src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEwSw_VdwNone_GeomW3P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomW3P1_VF_avx_256_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            None
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSw_VdwNone_GeomW3P1_VF_avx_256_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrE,jnrF,jnrG,jnrH;
  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              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  83     real             scratch[4*DIM];
  84     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     real *           vdwioffsetptr0;
  86     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     real *           vdwioffsetptr1;
  88     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     real *           vdwioffsetptr2;
  90     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  92     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     __m256i          ewitab;
  99     __m128i          ewitab_lo,ewitab_hi;
 100     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 101     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 102     real             *ewtab;
 103     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 104     real             rswitch_scalar,d_scalar;
 105     __m256           dummy_mask,cutoff_mask;
 106     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 107     __m256           one     = _mm256_set1_ps(1.0);
 108     __m256           two     = _mm256_set1_ps(2.0);
 109     x                = xx[0];
 110     f                = ff[0];
 111
 112     nri              = nlist->nri;
 113     iinr             = nlist->iinr;
 114     jindex           = nlist->jindex;
 115     jjnr             = nlist->jjnr;
 116     shiftidx         = nlist->shift;
 117     gid              = nlist->gid;
 118     shiftvec         = fr->shift_vec[0];
 119     fshift           = fr->fshift[0];
 120     facel            = _mm256_set1_ps(fr->ic->epsfac);
 121     charge           = mdatoms->chargeA;
 122
 123     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 124     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 125     beta2            = _mm256_mul_ps(beta,beta);
 126     beta3            = _mm256_mul_ps(beta,beta2);
 127
 128     ewtab            = fr->ic->tabq_coul_FDV0;
 129     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 130     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 131
 132     /* Setup water-specific parameters */
 133     inr              = nlist->iinr[0];
 134     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 135     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 136     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 137
 138     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 139     rcutoff_scalar   = fr->ic->rcoulomb;
 140     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 141     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 142
 143     rswitch_scalar   = fr->ic->rcoulomb_switch;
 144     rswitch          = _mm256_set1_ps(rswitch_scalar);
 145     /* Setup switch parameters */
 146     d_scalar         = rcutoff_scalar-rswitch_scalar;
 147     d                = _mm256_set1_ps(d_scalar);
 148     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 149     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 150     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 151     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 152     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 153     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 154
 155     /* Avoid stupid compiler warnings */
 156     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 157     j_coord_offsetA = 0;
 158     j_coord_offsetB = 0;
 159     j_coord_offsetC = 0;
 160     j_coord_offsetD = 0;
 161     j_coord_offsetE = 0;
 162     j_coord_offsetF = 0;
 163     j_coord_offsetG = 0;
 164     j_coord_offsetH = 0;
 165
 166     outeriter        = 0;
 167     inneriter        = 0;
 168
 169     for(iidx=0;iidx<4*DIM;iidx++)
 170     {
 171         scratch[iidx] = 0.0;
 172     }
 173
 174     /* Start outer loop over neighborlists */
 175     for(iidx=0; iidx<nri; iidx++)
 176     {
 177         /* Load shift vector for this list */
 178         i_shift_offset   = DIM*shiftidx[iidx];
 179
 180         /* Load limits for loop over neighbors */
 181         j_index_start    = jindex[iidx];
 182         j_index_end      = jindex[iidx+1];
 183
 184         /* Get outer coordinate index */
 185         inr              = iinr[iidx];
 186         i_coord_offset   = DIM*inr;
 187
 188         /* Load i particle coords and add shift vector */
 189         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 190                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 191
 192         fix0             = _mm256_setzero_ps();
 193         fiy0             = _mm256_setzero_ps();
 194         fiz0             = _mm256_setzero_ps();
 195         fix1             = _mm256_setzero_ps();
 196         fiy1             = _mm256_setzero_ps();
 197         fiz1             = _mm256_setzero_ps();
 198         fix2             = _mm256_setzero_ps();
 199         fiy2             = _mm256_setzero_ps();
 200         fiz2             = _mm256_setzero_ps();
 201
 202         /* Reset potential sums */
 203         velecsum         = _mm256_setzero_ps();
 204
 205         /* Start inner kernel loop */
 206         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 207         {
 208
 209             /* Get j neighbor index, and coordinate index */
 210             jnrA             = jjnr[jidx];
 211             jnrB             = jjnr[jidx+1];
 212             jnrC             = jjnr[jidx+2];
 213             jnrD             = jjnr[jidx+3];
 214             jnrE             = jjnr[jidx+4];
 215             jnrF             = jjnr[jidx+5];
 216             jnrG             = jjnr[jidx+6];
 217             jnrH             = jjnr[jidx+7];
 218             j_coord_offsetA  = DIM*jnrA;
 219             j_coord_offsetB  = DIM*jnrB;
 220             j_coord_offsetC  = DIM*jnrC;
 221             j_coord_offsetD  = DIM*jnrD;
 222             j_coord_offsetE  = DIM*jnrE;
 223             j_coord_offsetF  = DIM*jnrF;
 224             j_coord_offsetG  = DIM*jnrG;
 225             j_coord_offsetH  = DIM*jnrH;
 226
 227             /* load j atom coordinates */
 228             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 229                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 230                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 231                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 232                                                  &jx0,&jy0,&jz0);
 233
 234             /* Calculate displacement vector */
 235             dx00             = _mm256_sub_ps(ix0,jx0);
 236             dy00             = _mm256_sub_ps(iy0,jy0);
 237             dz00             = _mm256_sub_ps(iz0,jz0);
 238             dx10             = _mm256_sub_ps(ix1,jx0);
 239             dy10             = _mm256_sub_ps(iy1,jy0);
 240             dz10             = _mm256_sub_ps(iz1,jz0);
 241             dx20             = _mm256_sub_ps(ix2,jx0);
 242             dy20             = _mm256_sub_ps(iy2,jy0);
 243             dz20             = _mm256_sub_ps(iz2,jz0);
 244
 245             /* Calculate squared distance and things based on it */
 246             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 247             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 248             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 249
 250             rinv00           = avx256_invsqrt_f(rsq00);
 251             rinv10           = avx256_invsqrt_f(rsq10);
 252             rinv20           = avx256_invsqrt_f(rsq20);
 253
 254             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 255             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 256             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 257
 258             /* Load parameters for j particles */
 259             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 260                                                                  charge+jnrC+0,charge+jnrD+0,
 261                                                                  charge+jnrE+0,charge+jnrF+0,
 262                                                                  charge+jnrG+0,charge+jnrH+0);
 263
 264             fjx0             = _mm256_setzero_ps();
 265             fjy0             = _mm256_setzero_ps();
 266             fjz0             = _mm256_setzero_ps();
 267
 268             /**************************
 269              * CALCULATE INTERACTIONS *
 270              **************************/
 271
 272             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 273             {
 274
 275             r00              = _mm256_mul_ps(rsq00,rinv00);
 276
 277             /* Compute parameters for interactions between i and j atoms */
 278             qq00             = _mm256_mul_ps(iq0,jq0);
 279
 280             /* EWALD ELECTROSTATICS */
 281
 282             /* Analytical PME correction */
 283             zeta2            = _mm256_mul_ps(beta2,rsq00);
 284             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 285             pmecorrF         = avx256_pmecorrF_f(zeta2);
 286             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 287             felec            = _mm256_mul_ps(qq00,felec);
 288             pmecorrV         = avx256_pmecorrV_f(zeta2);
 289             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 290             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 291             velec            = _mm256_mul_ps(qq00,velec);
 292
 293             d                = _mm256_sub_ps(r00,rswitch);
 294             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 295             d2               = _mm256_mul_ps(d,d);
 296             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 297
 298             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 299
 300             /* Evaluate switch function */
 301             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 302             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
 303             velec            = _mm256_mul_ps(velec,sw);
 304             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 305
 306             /* Update potential sum for this i atom from the interaction with this j atom. */
 307             velec            = _mm256_and_ps(velec,cutoff_mask);
 308             velecsum         = _mm256_add_ps(velecsum,velec);
 309
 310             fscal            = felec;
 311
 312             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 313
 314             /* Calculate temporary vectorial force */
 315             tx               = _mm256_mul_ps(fscal,dx00);
 316             ty               = _mm256_mul_ps(fscal,dy00);
 317             tz               = _mm256_mul_ps(fscal,dz00);
 318
 319             /* Update vectorial force */
 320             fix0             = _mm256_add_ps(fix0,tx);
 321             fiy0             = _mm256_add_ps(fiy0,ty);
 322             fiz0             = _mm256_add_ps(fiz0,tz);
 323
 324             fjx0             = _mm256_add_ps(fjx0,tx);
 325             fjy0             = _mm256_add_ps(fjy0,ty);
 326             fjz0             = _mm256_add_ps(fjz0,tz);
 327
 328             }
 329
 330             /**************************
 331              * CALCULATE INTERACTIONS *
 332              **************************/
 333
 334             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 335             {
 336
 337             r10              = _mm256_mul_ps(rsq10,rinv10);
 338
 339             /* Compute parameters for interactions between i and j atoms */
 340             qq10             = _mm256_mul_ps(iq1,jq0);
 341
 342             /* EWALD ELECTROSTATICS */
 343
 344             /* Analytical PME correction */
 345             zeta2            = _mm256_mul_ps(beta2,rsq10);
 346             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 347             pmecorrF         = avx256_pmecorrF_f(zeta2);
 348             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 349             felec            = _mm256_mul_ps(qq10,felec);
 350             pmecorrV         = avx256_pmecorrV_f(zeta2);
 351             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 352             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 353             velec            = _mm256_mul_ps(qq10,velec);
 354
 355             d                = _mm256_sub_ps(r10,rswitch);
 356             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 357             d2               = _mm256_mul_ps(d,d);
 358             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 359
 360             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 361
 362             /* Evaluate switch function */
 363             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 364             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv10,_mm256_mul_ps(velec,dsw)) );
 365             velec            = _mm256_mul_ps(velec,sw);
 366             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 367
 368             /* Update potential sum for this i atom from the interaction with this j atom. */
 369             velec            = _mm256_and_ps(velec,cutoff_mask);
 370             velecsum         = _mm256_add_ps(velecsum,velec);
 371
 372             fscal            = felec;
 373
 374             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 375
 376             /* Calculate temporary vectorial force */
 377             tx               = _mm256_mul_ps(fscal,dx10);
 378             ty               = _mm256_mul_ps(fscal,dy10);
 379             tz               = _mm256_mul_ps(fscal,dz10);
 380
 381             /* Update vectorial force */
 382             fix1             = _mm256_add_ps(fix1,tx);
 383             fiy1             = _mm256_add_ps(fiy1,ty);
 384             fiz1             = _mm256_add_ps(fiz1,tz);
 385
 386             fjx0             = _mm256_add_ps(fjx0,tx);
 387             fjy0             = _mm256_add_ps(fjy0,ty);
 388             fjz0             = _mm256_add_ps(fjz0,tz);
 389
 390             }
 391
 392             /**************************
 393              * CALCULATE INTERACTIONS *
 394              **************************/
 395
 396             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 397             {
 398
 399             r20              = _mm256_mul_ps(rsq20,rinv20);
 400
 401             /* Compute parameters for interactions between i and j atoms */
 402             qq20             = _mm256_mul_ps(iq2,jq0);
 403
 404             /* EWALD ELECTROSTATICS */
 405
 406             /* Analytical PME correction */
 407             zeta2            = _mm256_mul_ps(beta2,rsq20);
 408             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 409             pmecorrF         = avx256_pmecorrF_f(zeta2);
 410             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 411             felec            = _mm256_mul_ps(qq20,felec);
 412             pmecorrV         = avx256_pmecorrV_f(zeta2);
 413             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 414             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 415             velec            = _mm256_mul_ps(qq20,velec);
 416
 417             d                = _mm256_sub_ps(r20,rswitch);
 418             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 419             d2               = _mm256_mul_ps(d,d);
 420             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 421
 422             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 423
 424             /* Evaluate switch function */
 425             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 426             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv20,_mm256_mul_ps(velec,dsw)) );
 427             velec            = _mm256_mul_ps(velec,sw);
 428             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 429
 430             /* Update potential sum for this i atom from the interaction with this j atom. */
 431             velec            = _mm256_and_ps(velec,cutoff_mask);
 432             velecsum         = _mm256_add_ps(velecsum,velec);
 433
 434             fscal            = felec;
 435
 436             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 437
 438             /* Calculate temporary vectorial force */
 439             tx               = _mm256_mul_ps(fscal,dx20);
 440             ty               = _mm256_mul_ps(fscal,dy20);
 441             tz               = _mm256_mul_ps(fscal,dz20);
 442
 443             /* Update vectorial force */
 444             fix2             = _mm256_add_ps(fix2,tx);
 445             fiy2             = _mm256_add_ps(fiy2,ty);
 446             fiz2             = _mm256_add_ps(fiz2,tz);
 447
 448             fjx0             = _mm256_add_ps(fjx0,tx);
 449             fjy0             = _mm256_add_ps(fjy0,ty);
 450             fjz0             = _mm256_add_ps(fjz0,tz);
 451
 452             }
 453
 454             fjptrA             = f+j_coord_offsetA;
 455             fjptrB             = f+j_coord_offsetB;
 456             fjptrC             = f+j_coord_offsetC;
 457             fjptrD             = f+j_coord_offsetD;
 458             fjptrE             = f+j_coord_offsetE;
 459             fjptrF             = f+j_coord_offsetF;
 460             fjptrG             = f+j_coord_offsetG;
 461             fjptrH             = f+j_coord_offsetH;
 462
 463             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 464
 465             /* Inner loop uses 327 flops */
 466         }
 467
 468         if(jidx<j_index_end)
 469         {
 470
 471             /* Get j neighbor index, and coordinate index */
 472             jnrlistA         = jjnr[jidx];
 473             jnrlistB         = jjnr[jidx+1];
 474             jnrlistC         = jjnr[jidx+2];
 475             jnrlistD         = jjnr[jidx+3];
 476             jnrlistE         = jjnr[jidx+4];
 477             jnrlistF         = jjnr[jidx+5];
 478             jnrlistG         = jjnr[jidx+6];
 479             jnrlistH         = jjnr[jidx+7];
 480             /* Sign of each element will be negative for non-real atoms.
 481              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 482              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 483              */
 484             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 485                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 486
 487             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 488             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 489             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 490             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 491             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 492             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 493             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 494             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 495             j_coord_offsetA  = DIM*jnrA;
 496             j_coord_offsetB  = DIM*jnrB;
 497             j_coord_offsetC  = DIM*jnrC;
 498             j_coord_offsetD  = DIM*jnrD;
 499             j_coord_offsetE  = DIM*jnrE;
 500             j_coord_offsetF  = DIM*jnrF;
 501             j_coord_offsetG  = DIM*jnrG;
 502             j_coord_offsetH  = DIM*jnrH;
 503
 504             /* load j atom coordinates */
 505             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 506                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 507                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 508                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 509                                                  &jx0,&jy0,&jz0);
 510
 511             /* Calculate displacement vector */
 512             dx00             = _mm256_sub_ps(ix0,jx0);
 513             dy00             = _mm256_sub_ps(iy0,jy0);
 514             dz00             = _mm256_sub_ps(iz0,jz0);
 515             dx10             = _mm256_sub_ps(ix1,jx0);
 516             dy10             = _mm256_sub_ps(iy1,jy0);
 517             dz10             = _mm256_sub_ps(iz1,jz0);
 518             dx20             = _mm256_sub_ps(ix2,jx0);
 519             dy20             = _mm256_sub_ps(iy2,jy0);
 520             dz20             = _mm256_sub_ps(iz2,jz0);
 521
 522             /* Calculate squared distance and things based on it */
 523             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 524             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 525             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 526
 527             rinv00           = avx256_invsqrt_f(rsq00);
 528             rinv10           = avx256_invsqrt_f(rsq10);
 529             rinv20           = avx256_invsqrt_f(rsq20);
 530
 531             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 532             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 533             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 534
 535             /* Load parameters for j particles */
 536             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 537                                                                  charge+jnrC+0,charge+jnrD+0,
 538                                                                  charge+jnrE+0,charge+jnrF+0,
 539                                                                  charge+jnrG+0,charge+jnrH+0);
 540
 541             fjx0             = _mm256_setzero_ps();
 542             fjy0             = _mm256_setzero_ps();
 543             fjz0             = _mm256_setzero_ps();
 544
 545             /**************************
 546              * CALCULATE INTERACTIONS *
 547              **************************/
 548
 549             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 550             {
 551
 552             r00              = _mm256_mul_ps(rsq00,rinv00);
 553             r00              = _mm256_andnot_ps(dummy_mask,r00);
 554
 555             /* Compute parameters for interactions between i and j atoms */
 556             qq00             = _mm256_mul_ps(iq0,jq0);
 557
 558             /* EWALD ELECTROSTATICS */
 559
 560             /* Analytical PME correction */
 561             zeta2            = _mm256_mul_ps(beta2,rsq00);
 562             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 563             pmecorrF         = avx256_pmecorrF_f(zeta2);
 564             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 565             felec            = _mm256_mul_ps(qq00,felec);
 566             pmecorrV         = avx256_pmecorrV_f(zeta2);
 567             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 568             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 569             velec            = _mm256_mul_ps(qq00,velec);
 570
 571             d                = _mm256_sub_ps(r00,rswitch);
 572             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 573             d2               = _mm256_mul_ps(d,d);
 574             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 575
 576             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 577
 578             /* Evaluate switch function */
 579             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 580             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
 581             velec            = _mm256_mul_ps(velec,sw);
 582             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 583
 584             /* Update potential sum for this i atom from the interaction with this j atom. */
 585             velec            = _mm256_and_ps(velec,cutoff_mask);
 586             velec            = _mm256_andnot_ps(dummy_mask,velec);
 587             velecsum         = _mm256_add_ps(velecsum,velec);
 588
 589             fscal            = felec;
 590
 591             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 592
 593             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 594
 595             /* Calculate temporary vectorial force */
 596             tx               = _mm256_mul_ps(fscal,dx00);
 597             ty               = _mm256_mul_ps(fscal,dy00);
 598             tz               = _mm256_mul_ps(fscal,dz00);
 599
 600             /* Update vectorial force */
 601             fix0             = _mm256_add_ps(fix0,tx);
 602             fiy0             = _mm256_add_ps(fiy0,ty);
 603             fiz0             = _mm256_add_ps(fiz0,tz);
 604
 605             fjx0             = _mm256_add_ps(fjx0,tx);
 606             fjy0             = _mm256_add_ps(fjy0,ty);
 607             fjz0             = _mm256_add_ps(fjz0,tz);
 608
 609             }
 610
 611             /**************************
 612              * CALCULATE INTERACTIONS *
 613              **************************/
 614
 615             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 616             {
 617
 618             r10              = _mm256_mul_ps(rsq10,rinv10);
 619             r10              = _mm256_andnot_ps(dummy_mask,r10);
 620
 621             /* Compute parameters for interactions between i and j atoms */
 622             qq10             = _mm256_mul_ps(iq1,jq0);
 623
 624             /* EWALD ELECTROSTATICS */
 625
 626             /* Analytical PME correction */
 627             zeta2            = _mm256_mul_ps(beta2,rsq10);
 628             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 629             pmecorrF         = avx256_pmecorrF_f(zeta2);
 630             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 631             felec            = _mm256_mul_ps(qq10,felec);
 632             pmecorrV         = avx256_pmecorrV_f(zeta2);
 633             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 634             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 635             velec            = _mm256_mul_ps(qq10,velec);
 636
 637             d                = _mm256_sub_ps(r10,rswitch);
 638             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 639             d2               = _mm256_mul_ps(d,d);
 640             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 641
 642             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 643
 644             /* Evaluate switch function */
 645             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 646             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv10,_mm256_mul_ps(velec,dsw)) );
 647             velec            = _mm256_mul_ps(velec,sw);
 648             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 649
 650             /* Update potential sum for this i atom from the interaction with this j atom. */
 651             velec            = _mm256_and_ps(velec,cutoff_mask);
 652             velec            = _mm256_andnot_ps(dummy_mask,velec);
 653             velecsum         = _mm256_add_ps(velecsum,velec);
 654
 655             fscal            = felec;
 656
 657             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 658
 659             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 660
 661             /* Calculate temporary vectorial force */
 662             tx               = _mm256_mul_ps(fscal,dx10);
 663             ty               = _mm256_mul_ps(fscal,dy10);
 664             tz               = _mm256_mul_ps(fscal,dz10);
 665
 666             /* Update vectorial force */
 667             fix1             = _mm256_add_ps(fix1,tx);
 668             fiy1             = _mm256_add_ps(fiy1,ty);
 669             fiz1             = _mm256_add_ps(fiz1,tz);
 670
 671             fjx0             = _mm256_add_ps(fjx0,tx);
 672             fjy0             = _mm256_add_ps(fjy0,ty);
 673             fjz0             = _mm256_add_ps(fjz0,tz);
 674
 675             }
 676
 677             /**************************
 678              * CALCULATE INTERACTIONS *
 679              **************************/
 680
 681             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 682             {
 683
 684             r20              = _mm256_mul_ps(rsq20,rinv20);
 685             r20              = _mm256_andnot_ps(dummy_mask,r20);
 686
 687             /* Compute parameters for interactions between i and j atoms */
 688             qq20             = _mm256_mul_ps(iq2,jq0);
 689
 690             /* EWALD ELECTROSTATICS */
 691
 692             /* Analytical PME correction */
 693             zeta2            = _mm256_mul_ps(beta2,rsq20);
 694             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 695             pmecorrF         = avx256_pmecorrF_f(zeta2);
 696             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 697             felec            = _mm256_mul_ps(qq20,felec);
 698             pmecorrV         = avx256_pmecorrV_f(zeta2);
 699             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 700             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 701             velec            = _mm256_mul_ps(qq20,velec);
 702
 703             d                = _mm256_sub_ps(r20,rswitch);
 704             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 705             d2               = _mm256_mul_ps(d,d);
 706             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 707
 708             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 709
 710             /* Evaluate switch function */
 711             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 712             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv20,_mm256_mul_ps(velec,dsw)) );
 713             velec            = _mm256_mul_ps(velec,sw);
 714             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 715
 716             /* Update potential sum for this i atom from the interaction with this j atom. */
 717             velec            = _mm256_and_ps(velec,cutoff_mask);
 718             velec            = _mm256_andnot_ps(dummy_mask,velec);
 719             velecsum         = _mm256_add_ps(velecsum,velec);
 720
 721             fscal            = felec;
 722
 723             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 724
 725             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 726
 727             /* Calculate temporary vectorial force */
 728             tx               = _mm256_mul_ps(fscal,dx20);
 729             ty               = _mm256_mul_ps(fscal,dy20);
 730             tz               = _mm256_mul_ps(fscal,dz20);
 731
 732             /* Update vectorial force */
 733             fix2             = _mm256_add_ps(fix2,tx);
 734             fiy2             = _mm256_add_ps(fiy2,ty);
 735             fiz2             = _mm256_add_ps(fiz2,tz);
 736
 737             fjx0             = _mm256_add_ps(fjx0,tx);
 738             fjy0             = _mm256_add_ps(fjy0,ty);
 739             fjz0             = _mm256_add_ps(fjz0,tz);
 740
 741             }
 742
 743             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 744             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 745             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 746             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 747             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 748             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 749             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 750             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 751
 752             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 753
 754             /* Inner loop uses 330 flops */
 755         }
 756
 757         /* End of innermost loop */
 758
 759         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 760                                                  f+i_coord_offset,fshift+i_shift_offset);
 761
 762         ggid                        = gid[iidx];
 763         /* Update potential energies */
 764         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 765
 766         /* Increment number of inner iterations */
 767         inneriter                  += j_index_end - j_index_start;
 768
 769         /* Outer loop uses 19 flops */
 770     }
 771
 772     /* Increment number of outer iterations */
 773     outeriter        += nri;
 774
 775     /* Update outer/inner flops */
 776
 777     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*330);
 778 }
 779 /*
 780  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomW3P1_F_avx_256_single
 781  * Electrostatics interaction: Ewald
 782  * VdW interaction:            None
 783  * Geometry:                   Water3-Particle
 784  * Calculate force/pot:        Force
 785  */
 786 void
 787 nb_kernel_ElecEwSw_VdwNone_GeomW3P1_F_avx_256_single
 788                     (t_nblist                    * gmx_restrict       nlist,
 789                      rvec                        * gmx_restrict          xx,
 790                      rvec                        * gmx_restrict          ff,
 791                      struct t_forcerec           * gmx_restrict          fr,
 792                      t_mdatoms                   * gmx_restrict     mdatoms,
 793                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 794                      t_nrnb                      * gmx_restrict        nrnb)
 795 {
 796     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 797      * just 0 for non-waters.
 798      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 799      * jnr indices corresponding to data put in the four positions in the SIMD register.
 800      */
 801     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 802     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 803     int              jnrA,jnrB,jnrC,jnrD;
 804     int              jnrE,jnrF,jnrG,jnrH;
 805     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 806     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 807     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 808     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 809     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 810     real             rcutoff_scalar;
 811     real             *shiftvec,*fshift,*x,*f;
 812     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 813     real             scratch[4*DIM];
 814     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 815     real *           vdwioffsetptr0;
 816     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 817     real *           vdwioffsetptr1;
 818     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 819     real *           vdwioffsetptr2;
 820     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 821     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 822     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 823     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 824     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 825     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 826     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 827     real             *charge;
 828     __m256i          ewitab;
 829     __m128i          ewitab_lo,ewitab_hi;
 830     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 831     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 832     real             *ewtab;
 833     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 834     real             rswitch_scalar,d_scalar;
 835     __m256           dummy_mask,cutoff_mask;
 836     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 837     __m256           one     = _mm256_set1_ps(1.0);
 838     __m256           two     = _mm256_set1_ps(2.0);
 839     x                = xx[0];
 840     f                = ff[0];
 841
 842     nri              = nlist->nri;
 843     iinr             = nlist->iinr;
 844     jindex           = nlist->jindex;
 845     jjnr             = nlist->jjnr;
 846     shiftidx         = nlist->shift;
 847     gid              = nlist->gid;
 848     shiftvec         = fr->shift_vec[0];
 849     fshift           = fr->fshift[0];
 850     facel            = _mm256_set1_ps(fr->ic->epsfac);
 851     charge           = mdatoms->chargeA;
 852
 853     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 854     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 855     beta2            = _mm256_mul_ps(beta,beta);
 856     beta3            = _mm256_mul_ps(beta,beta2);
 857
 858     ewtab            = fr->ic->tabq_coul_FDV0;
 859     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 860     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 861
 862     /* Setup water-specific parameters */
 863     inr              = nlist->iinr[0];
 864     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 865     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 866     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 867
 868     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 869     rcutoff_scalar   = fr->ic->rcoulomb;
 870     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 871     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 872
 873     rswitch_scalar   = fr->ic->rcoulomb_switch;
 874     rswitch          = _mm256_set1_ps(rswitch_scalar);
 875     /* Setup switch parameters */
 876     d_scalar         = rcutoff_scalar-rswitch_scalar;
 877     d                = _mm256_set1_ps(d_scalar);
 878     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 879     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 880     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 881     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 882     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 883     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 884
 885     /* Avoid stupid compiler warnings */
 886     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 887     j_coord_offsetA = 0;
 888     j_coord_offsetB = 0;
 889     j_coord_offsetC = 0;
 890     j_coord_offsetD = 0;
 891     j_coord_offsetE = 0;
 892     j_coord_offsetF = 0;
 893     j_coord_offsetG = 0;
 894     j_coord_offsetH = 0;
 895
 896     outeriter        = 0;
 897     inneriter        = 0;
 898
 899     for(iidx=0;iidx<4*DIM;iidx++)
 900     {
 901         scratch[iidx] = 0.0;
 902     }
 903
 904     /* Start outer loop over neighborlists */
 905     for(iidx=0; iidx<nri; iidx++)
 906     {
 907         /* Load shift vector for this list */
 908         i_shift_offset   = DIM*shiftidx[iidx];
 909
 910         /* Load limits for loop over neighbors */
 911         j_index_start    = jindex[iidx];
 912         j_index_end      = jindex[iidx+1];
 913
 914         /* Get outer coordinate index */
 915         inr              = iinr[iidx];
 916         i_coord_offset   = DIM*inr;
 917
 918         /* Load i particle coords and add shift vector */
 919         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 920                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 921
 922         fix0             = _mm256_setzero_ps();
 923         fiy0             = _mm256_setzero_ps();
 924         fiz0             = _mm256_setzero_ps();
 925         fix1             = _mm256_setzero_ps();
 926         fiy1             = _mm256_setzero_ps();
 927         fiz1             = _mm256_setzero_ps();
 928         fix2             = _mm256_setzero_ps();
 929         fiy2             = _mm256_setzero_ps();
 930         fiz2             = _mm256_setzero_ps();
 931
 932         /* Start inner kernel loop */
 933         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 934         {
 935
 936             /* Get j neighbor index, and coordinate index */
 937             jnrA             = jjnr[jidx];
 938             jnrB             = jjnr[jidx+1];
 939             jnrC             = jjnr[jidx+2];
 940             jnrD             = jjnr[jidx+3];
 941             jnrE             = jjnr[jidx+4];
 942             jnrF             = jjnr[jidx+5];
 943             jnrG             = jjnr[jidx+6];
 944             jnrH             = jjnr[jidx+7];
 945             j_coord_offsetA  = DIM*jnrA;
 946             j_coord_offsetB  = DIM*jnrB;
 947             j_coord_offsetC  = DIM*jnrC;
 948             j_coord_offsetD  = DIM*jnrD;
 949             j_coord_offsetE  = DIM*jnrE;
 950             j_coord_offsetF  = DIM*jnrF;
 951             j_coord_offsetG  = DIM*jnrG;
 952             j_coord_offsetH  = DIM*jnrH;
 953
 954             /* load j atom coordinates */
 955             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 956                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 957                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 958                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 959                                                  &jx0,&jy0,&jz0);
 960
 961             /* Calculate displacement vector */
 962             dx00             = _mm256_sub_ps(ix0,jx0);
 963             dy00             = _mm256_sub_ps(iy0,jy0);
 964             dz00             = _mm256_sub_ps(iz0,jz0);
 965             dx10             = _mm256_sub_ps(ix1,jx0);
 966             dy10             = _mm256_sub_ps(iy1,jy0);
 967             dz10             = _mm256_sub_ps(iz1,jz0);
 968             dx20             = _mm256_sub_ps(ix2,jx0);
 969             dy20             = _mm256_sub_ps(iy2,jy0);
 970             dz20             = _mm256_sub_ps(iz2,jz0);
 971
 972             /* Calculate squared distance and things based on it */
 973             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 974             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 975             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 976
 977             rinv00           = avx256_invsqrt_f(rsq00);
 978             rinv10           = avx256_invsqrt_f(rsq10);
 979             rinv20           = avx256_invsqrt_f(rsq20);
 980
 981             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 982             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 983             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 984
 985             /* Load parameters for j particles */
 986             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 987                                                                  charge+jnrC+0,charge+jnrD+0,
 988                                                                  charge+jnrE+0,charge+jnrF+0,
 989                                                                  charge+jnrG+0,charge+jnrH+0);
 990
 991             fjx0             = _mm256_setzero_ps();
 992             fjy0             = _mm256_setzero_ps();
 993             fjz0             = _mm256_setzero_ps();
 994
 995             /**************************
 996              * CALCULATE INTERACTIONS *
 997              **************************/
 998
 999             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1000             {
1001
1002             r00              = _mm256_mul_ps(rsq00,rinv00);
1003
1004             /* Compute parameters for interactions between i and j atoms */
1005             qq00             = _mm256_mul_ps(iq0,jq0);
1006
1007             /* EWALD ELECTROSTATICS */
1008
1009             /* Analytical PME correction */
1010             zeta2            = _mm256_mul_ps(beta2,rsq00);
1011             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1012             pmecorrF         = avx256_pmecorrF_f(zeta2);
1013             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1014             felec            = _mm256_mul_ps(qq00,felec);
1015             pmecorrV         = avx256_pmecorrV_f(zeta2);
1016             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1017             velec            = _mm256_sub_ps(rinv00,pmecorrV);
1018             velec            = _mm256_mul_ps(qq00,velec);
1019
1020             d                = _mm256_sub_ps(r00,rswitch);
1021             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1022             d2               = _mm256_mul_ps(d,d);
1023             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
1024
1025             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1026
1027             /* Evaluate switch function */
1028             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1029             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
1030             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1031
1032             fscal            = felec;
1033
1034             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1035
1036             /* Calculate temporary vectorial force */
1037             tx               = _mm256_mul_ps(fscal,dx00);
1038             ty               = _mm256_mul_ps(fscal,dy00);
1039             tz               = _mm256_mul_ps(fscal,dz00);
1040
1041             /* Update vectorial force */
1042             fix0             = _mm256_add_ps(fix0,tx);
1043             fiy0             = _mm256_add_ps(fiy0,ty);
1044             fiz0             = _mm256_add_ps(fiz0,tz);
1045
1046             fjx0             = _mm256_add_ps(fjx0,tx);
1047             fjy0             = _mm256_add_ps(fjy0,ty);
1048             fjz0             = _mm256_add_ps(fjz0,tz);
1049
1050             }
1051
1052             /**************************
1053              * CALCULATE INTERACTIONS *
1054              **************************/
1055
1056             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1057             {
1058
1059             r10              = _mm256_mul_ps(rsq10,rinv10);
1060
1061             /* Compute parameters for interactions between i and j atoms */
1062             qq10             = _mm256_mul_ps(iq1,jq0);
1063
1064             /* EWALD ELECTROSTATICS */
1065
1066             /* Analytical PME correction */
1067             zeta2            = _mm256_mul_ps(beta2,rsq10);
1068             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1069             pmecorrF         = avx256_pmecorrF_f(zeta2);
1070             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1071             felec            = _mm256_mul_ps(qq10,felec);
1072             pmecorrV         = avx256_pmecorrV_f(zeta2);
1073             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1074             velec            = _mm256_sub_ps(rinv10,pmecorrV);
1075             velec            = _mm256_mul_ps(qq10,velec);
1076
1077             d                = _mm256_sub_ps(r10,rswitch);
1078             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1079             d2               = _mm256_mul_ps(d,d);
1080             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
1081
1082             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1083
1084             /* Evaluate switch function */
1085             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1086             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv10,_mm256_mul_ps(velec,dsw)) );
1087             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1088
1089             fscal            = felec;
1090
1091             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1092
1093             /* Calculate temporary vectorial force */
1094             tx               = _mm256_mul_ps(fscal,dx10);
1095             ty               = _mm256_mul_ps(fscal,dy10);
1096             tz               = _mm256_mul_ps(fscal,dz10);
1097
1098             /* Update vectorial force */
1099             fix1             = _mm256_add_ps(fix1,tx);
1100             fiy1             = _mm256_add_ps(fiy1,ty);
1101             fiz1             = _mm256_add_ps(fiz1,tz);
1102
1103             fjx0             = _mm256_add_ps(fjx0,tx);
1104             fjy0             = _mm256_add_ps(fjy0,ty);
1105             fjz0             = _mm256_add_ps(fjz0,tz);
1106
1107             }
1108
1109             /**************************
1110              * CALCULATE INTERACTIONS *
1111              **************************/
1112
1113             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1114             {
1115
1116             r20              = _mm256_mul_ps(rsq20,rinv20);
1117
1118             /* Compute parameters for interactions between i and j atoms */
1119             qq20             = _mm256_mul_ps(iq2,jq0);
1120
1121             /* EWALD ELECTROSTATICS */
1122
1123             /* Analytical PME correction */
1124             zeta2            = _mm256_mul_ps(beta2,rsq20);
1125             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1126             pmecorrF         = avx256_pmecorrF_f(zeta2);
1127             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1128             felec            = _mm256_mul_ps(qq20,felec);
1129             pmecorrV         = avx256_pmecorrV_f(zeta2);
1130             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1131             velec            = _mm256_sub_ps(rinv20,pmecorrV);
1132             velec            = _mm256_mul_ps(qq20,velec);
1133
1134             d                = _mm256_sub_ps(r20,rswitch);
1135             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1136             d2               = _mm256_mul_ps(d,d);
1137             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
1138
1139             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1140
1141             /* Evaluate switch function */
1142             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1143             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv20,_mm256_mul_ps(velec,dsw)) );
1144             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1145
1146             fscal            = felec;
1147
1148             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1149
1150             /* Calculate temporary vectorial force */
1151             tx               = _mm256_mul_ps(fscal,dx20);
1152             ty               = _mm256_mul_ps(fscal,dy20);
1153             tz               = _mm256_mul_ps(fscal,dz20);
1154
1155             /* Update vectorial force */
1156             fix2             = _mm256_add_ps(fix2,tx);
1157             fiy2             = _mm256_add_ps(fiy2,ty);
1158             fiz2             = _mm256_add_ps(fiz2,tz);
1159
1160             fjx0             = _mm256_add_ps(fjx0,tx);
1161             fjy0             = _mm256_add_ps(fjy0,ty);
1162             fjz0             = _mm256_add_ps(fjz0,tz);
1163
1164             }
1165
1166             fjptrA             = f+j_coord_offsetA;
1167             fjptrB             = f+j_coord_offsetB;
1168             fjptrC             = f+j_coord_offsetC;
1169             fjptrD             = f+j_coord_offsetD;
1170             fjptrE             = f+j_coord_offsetE;
1171             fjptrF             = f+j_coord_offsetF;
1172             fjptrG             = f+j_coord_offsetG;
1173             fjptrH             = f+j_coord_offsetH;
1174
1175             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1176
1177             /* Inner loop uses 318 flops */
1178         }
1179
1180         if(jidx<j_index_end)
1181         {
1182
1183             /* Get j neighbor index, and coordinate index */
1184             jnrlistA         = jjnr[jidx];
1185             jnrlistB         = jjnr[jidx+1];
1186             jnrlistC         = jjnr[jidx+2];
1187             jnrlistD         = jjnr[jidx+3];
1188             jnrlistE         = jjnr[jidx+4];
1189             jnrlistF         = jjnr[jidx+5];
1190             jnrlistG         = jjnr[jidx+6];
1191             jnrlistH         = jjnr[jidx+7];
1192             /* Sign of each element will be negative for non-real atoms.
1193              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1194              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1195              */
1196             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1197                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1198
1199             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1200             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1201             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1202             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1203             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1204             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1205             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1206             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1207             j_coord_offsetA  = DIM*jnrA;
1208             j_coord_offsetB  = DIM*jnrB;
1209             j_coord_offsetC  = DIM*jnrC;
1210             j_coord_offsetD  = DIM*jnrD;
1211             j_coord_offsetE  = DIM*jnrE;
1212             j_coord_offsetF  = DIM*jnrF;
1213             j_coord_offsetG  = DIM*jnrG;
1214             j_coord_offsetH  = DIM*jnrH;
1215
1216             /* load j atom coordinates */
1217             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1218                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1219                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1220                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1221                                                  &jx0,&jy0,&jz0);
1222
1223             /* Calculate displacement vector */
1224             dx00             = _mm256_sub_ps(ix0,jx0);
1225             dy00             = _mm256_sub_ps(iy0,jy0);
1226             dz00             = _mm256_sub_ps(iz0,jz0);
1227             dx10             = _mm256_sub_ps(ix1,jx0);
1228             dy10             = _mm256_sub_ps(iy1,jy0);
1229             dz10             = _mm256_sub_ps(iz1,jz0);
1230             dx20             = _mm256_sub_ps(ix2,jx0);
1231             dy20             = _mm256_sub_ps(iy2,jy0);
1232             dz20             = _mm256_sub_ps(iz2,jz0);
1233
1234             /* Calculate squared distance and things based on it */
1235             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1236             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1237             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1238
1239             rinv00           = avx256_invsqrt_f(rsq00);
1240             rinv10           = avx256_invsqrt_f(rsq10);
1241             rinv20           = avx256_invsqrt_f(rsq20);
1242
1243             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1244             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1245             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1246
1247             /* Load parameters for j particles */
1248             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1249                                                                  charge+jnrC+0,charge+jnrD+0,
1250                                                                  charge+jnrE+0,charge+jnrF+0,
1251                                                                  charge+jnrG+0,charge+jnrH+0);
1252
1253             fjx0             = _mm256_setzero_ps();
1254             fjy0             = _mm256_setzero_ps();
1255             fjz0             = _mm256_setzero_ps();
1256
1257             /**************************
1258              * CALCULATE INTERACTIONS *
1259              **************************/
1260
1261             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1262             {
1263
1264             r00              = _mm256_mul_ps(rsq00,rinv00);
1265             r00              = _mm256_andnot_ps(dummy_mask,r00);
1266
1267             /* Compute parameters for interactions between i and j atoms */
1268             qq00             = _mm256_mul_ps(iq0,jq0);
1269
1270             /* EWALD ELECTROSTATICS */
1271
1272             /* Analytical PME correction */
1273             zeta2            = _mm256_mul_ps(beta2,rsq00);
1274             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1275             pmecorrF         = avx256_pmecorrF_f(zeta2);
1276             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1277             felec            = _mm256_mul_ps(qq00,felec);
1278             pmecorrV         = avx256_pmecorrV_f(zeta2);
1279             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1280             velec            = _mm256_sub_ps(rinv00,pmecorrV);
1281             velec            = _mm256_mul_ps(qq00,velec);
1282
1283             d                = _mm256_sub_ps(r00,rswitch);
1284             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1285             d2               = _mm256_mul_ps(d,d);
1286             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
1287
1288             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1289
1290             /* Evaluate switch function */
1291             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1292             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
1293             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1294
1295             fscal            = felec;
1296
1297             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1298
1299             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1300
1301             /* Calculate temporary vectorial force */
1302             tx               = _mm256_mul_ps(fscal,dx00);
1303             ty               = _mm256_mul_ps(fscal,dy00);
1304             tz               = _mm256_mul_ps(fscal,dz00);
1305
1306             /* Update vectorial force */
1307             fix0             = _mm256_add_ps(fix0,tx);
1308             fiy0             = _mm256_add_ps(fiy0,ty);
1309             fiz0             = _mm256_add_ps(fiz0,tz);
1310
1311             fjx0             = _mm256_add_ps(fjx0,tx);
1312             fjy0             = _mm256_add_ps(fjy0,ty);
1313             fjz0             = _mm256_add_ps(fjz0,tz);
1314
1315             }
1316
1317             /**************************
1318              * CALCULATE INTERACTIONS *
1319              **************************/
1320
1321             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1322             {
1323
1324             r10              = _mm256_mul_ps(rsq10,rinv10);
1325             r10              = _mm256_andnot_ps(dummy_mask,r10);
1326
1327             /* Compute parameters for interactions between i and j atoms */
1328             qq10             = _mm256_mul_ps(iq1,jq0);
1329
1330             /* EWALD ELECTROSTATICS */
1331
1332             /* Analytical PME correction */
1333             zeta2            = _mm256_mul_ps(beta2,rsq10);
1334             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1335             pmecorrF         = avx256_pmecorrF_f(zeta2);
1336             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1337             felec            = _mm256_mul_ps(qq10,felec);
1338             pmecorrV         = avx256_pmecorrV_f(zeta2);
1339             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1340             velec            = _mm256_sub_ps(rinv10,pmecorrV);
1341             velec            = _mm256_mul_ps(qq10,velec);
1342
1343             d                = _mm256_sub_ps(r10,rswitch);
1344             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1345             d2               = _mm256_mul_ps(d,d);
1346             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
1347
1348             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1349
1350             /* Evaluate switch function */
1351             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1352             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv10,_mm256_mul_ps(velec,dsw)) );
1353             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1354
1355             fscal            = felec;
1356
1357             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1358
1359             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1360
1361             /* Calculate temporary vectorial force */
1362             tx               = _mm256_mul_ps(fscal,dx10);
1363             ty               = _mm256_mul_ps(fscal,dy10);
1364             tz               = _mm256_mul_ps(fscal,dz10);
1365
1366             /* Update vectorial force */
1367             fix1             = _mm256_add_ps(fix1,tx);
1368             fiy1             = _mm256_add_ps(fiy1,ty);
1369             fiz1             = _mm256_add_ps(fiz1,tz);
1370
1371             fjx0             = _mm256_add_ps(fjx0,tx);
1372             fjy0             = _mm256_add_ps(fjy0,ty);
1373             fjz0             = _mm256_add_ps(fjz0,tz);
1374
1375             }
1376
1377             /**************************
1378              * CALCULATE INTERACTIONS *
1379              **************************/
1380
1381             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1382             {
1383
1384             r20              = _mm256_mul_ps(rsq20,rinv20);
1385             r20              = _mm256_andnot_ps(dummy_mask,r20);
1386
1387             /* Compute parameters for interactions between i and j atoms */
1388             qq20             = _mm256_mul_ps(iq2,jq0);
1389
1390             /* EWALD ELECTROSTATICS */
1391
1392             /* Analytical PME correction */
1393             zeta2            = _mm256_mul_ps(beta2,rsq20);
1394             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1395             pmecorrF         = avx256_pmecorrF_f(zeta2);
1396             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1397             felec            = _mm256_mul_ps(qq20,felec);
1398             pmecorrV         = avx256_pmecorrV_f(zeta2);
1399             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1400             velec            = _mm256_sub_ps(rinv20,pmecorrV);
1401             velec            = _mm256_mul_ps(qq20,velec);
1402
1403             d                = _mm256_sub_ps(r20,rswitch);
1404             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1405             d2               = _mm256_mul_ps(d,d);
1406             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
1407
1408             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1409
1410             /* Evaluate switch function */
1411             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1412             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv20,_mm256_mul_ps(velec,dsw)) );
1413             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1414
1415             fscal            = felec;
1416
1417             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1418
1419             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1420
1421             /* Calculate temporary vectorial force */
1422             tx               = _mm256_mul_ps(fscal,dx20);
1423             ty               = _mm256_mul_ps(fscal,dy20);
1424             tz               = _mm256_mul_ps(fscal,dz20);
1425
1426             /* Update vectorial force */
1427             fix2             = _mm256_add_ps(fix2,tx);
1428             fiy2             = _mm256_add_ps(fiy2,ty);
1429             fiz2             = _mm256_add_ps(fiz2,tz);
1430
1431             fjx0             = _mm256_add_ps(fjx0,tx);
1432             fjy0             = _mm256_add_ps(fjy0,ty);
1433             fjz0             = _mm256_add_ps(fjz0,tz);
1434
1435             }
1436
1437             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1438             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1439             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1440             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1441             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1442             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1443             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1444             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1445
1446             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1447
1448             /* Inner loop uses 321 flops */
1449         }
1450
1451         /* End of innermost loop */
1452
1453         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1454                                                  f+i_coord_offset,fshift+i_shift_offset);
1455
1456         /* Increment number of inner iterations */
1457         inneriter                  += j_index_end - j_index_start;
1458
1459         /* Outer loop uses 18 flops */
1460     }
1461
1462     /* Increment number of outer iterations */
1463     outeriter        += nri;
1464
1465     /* Update outer/inner flops */
1466
1467     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*321);
1468 }