     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 109     real             *ewtab;
 110     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 111     real             rswitch_scalar,d_scalar;
 112     __m256d          dummy_mask,cutoff_mask;
 113     __m128           tmpmask0,tmpmask1;
 114     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 115     __m256d          one     = _mm256_set1_pd(1.0);
 116     __m256d          two     = _mm256_set1_pd(2.0);
 117     x                = xx[0];
 118     f                = ff[0];
 119
 120     nri              = nlist->nri;
 121     iinr             = nlist->iinr;
 122     jindex           = nlist->jindex;
 123     jjnr             = nlist->jjnr;
 124     shiftidx         = nlist->shift;
 125     gid              = nlist->gid;
 126     shiftvec         = fr->shift_vec[0];
 127     fshift           = fr->fshift[0];
 128     facel            = _mm256_set1_pd(fr->epsfac);
 129     charge           = mdatoms->chargeA;
 130     nvdwtype         = fr->ntype;
 131     vdwparam         = fr->nbfp;
 132     vdwtype          = mdatoms->typeA;
 133
 134     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 135     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 136     beta2            = _mm256_mul_pd(beta,beta);
 137     beta3            = _mm256_mul_pd(beta,beta2);
 138
 139     ewtab            = fr->ic->tabq_coul_FDV0;
 140     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 141     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 142
 143     /* Setup water-specific parameters */
 144     inr              = nlist->iinr[0];
 145     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 146     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 147     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 148     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 149
 150     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 151     rcutoff_scalar   = fr->rcoulomb;
 152     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 153     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 154
 155     rswitch_scalar   = fr->rcoulomb_switch;
 156     rswitch          = _mm256_set1_pd(rswitch_scalar);
 157     /* Setup switch parameters */
 158     d_scalar         = rcutoff_scalar-rswitch_scalar;
 159     d                = _mm256_set1_pd(d_scalar);
 160     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 161     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 162     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 163     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 164     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 165     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 166
 167     /* Avoid stupid compiler warnings */
 168     jnrA = jnrB = jnrC = jnrD = 0;
 169     j_coord_offsetA = 0;
 170     j_coord_offsetB = 0;
 171     j_coord_offsetC = 0;
 172     j_coord_offsetD = 0;
 173
 174     outeriter        = 0;
 175     inneriter        = 0;
 176
 177     for(iidx=0;iidx<4*DIM;iidx++)
 178     {
 179         scratch[iidx] = 0.0;
 180     }
 181
 182     /* Start outer loop over neighborlists */
 183     for(iidx=0; iidx<nri; iidx++)
 184     {
 185         /* Load shift vector for this list */
 186         i_shift_offset   = DIM*shiftidx[iidx];
 187
 188         /* Load limits for loop over neighbors */
 189         j_index_start    = jindex[iidx];
 190         j_index_end      = jindex[iidx+1];
 191
 192         /* Get outer coordinate index */
 193         inr              = iinr[iidx];
 194         i_coord_offset   = DIM*inr;
 195
 196         /* Load i particle coords and add shift vector */
 197         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 198                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 199
 200         fix0             = _mm256_setzero_pd();
 201         fiy0             = _mm256_setzero_pd();
 202         fiz0             = _mm256_setzero_pd();
 203         fix1             = _mm256_setzero_pd();
 204         fiy1             = _mm256_setzero_pd();
 205         fiz1             = _mm256_setzero_pd();
 206         fix2             = _mm256_setzero_pd();
 207         fiy2             = _mm256_setzero_pd();
 208         fiz2             = _mm256_setzero_pd();
 209         fix3             = _mm256_setzero_pd();
 210         fiy3             = _mm256_setzero_pd();
 211         fiz3             = _mm256_setzero_pd();
 212
 213         /* Reset potential sums */
 214         velecsum         = _mm256_setzero_pd();
 215         vvdwsum          = _mm256_setzero_pd();
 216
 217         /* Start inner kernel loop */
 218         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 219         {
 220
 221             /* Get j neighbor index, and coordinate index */
 222             jnrA             = jjnr[jidx];
 223             jnrB             = jjnr[jidx+1];
 224             jnrC             = jjnr[jidx+2];
 225             jnrD             = jjnr[jidx+3];
 226             j_coord_offsetA  = DIM*jnrA;
 227             j_coord_offsetB  = DIM*jnrB;
 228             j_coord_offsetC  = DIM*jnrC;
 229             j_coord_offsetD  = DIM*jnrD;
 230
 231             /* load j atom coordinates */
 232             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 233                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 234                                                  &jx0,&jy0,&jz0);
 235
 236             /* Calculate displacement vector */
 237             dx00             = _mm256_sub_pd(ix0,jx0);
 238             dy00             = _mm256_sub_pd(iy0,jy0);
 239             dz00             = _mm256_sub_pd(iz0,jz0);
 240             dx10             = _mm256_sub_pd(ix1,jx0);
 241             dy10             = _mm256_sub_pd(iy1,jy0);
 242             dz10             = _mm256_sub_pd(iz1,jz0);
 243             dx20             = _mm256_sub_pd(ix2,jx0);
 244             dy20             = _mm256_sub_pd(iy2,jy0);
 245             dz20             = _mm256_sub_pd(iz2,jz0);
 246             dx30             = _mm256_sub_pd(ix3,jx0);
 247             dy30             = _mm256_sub_pd(iy3,jy0);
 248             dz30             = _mm256_sub_pd(iz3,jz0);
 249
 250             /* Calculate squared distance and things based on it */
 251             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 252             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 253             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 254             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 255
 256             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 257             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 258             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 259             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 260
 261             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 262             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 263             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 264             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 265
 266             /* Load parameters for j particles */
 267             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 268                                                                  charge+jnrC+0,charge+jnrD+0);
 269             vdwjidx0A        = 2*vdwtype[jnrA+0];
 270             vdwjidx0B        = 2*vdwtype[jnrB+0];
 271             vdwjidx0C        = 2*vdwtype[jnrC+0];
 272             vdwjidx0D        = 2*vdwtype[jnrD+0];
 273
 274             fjx0             = _mm256_setzero_pd();
 275             fjy0             = _mm256_setzero_pd();
 276             fjz0             = _mm256_setzero_pd();
 277
 278             /**************************
 279              * CALCULATE INTERACTIONS *
 280              **************************/
 281
 282             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 283             {
 284
 285             r00              = _mm256_mul_pd(rsq00,rinv00);
 286
 287             /* Compute parameters for interactions between i and j atoms */
 288             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 289                                             vdwioffsetptr0+vdwjidx0B,
 290                                             vdwioffsetptr0+vdwjidx0C,
 291                                             vdwioffsetptr0+vdwjidx0D,
 292                                             &c6_00,&c12_00);
 293
 294             /* LENNARD-JONES DISPERSION/REPULSION */
 295
 296             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 297             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 298             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 299             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 300             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 301
 302             d                = _mm256_sub_pd(r00,rswitch);
 303             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 304             d2               = _mm256_mul_pd(d,d);
 305             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 306
 307             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 308
 309             /* Evaluate switch function */
 310             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 311             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
 312             vvdw             = _mm256_mul_pd(vvdw,sw);
 313             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 314
 315             /* Update potential sum for this i atom from the interaction with this j atom. */
 316             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 317             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 318
 319             fscal            = fvdw;
 320
 321             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 322
 323             /* Calculate temporary vectorial force */
 324             tx               = _mm256_mul_pd(fscal,dx00);
 325             ty               = _mm256_mul_pd(fscal,dy00);
 326             tz               = _mm256_mul_pd(fscal,dz00);
 327
 328             /* Update vectorial force */
 329             fix0             = _mm256_add_pd(fix0,tx);
 330             fiy0             = _mm256_add_pd(fiy0,ty);
 331             fiz0             = _mm256_add_pd(fiz0,tz);
 332
 333             fjx0             = _mm256_add_pd(fjx0,tx);
 334             fjy0             = _mm256_add_pd(fjy0,ty);
 335             fjz0             = _mm256_add_pd(fjz0,tz);
 336
 337             }
 338
 339             /**************************
 340              * CALCULATE INTERACTIONS *
 341              **************************/
 342
 343             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 344             {
 345
 346             r10              = _mm256_mul_pd(rsq10,rinv10);
 347
 348             /* Compute parameters for interactions between i and j atoms */
 349             qq10             = _mm256_mul_pd(iq1,jq0);
 350
 351             /* EWALD ELECTROSTATICS */
 352
 353             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 354             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 355             ewitab           = _mm256_cvttpd_epi32(ewrt);
 356             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 357             ewitab           = _mm_slli_epi32(ewitab,2);
 358             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 359             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 360             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 361             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 362             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 363             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 364             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 365             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 366             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 367
 368             d                = _mm256_sub_pd(r10,rswitch);
 369             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 370             d2               = _mm256_mul_pd(d,d);
 371             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 372
 373             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 374
 375             /* Evaluate switch function */
 376             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 377             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv10,_mm256_mul_pd(velec,dsw)) );
 378             velec            = _mm256_mul_pd(velec,sw);
 379             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 380
 381             /* Update potential sum for this i atom from the interaction with this j atom. */
 382             velec            = _mm256_and_pd(velec,cutoff_mask);
 383             velecsum         = _mm256_add_pd(velecsum,velec);
 384
 385             fscal            = felec;
 386
 387             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 388
 389             /* Calculate temporary vectorial force */
 390             tx               = _mm256_mul_pd(fscal,dx10);
 391             ty               = _mm256_mul_pd(fscal,dy10);
 392             tz               = _mm256_mul_pd(fscal,dz10);
 393
 394             /* Update vectorial force */
 395             fix1             = _mm256_add_pd(fix1,tx);
 396             fiy1             = _mm256_add_pd(fiy1,ty);
 397             fiz1             = _mm256_add_pd(fiz1,tz);
 398
 399             fjx0             = _mm256_add_pd(fjx0,tx);
 400             fjy0             = _mm256_add_pd(fjy0,ty);
 401             fjz0             = _mm256_add_pd(fjz0,tz);
 402
 403             }
 404
 405             /**************************
 406              * CALCULATE INTERACTIONS *
 407              **************************/
 408
 409             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 410             {
 411
 412             r20              = _mm256_mul_pd(rsq20,rinv20);
 413
 414             /* Compute parameters for interactions between i and j atoms */
 415             qq20             = _mm256_mul_pd(iq2,jq0);
 416
 417             /* EWALD ELECTROSTATICS */
 418
 419             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 420             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 421             ewitab           = _mm256_cvttpd_epi32(ewrt);
 422             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 423             ewitab           = _mm_slli_epi32(ewitab,2);
 424             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 425             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 426             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 427             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 428             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 429             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 430             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 431             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 432             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 433
 434             d                = _mm256_sub_pd(r20,rswitch);
 435             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 436             d2               = _mm256_mul_pd(d,d);
 437             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 438
 439             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 440
 441             /* Evaluate switch function */
 442             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 443             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv20,_mm256_mul_pd(velec,dsw)) );
 444             velec            = _mm256_mul_pd(velec,sw);
 445             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 446
 447             /* Update potential sum for this i atom from the interaction with this j atom. */
 448             velec            = _mm256_and_pd(velec,cutoff_mask);
 449             velecsum         = _mm256_add_pd(velecsum,velec);
 450
 451             fscal            = felec;
 452
 453             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 454
 455             /* Calculate temporary vectorial force */
 456             tx               = _mm256_mul_pd(fscal,dx20);
 457             ty               = _mm256_mul_pd(fscal,dy20);
 458             tz               = _mm256_mul_pd(fscal,dz20);
 459
 460             /* Update vectorial force */
 461             fix2             = _mm256_add_pd(fix2,tx);
 462             fiy2             = _mm256_add_pd(fiy2,ty);
 463             fiz2             = _mm256_add_pd(fiz2,tz);
 464
 465             fjx0             = _mm256_add_pd(fjx0,tx);
 466             fjy0             = _mm256_add_pd(fjy0,ty);
 467             fjz0             = _mm256_add_pd(fjz0,tz);
 468
 469             }
 470
 471             /**************************
 472              * CALCULATE INTERACTIONS *
 473              **************************/
 474
 475             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 476             {
 477
 478             r30              = _mm256_mul_pd(rsq30,rinv30);
 479
 480             /* Compute parameters for interactions between i and j atoms */
 481             qq30             = _mm256_mul_pd(iq3,jq0);
 482
 483             /* EWALD ELECTROSTATICS */
 484
 485             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 486             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 487             ewitab           = _mm256_cvttpd_epi32(ewrt);
 488             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 489             ewitab           = _mm_slli_epi32(ewitab,2);
 490             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 491             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 492             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 493             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 494             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 495             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 496             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 497             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(rinv30,velec));
 498             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 499
 500             d                = _mm256_sub_pd(r30,rswitch);
 501             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 502             d2               = _mm256_mul_pd(d,d);
 503             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 504
 505             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 506
 507             /* Evaluate switch function */
 508             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 509             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv30,_mm256_mul_pd(velec,dsw)) );
 510             velec            = _mm256_mul_pd(velec,sw);
 511             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 512
 513             /* Update potential sum for this i atom from the interaction with this j atom. */
 514             velec            = _mm256_and_pd(velec,cutoff_mask);
 515             velecsum         = _mm256_add_pd(velecsum,velec);
 516
 517             fscal            = felec;
 518
 519             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 520
 521             /* Calculate temporary vectorial force */
 522             tx               = _mm256_mul_pd(fscal,dx30);
 523             ty               = _mm256_mul_pd(fscal,dy30);
 524             tz               = _mm256_mul_pd(fscal,dz30);
 525
 526             /* Update vectorial force */
 527             fix3             = _mm256_add_pd(fix3,tx);
 528             fiy3             = _mm256_add_pd(fiy3,ty);
 529             fiz3             = _mm256_add_pd(fiz3,tz);
 530
 531             fjx0             = _mm256_add_pd(fjx0,tx);
 532             fjy0             = _mm256_add_pd(fjy0,ty);
 533             fjz0             = _mm256_add_pd(fjz0,tz);
 534
 535             }
 536
 537             fjptrA             = f+j_coord_offsetA;
 538             fjptrB             = f+j_coord_offsetB;
 539             fjptrC             = f+j_coord_offsetC;
 540             fjptrD             = f+j_coord_offsetD;
 541
 542             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 543
 544             /* Inner loop uses 257 flops */
 545         }
 546
 547         if(jidx<j_index_end)
 548         {
 549
 550             /* Get j neighbor index, and coordinate index */
 551             jnrlistA         = jjnr[jidx];
 552             jnrlistB         = jjnr[jidx+1];
 553             jnrlistC         = jjnr[jidx+2];
 554             jnrlistD         = jjnr[jidx+3];
 555             /* Sign of each element will be negative for non-real atoms.
 556              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 557              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 558              */
 559             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 560
 561             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 562             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 563             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 564
 565             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 566             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 567             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 568             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 569             j_coord_offsetA  = DIM*jnrA;
 570             j_coord_offsetB  = DIM*jnrB;
 571             j_coord_offsetC  = DIM*jnrC;
 572             j_coord_offsetD  = DIM*jnrD;
 573
 574             /* load j atom coordinates */
 575             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 576                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 577                                                  &jx0,&jy0,&jz0);
 578
 579             /* Calculate displacement vector */
 580             dx00             = _mm256_sub_pd(ix0,jx0);
 581             dy00             = _mm256_sub_pd(iy0,jy0);
 582             dz00             = _mm256_sub_pd(iz0,jz0);
 583             dx10             = _mm256_sub_pd(ix1,jx0);
 584             dy10             = _mm256_sub_pd(iy1,jy0);
 585             dz10             = _mm256_sub_pd(iz1,jz0);
 586             dx20             = _mm256_sub_pd(ix2,jx0);
 587             dy20             = _mm256_sub_pd(iy2,jy0);
 588             dz20             = _mm256_sub_pd(iz2,jz0);
 589             dx30             = _mm256_sub_pd(ix3,jx0);
 590             dy30             = _mm256_sub_pd(iy3,jy0);
 591             dz30             = _mm256_sub_pd(iz3,jz0);
 592
 593             /* Calculate squared distance and things based on it */
 594             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 595             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 596             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 597             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 598
 599             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 600             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 601             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 602             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 603
 604             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 605             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 606             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 607             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 608
 609             /* Load parameters for j particles */
 610             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 611                                                                  charge+jnrC+0,charge+jnrD+0);
 612             vdwjidx0A        = 2*vdwtype[jnrA+0];
 613             vdwjidx0B        = 2*vdwtype[jnrB+0];
 614             vdwjidx0C        = 2*vdwtype[jnrC+0];
 615             vdwjidx0D        = 2*vdwtype[jnrD+0];
 616
 617             fjx0             = _mm256_setzero_pd();
 618             fjy0             = _mm256_setzero_pd();
 619             fjz0             = _mm256_setzero_pd();
 620
 621             /**************************
 622              * CALCULATE INTERACTIONS *
 623              **************************/
 624
 625             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 626             {
 627
 628             r00              = _mm256_mul_pd(rsq00,rinv00);
 629             r00              = _mm256_andnot_pd(dummy_mask,r00);
 630
 631             /* Compute parameters for interactions between i and j atoms */
 632             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 633                                             vdwioffsetptr0+vdwjidx0B,
 634                                             vdwioffsetptr0+vdwjidx0C,
 635                                             vdwioffsetptr0+vdwjidx0D,
 636                                             &c6_00,&c12_00);
 637
 638             /* LENNARD-JONES DISPERSION/REPULSION */
 639
 640             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 641             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 642             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 643             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 644             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 645
 646             d                = _mm256_sub_pd(r00,rswitch);
 647             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 648             d2               = _mm256_mul_pd(d,d);
 649             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 650
 651             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 652
 653             /* Evaluate switch function */
 654             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 655             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
 656             vvdw             = _mm256_mul_pd(vvdw,sw);
 657             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 658
 659             /* Update potential sum for this i atom from the interaction with this j atom. */
 660             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 661             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 662             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 663
 664             fscal            = fvdw;
 665
 666             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 667
 668             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 669
 670             /* Calculate temporary vectorial force */
 671             tx               = _mm256_mul_pd(fscal,dx00);
 672             ty               = _mm256_mul_pd(fscal,dy00);
 673             tz               = _mm256_mul_pd(fscal,dz00);
 674
 675             /* Update vectorial force */
 676             fix0             = _mm256_add_pd(fix0,tx);
 677             fiy0             = _mm256_add_pd(fiy0,ty);
 678             fiz0             = _mm256_add_pd(fiz0,tz);
 679
 680             fjx0             = _mm256_add_pd(fjx0,tx);
 681             fjy0             = _mm256_add_pd(fjy0,ty);
 682             fjz0             = _mm256_add_pd(fjz0,tz);
 683
 684             }
 685
 686             /**************************
 687              * CALCULATE INTERACTIONS *
 688              **************************/
 689
 690             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 691             {
 692
 693             r10              = _mm256_mul_pd(rsq10,rinv10);
 694             r10              = _mm256_andnot_pd(dummy_mask,r10);
 695
 696             /* Compute parameters for interactions between i and j atoms */
 697             qq10             = _mm256_mul_pd(iq1,jq0);
 698
 699             /* EWALD ELECTROSTATICS */
 700
 701             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 702             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 703             ewitab           = _mm256_cvttpd_epi32(ewrt);
 704             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 705             ewitab           = _mm_slli_epi32(ewitab,2);
 706             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 707             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 708             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 709             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 710             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 711             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 712             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 713             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 714             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 715
 716             d                = _mm256_sub_pd(r10,rswitch);
 717             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 718             d2               = _mm256_mul_pd(d,d);
 719             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 720
 721             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 722
 723             /* Evaluate switch function */
 724             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 725             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv10,_mm256_mul_pd(velec,dsw)) );
 726             velec            = _mm256_mul_pd(velec,sw);
 727             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 728
 729             /* Update potential sum for this i atom from the interaction with this j atom. */
 730             velec            = _mm256_and_pd(velec,cutoff_mask);
 731             velec            = _mm256_andnot_pd(dummy_mask,velec);
 732             velecsum         = _mm256_add_pd(velecsum,velec);
 733
 734             fscal            = felec;
 735
 736             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 737
 738             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 739
 740             /* Calculate temporary vectorial force */
 741             tx               = _mm256_mul_pd(fscal,dx10);
 742             ty               = _mm256_mul_pd(fscal,dy10);
 743             tz               = _mm256_mul_pd(fscal,dz10);
 744
 745             /* Update vectorial force */
 746             fix1             = _mm256_add_pd(fix1,tx);
 747             fiy1             = _mm256_add_pd(fiy1,ty);
 748             fiz1             = _mm256_add_pd(fiz1,tz);
 749
 750             fjx0             = _mm256_add_pd(fjx0,tx);
 751             fjy0             = _mm256_add_pd(fjy0,ty);
 752             fjz0             = _mm256_add_pd(fjz0,tz);
 753
 754             }
 755
 756             /**************************
 757              * CALCULATE INTERACTIONS *
 758              **************************/
 759
 760             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 761             {
 762
 763             r20              = _mm256_mul_pd(rsq20,rinv20);
 764             r20              = _mm256_andnot_pd(dummy_mask,r20);
 765
 766             /* Compute parameters for interactions between i and j atoms */
 767             qq20             = _mm256_mul_pd(iq2,jq0);
 768
 769             /* EWALD ELECTROSTATICS */
 770
 771             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 772             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 773             ewitab           = _mm256_cvttpd_epi32(ewrt);
 774             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 775             ewitab           = _mm_slli_epi32(ewitab,2);
 776             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 777             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 778             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 779             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 780             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 781             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 782             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 783             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 784             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 785
 786             d                = _mm256_sub_pd(r20,rswitch);
 787             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 788             d2               = _mm256_mul_pd(d,d);
 789             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 790
 791             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 792
 793             /* Evaluate switch function */
 794             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 795             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv20,_mm256_mul_pd(velec,dsw)) );
 796             velec            = _mm256_mul_pd(velec,sw);
 797             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 798
 799             /* Update potential sum for this i atom from the interaction with this j atom. */
 800             velec            = _mm256_and_pd(velec,cutoff_mask);
 801             velec            = _mm256_andnot_pd(dummy_mask,velec);
 802             velecsum         = _mm256_add_pd(velecsum,velec);
 803
 804             fscal            = felec;
 805
 806             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 807
 808             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 809
 810             /* Calculate temporary vectorial force */
 811             tx               = _mm256_mul_pd(fscal,dx20);
 812             ty               = _mm256_mul_pd(fscal,dy20);
 813             tz               = _mm256_mul_pd(fscal,dz20);
 814
 815             /* Update vectorial force */
 816             fix2             = _mm256_add_pd(fix2,tx);
 817             fiy2             = _mm256_add_pd(fiy2,ty);
 818             fiz2             = _mm256_add_pd(fiz2,tz);
 819
 820             fjx0             = _mm256_add_pd(fjx0,tx);
 821             fjy0             = _mm256_add_pd(fjy0,ty);
 822             fjz0             = _mm256_add_pd(fjz0,tz);
 823
 824             }
 825
 826             /**************************
 827              * CALCULATE INTERACTIONS *
 828              **************************/
 829
 830             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 831             {
 832
 833             r30              = _mm256_mul_pd(rsq30,rinv30);
 834             r30              = _mm256_andnot_pd(dummy_mask,r30);
 835
 836             /* Compute parameters for interactions between i and j atoms */
 837             qq30             = _mm256_mul_pd(iq3,jq0);
 838
 839             /* EWALD ELECTROSTATICS */
 840
 841             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 842             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 843             ewitab           = _mm256_cvttpd_epi32(ewrt);
 844             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 845             ewitab           = _mm_slli_epi32(ewitab,2);
 846             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 847             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 848             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 849             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 850             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 851             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 852             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 853             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(rinv30,velec));
 854             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 855
 856             d                = _mm256_sub_pd(r30,rswitch);
 857             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 858             d2               = _mm256_mul_pd(d,d);
 859             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 860
 861             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 862
 863             /* Evaluate switch function */
 864             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 865             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv30,_mm256_mul_pd(velec,dsw)) );
 866             velec            = _mm256_mul_pd(velec,sw);
 867             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 868
 869             /* Update potential sum for this i atom from the interaction with this j atom. */
 870             velec            = _mm256_and_pd(velec,cutoff_mask);
 871             velec            = _mm256_andnot_pd(dummy_mask,velec);
 872             velecsum         = _mm256_add_pd(velecsum,velec);
 873
 874             fscal            = felec;
 875
 876             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 877
 878             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 879
 880             /* Calculate temporary vectorial force */
 881             tx               = _mm256_mul_pd(fscal,dx30);
 882             ty               = _mm256_mul_pd(fscal,dy30);
 883             tz               = _mm256_mul_pd(fscal,dz30);
 884
 885             /* Update vectorial force */
 886             fix3             = _mm256_add_pd(fix3,tx);
 887             fiy3             = _mm256_add_pd(fiy3,ty);
 888             fiz3             = _mm256_add_pd(fiz3,tz);
 889
 890             fjx0             = _mm256_add_pd(fjx0,tx);
 891             fjy0             = _mm256_add_pd(fjy0,ty);
 892             fjz0             = _mm256_add_pd(fjz0,tz);
 893
 894             }
 895
 896             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 897             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 898             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 899             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 900
 901             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 902
 903             /* Inner loop uses 261 flops */
 904         }
 905
 906         /* End of innermost loop */
 907
 908         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 909                                                  f+i_coord_offset,fshift+i_shift_offset);
 910
 911         ggid                        = gid[iidx];
 912         /* Update potential energies */
 913         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 914         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 915
 916         /* Increment number of inner iterations */
 917         inneriter                  += j_index_end - j_index_start;
 918
 919         /* Outer loop uses 26 flops */
 920     }
 921
 922     /* Increment number of outer iterations */
 923     outeriter        += nri;
 924
 925     /* Update outer/inner flops */
 926
 927     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*261);
 928 }
 929 /*
 930  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_F_avx_256_double
 931  * Electrostatics interaction: Ewald
 932  * VdW interaction:            LennardJones
 933  * Geometry:                   Water4-Particle
 934  * Calculate force/pot:        Force
 935  */
 936 void
 937 nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_F_avx_256_double
 938                     (t_nblist                    * gmx_restrict       nlist,
 939                      rvec                        * gmx_restrict          xx,
 940                      rvec                        * gmx_restrict          ff,
 941                      t_forcerec                  * gmx_restrict          fr,
 942                      t_mdatoms                   * gmx_restrict     mdatoms,
 943                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 944                      t_nrnb                      * gmx_restrict        nrnb)
 945 {
 946     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 947      * just 0 for non-waters.
 948      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 949      * jnr indices corresponding to data put in the four positions in the SIMD register.
 950      */
 951     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 952     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 953     int              jnrA,jnrB,jnrC,jnrD;
 954     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 955     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 956     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 957     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 958     real             rcutoff_scalar;
 959     real             *shiftvec,*fshift,*x,*f;
 960     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 961     real             scratch[4*DIM];
 962     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 963     real *           vdwioffsetptr0;
 964     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 965     real *           vdwioffsetptr1;
 966     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 967     real *           vdwioffsetptr2;
 968     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 969     real *           vdwioffsetptr3;
 970     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 971     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 972     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 973     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 974     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 975     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 976     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 977     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 978     real             *charge;
 979     int              nvdwtype;
 980     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 981     int              *vdwtype;
 982     real             *vdwparam;
 983     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 984     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 985     __m128i          ewitab;
 986     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 987     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 988     real             *ewtab;
 989     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 990     real             rswitch_scalar,d_scalar;
 991     __m256d          dummy_mask,cutoff_mask;
 992     __m128           tmpmask0,tmpmask1;
 993     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 994     __m256d          one     = _mm256_set1_pd(1.0);
 995     __m256d          two     = _mm256_set1_pd(2.0);
 996     x                = xx[0];
 997     f                = ff[0];
 998
 999     nri              = nlist->nri;
1000     iinr             = nlist->iinr;
1001     jindex           = nlist->jindex;
1002     jjnr             = nlist->jjnr;
1003     shiftidx         = nlist->shift;
1004     gid              = nlist->gid;
1005     shiftvec         = fr->shift_vec[0];
1006     fshift           = fr->fshift[0];
1007     facel            = _mm256_set1_pd(fr->epsfac);
1008     charge           = mdatoms->chargeA;
1009     nvdwtype         = fr->ntype;
1010     vdwparam         = fr->nbfp;
1011     vdwtype          = mdatoms->typeA;
1012
1013     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
1014     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
1015     beta2            = _mm256_mul_pd(beta,beta);
1016     beta3            = _mm256_mul_pd(beta,beta2);
1017
1018     ewtab            = fr->ic->tabq_coul_FDV0;
1019     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
1020     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
1021
1022     /* Setup water-specific parameters */
1023     inr              = nlist->iinr[0];
1024     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1025     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1026     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
1027     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1028
1029     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1030     rcutoff_scalar   = fr->rcoulomb;
1031     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
1032     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
1033
1034     rswitch_scalar   = fr->rcoulomb_switch;
1035     rswitch          = _mm256_set1_pd(rswitch_scalar);
1036     /* Setup switch parameters */
1037     d_scalar         = rcutoff_scalar-rswitch_scalar;
1038     d                = _mm256_set1_pd(d_scalar);
1039     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
1040     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
1041     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
1042     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
1043     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
1044     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
1045
1046     /* Avoid stupid compiler warnings */
1047     jnrA = jnrB = jnrC = jnrD = 0;
1048     j_coord_offsetA = 0;
1049     j_coord_offsetB = 0;
1050     j_coord_offsetC = 0;
1051     j_coord_offsetD = 0;
1052
1053     outeriter        = 0;
1054     inneriter        = 0;
1055
1056     for(iidx=0;iidx<4*DIM;iidx++)
1057     {
1058         scratch[iidx] = 0.0;
1059     }
1060
1061     /* Start outer loop over neighborlists */
1062     for(iidx=0; iidx<nri; iidx++)
1063     {
1064         /* Load shift vector for this list */
1065         i_shift_offset   = DIM*shiftidx[iidx];
1066
1067         /* Load limits for loop over neighbors */
1068         j_index_start    = jindex[iidx];
1069         j_index_end      = jindex[iidx+1];
1070
1071         /* Get outer coordinate index */
1072         inr              = iinr[iidx];
1073         i_coord_offset   = DIM*inr;
1074
1075         /* Load i particle coords and add shift vector */
1076         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1077                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1078
1079         fix0             = _mm256_setzero_pd();
1080         fiy0             = _mm256_setzero_pd();
1081         fiz0             = _mm256_setzero_pd();
1082         fix1             = _mm256_setzero_pd();
1083         fiy1             = _mm256_setzero_pd();
1084         fiz1             = _mm256_setzero_pd();
1085         fix2             = _mm256_setzero_pd();
1086         fiy2             = _mm256_setzero_pd();
1087         fiz2             = _mm256_setzero_pd();
1088         fix3             = _mm256_setzero_pd();
1089         fiy3             = _mm256_setzero_pd();
1090         fiz3             = _mm256_setzero_pd();
1091
1092         /* Start inner kernel loop */
1093         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1094         {
1095
1096             /* Get j neighbor index, and coordinate index */
1097             jnrA             = jjnr[jidx];
1098             jnrB             = jjnr[jidx+1];
1099             jnrC             = jjnr[jidx+2];
1100             jnrD             = jjnr[jidx+3];
1101             j_coord_offsetA  = DIM*jnrA;
1102             j_coord_offsetB  = DIM*jnrB;
1103             j_coord_offsetC  = DIM*jnrC;
1104             j_coord_offsetD  = DIM*jnrD;
1105
1106             /* load j atom coordinates */
1107             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1108                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1109                                                  &jx0,&jy0,&jz0);
1110
1111             /* Calculate displacement vector */
1112             dx00             = _mm256_sub_pd(ix0,jx0);
1113             dy00             = _mm256_sub_pd(iy0,jy0);
1114             dz00             = _mm256_sub_pd(iz0,jz0);
1115             dx10             = _mm256_sub_pd(ix1,jx0);
1116             dy10             = _mm256_sub_pd(iy1,jy0);
1117             dz10             = _mm256_sub_pd(iz1,jz0);
1118             dx20             = _mm256_sub_pd(ix2,jx0);
1119             dy20             = _mm256_sub_pd(iy2,jy0);
1120             dz20             = _mm256_sub_pd(iz2,jz0);
1121             dx30             = _mm256_sub_pd(ix3,jx0);
1122             dy30             = _mm256_sub_pd(iy3,jy0);
1123             dz30             = _mm256_sub_pd(iz3,jz0);
1124
1125             /* Calculate squared distance and things based on it */
1126             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1127             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1128             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1129             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1130
1131             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1132             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1133             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1134             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1135
1136             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1137             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1138             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1139             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1140
1141             /* Load parameters for j particles */
1142             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1143                                                                  charge+jnrC+0,charge+jnrD+0);
1144             vdwjidx0A        = 2*vdwtype[jnrA+0];
1145             vdwjidx0B        = 2*vdwtype[jnrB+0];
1146             vdwjidx0C        = 2*vdwtype[jnrC+0];
1147             vdwjidx0D        = 2*vdwtype[jnrD+0];
1148
1149             fjx0             = _mm256_setzero_pd();
1150             fjy0             = _mm256_setzero_pd();
1151             fjz0             = _mm256_setzero_pd();
1152
1153             /**************************
1154              * CALCULATE INTERACTIONS *
1155              **************************/
1156
1157             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1158             {
1159
1160             r00              = _mm256_mul_pd(rsq00,rinv00);
1161
1162             /* Compute parameters for interactions between i and j atoms */
1163             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1164                                             vdwioffsetptr0+vdwjidx0B,
1165                                             vdwioffsetptr0+vdwjidx0C,
1166                                             vdwioffsetptr0+vdwjidx0D,
1167                                             &c6_00,&c12_00);
1168
1169             /* LENNARD-JONES DISPERSION/REPULSION */
1170
1171             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1172             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
1173             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
1174             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
1175             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
1176
1177             d                = _mm256_sub_pd(r00,rswitch);
1178             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1179             d2               = _mm256_mul_pd(d,d);
1180             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1181
1182             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1183
1184             /* Evaluate switch function */
1185             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1186             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
1187             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1188
1189             fscal            = fvdw;
1190
1191             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1192
1193             /* Calculate temporary vectorial force */
1194             tx               = _mm256_mul_pd(fscal,dx00);
1195             ty               = _mm256_mul_pd(fscal,dy00);
1196             tz               = _mm256_mul_pd(fscal,dz00);
1197
1198             /* Update vectorial force */
1199             fix0             = _mm256_add_pd(fix0,tx);
1200             fiy0             = _mm256_add_pd(fiy0,ty);
1201             fiz0             = _mm256_add_pd(fiz0,tz);
1202
1203             fjx0             = _mm256_add_pd(fjx0,tx);
1204             fjy0             = _mm256_add_pd(fjy0,ty);
1205             fjz0             = _mm256_add_pd(fjz0,tz);
1206
1207             }
1208
1209             /**************************
1210              * CALCULATE INTERACTIONS *
1211              **************************/
1212
1213             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1214             {
1215
1216             r10              = _mm256_mul_pd(rsq10,rinv10);
1217
1218             /* Compute parameters for interactions between i and j atoms */
1219             qq10             = _mm256_mul_pd(iq1,jq0);
1220
1221             /* EWALD ELECTROSTATICS */
1222
1223             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1224             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1225             ewitab           = _mm256_cvttpd_epi32(ewrt);
1226             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1227             ewitab           = _mm_slli_epi32(ewitab,2);
1228             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1229             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1230             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1231             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1232             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1233             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1234             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1235             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
1236             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1237
1238             d                = _mm256_sub_pd(r10,rswitch);
1239             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1240             d2               = _mm256_mul_pd(d,d);
1241             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1242
1243             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1244
1245             /* Evaluate switch function */
1246             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1247             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv10,_mm256_mul_pd(velec,dsw)) );
1248             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1249
1250             fscal            = felec;
1251
1252             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1253
1254             /* Calculate temporary vectorial force */
1255             tx               = _mm256_mul_pd(fscal,dx10);
1256             ty               = _mm256_mul_pd(fscal,dy10);
1257             tz               = _mm256_mul_pd(fscal,dz10);
1258
1259             /* Update vectorial force */
1260             fix1             = _mm256_add_pd(fix1,tx);
1261             fiy1             = _mm256_add_pd(fiy1,ty);
1262             fiz1             = _mm256_add_pd(fiz1,tz);
1263
1264             fjx0             = _mm256_add_pd(fjx0,tx);
1265             fjy0             = _mm256_add_pd(fjy0,ty);
1266             fjz0             = _mm256_add_pd(fjz0,tz);
1267
1268             }
1269
1270             /**************************
1271              * CALCULATE INTERACTIONS *
1272              **************************/
1273
1274             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1275             {
1276
1277             r20              = _mm256_mul_pd(rsq20,rinv20);
1278
1279             /* Compute parameters for interactions between i and j atoms */
1280             qq20             = _mm256_mul_pd(iq2,jq0);
1281
1282             /* EWALD ELECTROSTATICS */
1283
1284             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1285             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1286             ewitab           = _mm256_cvttpd_epi32(ewrt);
1287             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1288             ewitab           = _mm_slli_epi32(ewitab,2);
1289             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1290             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1291             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1292             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1293             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1294             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1295             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1296             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
1297             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1298
1299             d                = _mm256_sub_pd(r20,rswitch);
1300             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1301             d2               = _mm256_mul_pd(d,d);
1302             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1303
1304             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1305
1306             /* Evaluate switch function */
1307             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1308             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv20,_mm256_mul_pd(velec,dsw)) );
1309             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1310
1311             fscal            = felec;
1312
1313             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1314
1315             /* Calculate temporary vectorial force */
1316             tx               = _mm256_mul_pd(fscal,dx20);
1317             ty               = _mm256_mul_pd(fscal,dy20);
1318             tz               = _mm256_mul_pd(fscal,dz20);
1319
1320             /* Update vectorial force */
1321             fix2             = _mm256_add_pd(fix2,tx);
1322             fiy2             = _mm256_add_pd(fiy2,ty);
1323             fiz2             = _mm256_add_pd(fiz2,tz);
1324
1325             fjx0             = _mm256_add_pd(fjx0,tx);
1326             fjy0             = _mm256_add_pd(fjy0,ty);
1327             fjz0             = _mm256_add_pd(fjz0,tz);
1328
1329             }
1330
1331             /**************************
1332              * CALCULATE INTERACTIONS *
1333              **************************/
1334
1335             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1336             {
1337
1338             r30              = _mm256_mul_pd(rsq30,rinv30);
1339
1340             /* Compute parameters for interactions between i and j atoms */
1341             qq30             = _mm256_mul_pd(iq3,jq0);
1342
1343             /* EWALD ELECTROSTATICS */
1344
1345             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1346             ewrt             = _mm256_mul_pd(r30,ewtabscale);
1347             ewitab           = _mm256_cvttpd_epi32(ewrt);
1348             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1349             ewitab           = _mm_slli_epi32(ewitab,2);
1350             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1351             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1352             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1353             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1354             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1355             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1356             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1357             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(rinv30,velec));
1358             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
1359
1360             d                = _mm256_sub_pd(r30,rswitch);
1361             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1362             d2               = _mm256_mul_pd(d,d);
1363             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1364
1365             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1366
1367             /* Evaluate switch function */
1368             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1369             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv30,_mm256_mul_pd(velec,dsw)) );
1370             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1371
1372             fscal            = felec;
1373
1374             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1375
1376             /* Calculate temporary vectorial force */
1377             tx               = _mm256_mul_pd(fscal,dx30);
1378             ty               = _mm256_mul_pd(fscal,dy30);
1379             tz               = _mm256_mul_pd(fscal,dz30);
1380
1381             /* Update vectorial force */
1382             fix3             = _mm256_add_pd(fix3,tx);
1383             fiy3             = _mm256_add_pd(fiy3,ty);
1384             fiz3             = _mm256_add_pd(fiz3,tz);
1385
1386             fjx0             = _mm256_add_pd(fjx0,tx);
1387             fjy0             = _mm256_add_pd(fjy0,ty);
1388             fjz0             = _mm256_add_pd(fjz0,tz);
1389
1390             }
1391
1392             fjptrA             = f+j_coord_offsetA;
1393             fjptrB             = f+j_coord_offsetB;
1394             fjptrC             = f+j_coord_offsetC;
1395             fjptrD             = f+j_coord_offsetD;
1396
1397             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1398
1399             /* Inner loop uses 245 flops */
1400         }
1401
1402         if(jidx<j_index_end)
1403         {
1404
1405             /* Get j neighbor index, and coordinate index */
1406             jnrlistA         = jjnr[jidx];
1407             jnrlistB         = jjnr[jidx+1];
1408             jnrlistC         = jjnr[jidx+2];
1409             jnrlistD         = jjnr[jidx+3];
1410             /* Sign of each element will be negative for non-real atoms.
1411              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1412              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1413              */
1414             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1415
1416             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1417             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1418             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1419
1420             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1421             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1422             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1423             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1424             j_coord_offsetA  = DIM*jnrA;
1425             j_coord_offsetB  = DIM*jnrB;
1426             j_coord_offsetC  = DIM*jnrC;
1427             j_coord_offsetD  = DIM*jnrD;
1428
1429             /* load j atom coordinates */
1430             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1431                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1432                                                  &jx0,&jy0,&jz0);
1433
1434             /* Calculate displacement vector */
1435             dx00             = _mm256_sub_pd(ix0,jx0);
1436             dy00             = _mm256_sub_pd(iy0,jy0);
1437             dz00             = _mm256_sub_pd(iz0,jz0);
1438             dx10             = _mm256_sub_pd(ix1,jx0);
1439             dy10             = _mm256_sub_pd(iy1,jy0);
1440             dz10             = _mm256_sub_pd(iz1,jz0);
1441             dx20             = _mm256_sub_pd(ix2,jx0);
1442             dy20             = _mm256_sub_pd(iy2,jy0);
1443             dz20             = _mm256_sub_pd(iz2,jz0);
1444             dx30             = _mm256_sub_pd(ix3,jx0);
1445             dy30             = _mm256_sub_pd(iy3,jy0);
1446             dz30             = _mm256_sub_pd(iz3,jz0);
1447
1448             /* Calculate squared distance and things based on it */
1449             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1450             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1451             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1452             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1453
1454             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1455             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1456             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1457             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1458
1459             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1460             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1461             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1462             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1463
1464             /* Load parameters for j particles */
1465             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1466                                                                  charge+jnrC+0,charge+jnrD+0);
1467             vdwjidx0A        = 2*vdwtype[jnrA+0];
1468             vdwjidx0B        = 2*vdwtype[jnrB+0];
1469             vdwjidx0C        = 2*vdwtype[jnrC+0];
1470             vdwjidx0D        = 2*vdwtype[jnrD+0];
1471
1472             fjx0             = _mm256_setzero_pd();
1473             fjy0             = _mm256_setzero_pd();
1474             fjz0             = _mm256_setzero_pd();
1475
1476             /**************************
1477              * CALCULATE INTERACTIONS *
1478              **************************/
1479
1480             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1481             {
1482
1483             r00              = _mm256_mul_pd(rsq00,rinv00);
1484             r00              = _mm256_andnot_pd(dummy_mask,r00);
1485
1486             /* Compute parameters for interactions between i and j atoms */
1487             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1488                                             vdwioffsetptr0+vdwjidx0B,
1489                                             vdwioffsetptr0+vdwjidx0C,
1490                                             vdwioffsetptr0+vdwjidx0D,
1491                                             &c6_00,&c12_00);
1492
1493             /* LENNARD-JONES DISPERSION/REPULSION */
1494
1495             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1496             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
1497             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
1498             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
1499             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
1500
1501             d                = _mm256_sub_pd(r00,rswitch);
1502             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1503             d2               = _mm256_mul_pd(d,d);
1504             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1505
1506             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1507
1508             /* Evaluate switch function */
1509             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1510             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
1511             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1512
1513             fscal            = fvdw;
1514
1515             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1516
1517             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1518
1519             /* Calculate temporary vectorial force */
1520             tx               = _mm256_mul_pd(fscal,dx00);
1521             ty               = _mm256_mul_pd(fscal,dy00);
1522             tz               = _mm256_mul_pd(fscal,dz00);
1523
1524             /* Update vectorial force */
1525             fix0             = _mm256_add_pd(fix0,tx);
1526             fiy0             = _mm256_add_pd(fiy0,ty);
1527             fiz0             = _mm256_add_pd(fiz0,tz);
1528
1529             fjx0             = _mm256_add_pd(fjx0,tx);
1530             fjy0             = _mm256_add_pd(fjy0,ty);
1531             fjz0             = _mm256_add_pd(fjz0,tz);
1532
1533             }
1534
1535             /**************************
1536              * CALCULATE INTERACTIONS *
1537              **************************/
1538
1539             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1540             {
1541
1542             r10              = _mm256_mul_pd(rsq10,rinv10);
1543             r10              = _mm256_andnot_pd(dummy_mask,r10);
1544
1545             /* Compute parameters for interactions between i and j atoms */
1546             qq10             = _mm256_mul_pd(iq1,jq0);
1547
1548             /* EWALD ELECTROSTATICS */
1549
1550             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1551             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1552             ewitab           = _mm256_cvttpd_epi32(ewrt);
1553             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1554             ewitab           = _mm_slli_epi32(ewitab,2);
1555             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1556             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1557             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1558             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1559             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1560             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1561             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1562             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
1563             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1564
1565             d                = _mm256_sub_pd(r10,rswitch);
1566             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1567             d2               = _mm256_mul_pd(d,d);
1568             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1569
1570             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1571
1572             /* Evaluate switch function */
1573             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1574             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv10,_mm256_mul_pd(velec,dsw)) );
1575             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1576
1577             fscal            = felec;
1578
1579             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1580
1581             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1582
1583             /* Calculate temporary vectorial force */
1584             tx               = _mm256_mul_pd(fscal,dx10);
1585             ty               = _mm256_mul_pd(fscal,dy10);
1586             tz               = _mm256_mul_pd(fscal,dz10);
1587
1588             /* Update vectorial force */
1589             fix1             = _mm256_add_pd(fix1,tx);
1590             fiy1             = _mm256_add_pd(fiy1,ty);
1591             fiz1             = _mm256_add_pd(fiz1,tz);
1592
1593             fjx0             = _mm256_add_pd(fjx0,tx);
1594             fjy0             = _mm256_add_pd(fjy0,ty);
1595             fjz0             = _mm256_add_pd(fjz0,tz);
1596
1597             }
1598
1599             /**************************
1600              * CALCULATE INTERACTIONS *
1601              **************************/
1602
1603             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1604             {
1605
1606             r20              = _mm256_mul_pd(rsq20,rinv20);
1607             r20              = _mm256_andnot_pd(dummy_mask,r20);
1608
1609             /* Compute parameters for interactions between i and j atoms */
1610             qq20             = _mm256_mul_pd(iq2,jq0);
1611
1612             /* EWALD ELECTROSTATICS */
1613
1614             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1615             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1616             ewitab           = _mm256_cvttpd_epi32(ewrt);
1617             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1618             ewitab           = _mm_slli_epi32(ewitab,2);
1619             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1620             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1621             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1622             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1623             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1624             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1625             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1626             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
1627             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1628
1629             d                = _mm256_sub_pd(r20,rswitch);
1630             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1631             d2               = _mm256_mul_pd(d,d);
1632             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1633
1634             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1635
1636             /* Evaluate switch function */
1637             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1638             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv20,_mm256_mul_pd(velec,dsw)) );
1639             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1640
1641             fscal            = felec;
1642
1643             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1644
1645             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1646
1647             /* Calculate temporary vectorial force */
1648             tx               = _mm256_mul_pd(fscal,dx20);
1649             ty               = _mm256_mul_pd(fscal,dy20);
1650             tz               = _mm256_mul_pd(fscal,dz20);
1651
1652             /* Update vectorial force */
1653             fix2             = _mm256_add_pd(fix2,tx);
1654             fiy2             = _mm256_add_pd(fiy2,ty);
1655             fiz2             = _mm256_add_pd(fiz2,tz);
1656
1657             fjx0             = _mm256_add_pd(fjx0,tx);
1658             fjy0             = _mm256_add_pd(fjy0,ty);
1659             fjz0             = _mm256_add_pd(fjz0,tz);
1660
1661             }
1662
1663             /**************************
1664              * CALCULATE INTERACTIONS *
1665              **************************/
1666
1667             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1668             {
1669
1670             r30              = _mm256_mul_pd(rsq30,rinv30);
1671             r30              = _mm256_andnot_pd(dummy_mask,r30);
1672
1673             /* Compute parameters for interactions between i and j atoms */
1674             qq30             = _mm256_mul_pd(iq3,jq0);
1675
1676             /* EWALD ELECTROSTATICS */
1677
1678             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1679             ewrt             = _mm256_mul_pd(r30,ewtabscale);
1680             ewitab           = _mm256_cvttpd_epi32(ewrt);
1681             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1682             ewitab           = _mm_slli_epi32(ewitab,2);
1683             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1684             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1685             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1686             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1687             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1688             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1689             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1690             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(rinv30,velec));
1691             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
1692
1693             d                = _mm256_sub_pd(r30,rswitch);
1694             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1695             d2               = _mm256_mul_pd(d,d);
1696             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1697
1698             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(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_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv30,_mm256_mul_pd(velec,dsw)) );
1703             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1704
1705             fscal            = felec;
1706
1707             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1708
1709             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1710
1711             /* Calculate temporary vectorial force */
1712             tx               = _mm256_mul_pd(fscal,dx30);
1713             ty               = _mm256_mul_pd(fscal,dy30);
1714             tz               = _mm256_mul_pd(fscal,dz30);
1715
1716             /* Update vectorial force */
1717             fix3             = _mm256_add_pd(fix3,tx);
1718             fiy3             = _mm256_add_pd(fiy3,ty);
1719             fiz3             = _mm256_add_pd(fiz3,tz);
1720
1721             fjx0             = _mm256_add_pd(fjx0,tx);
1722             fjy0             = _mm256_add_pd(fjy0,ty);
1723             fjz0             = _mm256_add_pd(fjz0,tz);
1724
1725             }
1726
1727             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1728             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1729             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1730             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1731
1732             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1733
1734             /* Inner loop uses 249 flops */
1735         }
1736
1737         /* End of innermost loop */
1738
1739         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1740                                                  f+i_coord_offset,fshift+i_shift_offset);
1741
1742         /* Increment number of inner iterations */
1743         inneriter                  += j_index_end - j_index_start;
1744
1745         /* Outer loop uses 24 flops */
1746     }
1747
1748     /* Increment number of outer iterations */
1749     outeriter        += nri;
1750
1751     /* Update outer/inner flops */
1752
1753     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*249);
1754 }