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, 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/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomW3P1_VF_avx_256_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            None
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSw_VdwNone_GeomW3P1_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     real *           vdwioffsetptr1;
  91     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  92     real *           vdwioffsetptr2;
  93     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  95     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  97     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  99     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     __m256i          ewitab;
 102     __m128i          ewitab_lo,ewitab_hi;
 103     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 104     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 105     real             *ewtab;
 106     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 107     real             rswitch_scalar,d_scalar;
 108     __m256           dummy_mask,cutoff_mask;
 109     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 110     __m256           one     = _mm256_set1_ps(1.0);
 111     __m256           two     = _mm256_set1_ps(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm256_set1_ps(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125
 126     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 127     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 128     beta2            = _mm256_mul_ps(beta,beta);
 129     beta3            = _mm256_mul_ps(beta,beta2);
 130
 131     ewtab            = fr->ic->tabq_coul_FDV0;
 132     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 133     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 134
 135     /* Setup water-specific parameters */
 136     inr              = nlist->iinr[0];
 137     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 138     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 139     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 140
 141     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 142     rcutoff_scalar   = fr->rcoulomb;
 143     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 144     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 145
 146     rswitch_scalar   = fr->rcoulomb_switch;
 147     rswitch          = _mm256_set1_ps(rswitch_scalar);
 148     /* Setup switch parameters */
 149     d_scalar         = rcutoff_scalar-rswitch_scalar;
 150     d                = _mm256_set1_ps(d_scalar);
 151     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 152     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 153     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 154     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 155     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 156     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 157
 158     /* Avoid stupid compiler warnings */
 159     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 160     j_coord_offsetA = 0;
 161     j_coord_offsetB = 0;
 162     j_coord_offsetC = 0;
 163     j_coord_offsetD = 0;
 164     j_coord_offsetE = 0;
 165     j_coord_offsetF = 0;
 166     j_coord_offsetG = 0;
 167     j_coord_offsetH = 0;
 168
 169     outeriter        = 0;
 170     inneriter        = 0;
 171
 172     for(iidx=0;iidx<4*DIM;iidx++)
 173     {
 174         scratch[iidx] = 0.0;
 175     }
 176
 177     /* Start outer loop over neighborlists */
 178     for(iidx=0; iidx<nri; iidx++)
 179     {
 180         /* Load shift vector for this list */
 181         i_shift_offset   = DIM*shiftidx[iidx];
 182
 183         /* Load limits for loop over neighbors */
 184         j_index_start    = jindex[iidx];
 185         j_index_end      = jindex[iidx+1];
 186
 187         /* Get outer coordinate index */
 188         inr              = iinr[iidx];
 189         i_coord_offset   = DIM*inr;
 190
 191         /* Load i particle coords and add shift vector */
 192         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 193                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 194
 195         fix0             = _mm256_setzero_ps();
 196         fiy0             = _mm256_setzero_ps();
 197         fiz0             = _mm256_setzero_ps();
 198         fix1             = _mm256_setzero_ps();
 199         fiy1             = _mm256_setzero_ps();
 200         fiz1             = _mm256_setzero_ps();
 201         fix2             = _mm256_setzero_ps();
 202         fiy2             = _mm256_setzero_ps();
 203         fiz2             = _mm256_setzero_ps();
 204
 205         /* Reset potential sums */
 206         velecsum         = _mm256_setzero_ps();
 207
 208         /* Start inner kernel loop */
 209         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 210         {
 211
 212             /* Get j neighbor index, and coordinate index */
 213             jnrA             = jjnr[jidx];
 214             jnrB             = jjnr[jidx+1];
 215             jnrC             = jjnr[jidx+2];
 216             jnrD             = jjnr[jidx+3];
 217             jnrE             = jjnr[jidx+4];
 218             jnrF             = jjnr[jidx+5];
 219             jnrG             = jjnr[jidx+6];
 220             jnrH             = jjnr[jidx+7];
 221             j_coord_offsetA  = DIM*jnrA;
 222             j_coord_offsetB  = DIM*jnrB;
 223             j_coord_offsetC  = DIM*jnrC;
 224             j_coord_offsetD  = DIM*jnrD;
 225             j_coord_offsetE  = DIM*jnrE;
 226             j_coord_offsetF  = DIM*jnrF;
 227             j_coord_offsetG  = DIM*jnrG;
 228             j_coord_offsetH  = DIM*jnrH;
 229
 230             /* load j atom coordinates */
 231             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 232                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 233                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 234                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 235                                                  &jx0,&jy0,&jz0);
 236
 237             /* Calculate displacement vector */
 238             dx00             = _mm256_sub_ps(ix0,jx0);
 239             dy00             = _mm256_sub_ps(iy0,jy0);
 240             dz00             = _mm256_sub_ps(iz0,jz0);
 241             dx10             = _mm256_sub_ps(ix1,jx0);
 242             dy10             = _mm256_sub_ps(iy1,jy0);
 243             dz10             = _mm256_sub_ps(iz1,jz0);
 244             dx20             = _mm256_sub_ps(ix2,jx0);
 245             dy20             = _mm256_sub_ps(iy2,jy0);
 246             dz20             = _mm256_sub_ps(iz2,jz0);
 247
 248             /* Calculate squared distance and things based on it */
 249             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 250             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 251             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 252
 253             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 254             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 255             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 256
 257             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 258             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 259             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 260
 261             /* Load parameters for j particles */
 262             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 263                                                                  charge+jnrC+0,charge+jnrD+0,
 264                                                                  charge+jnrE+0,charge+jnrF+0,
 265                                                                  charge+jnrG+0,charge+jnrH+0);
 266
 267             fjx0             = _mm256_setzero_ps();
 268             fjy0             = _mm256_setzero_ps();
 269             fjz0             = _mm256_setzero_ps();
 270
 271             /**************************
 272              * CALCULATE INTERACTIONS *
 273              **************************/
 274
 275             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 276             {
 277
 278             r00              = _mm256_mul_ps(rsq00,rinv00);
 279
 280             /* Compute parameters for interactions between i and j atoms */
 281             qq00             = _mm256_mul_ps(iq0,jq0);
 282
 283             /* EWALD ELECTROSTATICS */
 284
 285             /* Analytical PME correction */
 286             zeta2            = _mm256_mul_ps(beta2,rsq00);
 287             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 288             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 289             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 290             felec            = _mm256_mul_ps(qq00,felec);
 291             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 292             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 293             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 294             velec            = _mm256_mul_ps(qq00,velec);
 295
 296             d                = _mm256_sub_ps(r00,rswitch);
 297             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 298             d2               = _mm256_mul_ps(d,d);
 299             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)))))));
 300
 301             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 302
 303             /* Evaluate switch function */
 304             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 305             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
 306             velec            = _mm256_mul_ps(velec,sw);
 307             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 308
 309             /* Update potential sum for this i atom from the interaction with this j atom. */
 310             velec            = _mm256_and_ps(velec,cutoff_mask);
 311             velecsum         = _mm256_add_ps(velecsum,velec);
 312
 313             fscal            = felec;
 314
 315             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 316
 317             /* Calculate temporary vectorial force */
 318             tx               = _mm256_mul_ps(fscal,dx00);
 319             ty               = _mm256_mul_ps(fscal,dy00);
 320             tz               = _mm256_mul_ps(fscal,dz00);
 321
 322             /* Update vectorial force */
 323             fix0             = _mm256_add_ps(fix0,tx);
 324             fiy0             = _mm256_add_ps(fiy0,ty);
 325             fiz0             = _mm256_add_ps(fiz0,tz);
 326
 327             fjx0             = _mm256_add_ps(fjx0,tx);
 328             fjy0             = _mm256_add_ps(fjy0,ty);
 329             fjz0             = _mm256_add_ps(fjz0,tz);
 330
 331             }
 332
 333             /**************************
 334              * CALCULATE INTERACTIONS *
 335              **************************/
 336
 337             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 338             {
 339
 340             r10              = _mm256_mul_ps(rsq10,rinv10);
 341
 342             /* Compute parameters for interactions between i and j atoms */
 343             qq10             = _mm256_mul_ps(iq1,jq0);
 344
 345             /* EWALD ELECTROSTATICS */
 346
 347             /* Analytical PME correction */
 348             zeta2            = _mm256_mul_ps(beta2,rsq10);
 349             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 350             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 351             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 352             felec            = _mm256_mul_ps(qq10,felec);
 353             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 354             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 355             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 356             velec            = _mm256_mul_ps(qq10,velec);
 357
 358             d                = _mm256_sub_ps(r10,rswitch);
 359             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 360             d2               = _mm256_mul_ps(d,d);
 361             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)))))));
 362
 363             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 364
 365             /* Evaluate switch function */
 366             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 367             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv10,_mm256_mul_ps(velec,dsw)) );
 368             velec            = _mm256_mul_ps(velec,sw);
 369             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 370
 371             /* Update potential sum for this i atom from the interaction with this j atom. */
 372             velec            = _mm256_and_ps(velec,cutoff_mask);
 373             velecsum         = _mm256_add_ps(velecsum,velec);
 374
 375             fscal            = felec;
 376
 377             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 378
 379             /* Calculate temporary vectorial force */
 380             tx               = _mm256_mul_ps(fscal,dx10);
 381             ty               = _mm256_mul_ps(fscal,dy10);
 382             tz               = _mm256_mul_ps(fscal,dz10);
 383
 384             /* Update vectorial force */
 385             fix1             = _mm256_add_ps(fix1,tx);
 386             fiy1             = _mm256_add_ps(fiy1,ty);
 387             fiz1             = _mm256_add_ps(fiz1,tz);
 388
 389             fjx0             = _mm256_add_ps(fjx0,tx);
 390             fjy0             = _mm256_add_ps(fjy0,ty);
 391             fjz0             = _mm256_add_ps(fjz0,tz);
 392
 393             }
 394
 395             /**************************
 396              * CALCULATE INTERACTIONS *
 397              **************************/
 398
 399             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 400             {
 401
 402             r20              = _mm256_mul_ps(rsq20,rinv20);
 403
 404             /* Compute parameters for interactions between i and j atoms */
 405             qq20             = _mm256_mul_ps(iq2,jq0);
 406
 407             /* EWALD ELECTROSTATICS */
 408
 409             /* Analytical PME correction */
 410             zeta2            = _mm256_mul_ps(beta2,rsq20);
 411             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 412             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 413             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 414             felec            = _mm256_mul_ps(qq20,felec);
 415             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 416             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 417             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 418             velec            = _mm256_mul_ps(qq20,velec);
 419
 420             d                = _mm256_sub_ps(r20,rswitch);
 421             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 422             d2               = _mm256_mul_ps(d,d);
 423             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)))))));
 424
 425             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 426
 427             /* Evaluate switch function */
 428             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 429             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv20,_mm256_mul_ps(velec,dsw)) );
 430             velec            = _mm256_mul_ps(velec,sw);
 431             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 432
 433             /* Update potential sum for this i atom from the interaction with this j atom. */
 434             velec            = _mm256_and_ps(velec,cutoff_mask);
 435             velecsum         = _mm256_add_ps(velecsum,velec);
 436
 437             fscal            = felec;
 438
 439             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 440
 441             /* Calculate temporary vectorial force */
 442             tx               = _mm256_mul_ps(fscal,dx20);
 443             ty               = _mm256_mul_ps(fscal,dy20);
 444             tz               = _mm256_mul_ps(fscal,dz20);
 445
 446             /* Update vectorial force */
 447             fix2             = _mm256_add_ps(fix2,tx);
 448             fiy2             = _mm256_add_ps(fiy2,ty);
 449             fiz2             = _mm256_add_ps(fiz2,tz);
 450
 451             fjx0             = _mm256_add_ps(fjx0,tx);
 452             fjy0             = _mm256_add_ps(fjy0,ty);
 453             fjz0             = _mm256_add_ps(fjz0,tz);
 454
 455             }
 456
 457             fjptrA             = f+j_coord_offsetA;
 458             fjptrB             = f+j_coord_offsetB;
 459             fjptrC             = f+j_coord_offsetC;
 460             fjptrD             = f+j_coord_offsetD;
 461             fjptrE             = f+j_coord_offsetE;
 462             fjptrF             = f+j_coord_offsetF;
 463             fjptrG             = f+j_coord_offsetG;
 464             fjptrH             = f+j_coord_offsetH;
 465
 466             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 467
 468             /* Inner loop uses 327 flops */
 469         }
 470
 471         if(jidx<j_index_end)
 472         {
 473
 474             /* Get j neighbor index, and coordinate index */
 475             jnrlistA         = jjnr[jidx];
 476             jnrlistB         = jjnr[jidx+1];
 477             jnrlistC         = jjnr[jidx+2];
 478             jnrlistD         = jjnr[jidx+3];
 479             jnrlistE         = jjnr[jidx+4];
 480             jnrlistF         = jjnr[jidx+5];
 481             jnrlistG         = jjnr[jidx+6];
 482             jnrlistH         = jjnr[jidx+7];
 483             /* Sign of each element will be negative for non-real atoms.
 484              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 485              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 486              */
 487             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 488                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 489
 490             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 491             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 492             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 493             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 494             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 495             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 496             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 497             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 498             j_coord_offsetA  = DIM*jnrA;
 499             j_coord_offsetB  = DIM*jnrB;
 500             j_coord_offsetC  = DIM*jnrC;
 501             j_coord_offsetD  = DIM*jnrD;
 502             j_coord_offsetE  = DIM*jnrE;
 503             j_coord_offsetF  = DIM*jnrF;
 504             j_coord_offsetG  = DIM*jnrG;
 505             j_coord_offsetH  = DIM*jnrH;
 506
 507             /* load j atom coordinates */
 508             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 509                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 510                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 511                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 512                                                  &jx0,&jy0,&jz0);
 513
 514             /* Calculate displacement vector */
 515             dx00             = _mm256_sub_ps(ix0,jx0);
 516             dy00             = _mm256_sub_ps(iy0,jy0);
 517             dz00             = _mm256_sub_ps(iz0,jz0);
 518             dx10             = _mm256_sub_ps(ix1,jx0);
 519             dy10             = _mm256_sub_ps(iy1,jy0);
 520             dz10             = _mm256_sub_ps(iz1,jz0);
 521             dx20             = _mm256_sub_ps(ix2,jx0);
 522             dy20             = _mm256_sub_ps(iy2,jy0);
 523             dz20             = _mm256_sub_ps(iz2,jz0);
 524
 525             /* Calculate squared distance and things based on it */
 526             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 527             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 528             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 529
 530             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 531             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 532             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 533
 534             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 535             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 536             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 537
 538             /* Load parameters for j particles */
 539             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 540                                                                  charge+jnrC+0,charge+jnrD+0,
 541                                                                  charge+jnrE+0,charge+jnrF+0,
 542                                                                  charge+jnrG+0,charge+jnrH+0);
 543
 544             fjx0             = _mm256_setzero_ps();
 545             fjy0             = _mm256_setzero_ps();
 546             fjz0             = _mm256_setzero_ps();
 547
 548             /**************************
 549              * CALCULATE INTERACTIONS *
 550              **************************/
 551
 552             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 553             {
 554
 555             r00              = _mm256_mul_ps(rsq00,rinv00);
 556             r00              = _mm256_andnot_ps(dummy_mask,r00);
 557
 558             /* Compute parameters for interactions between i and j atoms */
 559             qq00             = _mm256_mul_ps(iq0,jq0);
 560
 561             /* EWALD ELECTROSTATICS */
 562
 563             /* Analytical PME correction */
 564             zeta2            = _mm256_mul_ps(beta2,rsq00);
 565             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 566             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 567             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 568             felec            = _mm256_mul_ps(qq00,felec);
 569             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 570             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 571             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 572             velec            = _mm256_mul_ps(qq00,velec);
 573
 574             d                = _mm256_sub_ps(r00,rswitch);
 575             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 576             d2               = _mm256_mul_ps(d,d);
 577             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)))))));
 578
 579             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 580
 581             /* Evaluate switch function */
 582             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 583             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
 584             velec            = _mm256_mul_ps(velec,sw);
 585             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 586
 587             /* Update potential sum for this i atom from the interaction with this j atom. */
 588             velec            = _mm256_and_ps(velec,cutoff_mask);
 589             velec            = _mm256_andnot_ps(dummy_mask,velec);
 590             velecsum         = _mm256_add_ps(velecsum,velec);
 591
 592             fscal            = felec;
 593
 594             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 595
 596             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 597
 598             /* Calculate temporary vectorial force */
 599             tx               = _mm256_mul_ps(fscal,dx00);
 600             ty               = _mm256_mul_ps(fscal,dy00);
 601             tz               = _mm256_mul_ps(fscal,dz00);
 602
 603             /* Update vectorial force */
 604             fix0             = _mm256_add_ps(fix0,tx);
 605             fiy0             = _mm256_add_ps(fiy0,ty);
 606             fiz0             = _mm256_add_ps(fiz0,tz);
 607
 608             fjx0             = _mm256_add_ps(fjx0,tx);
 609             fjy0             = _mm256_add_ps(fjy0,ty);
 610             fjz0             = _mm256_add_ps(fjz0,tz);
 611
 612             }
 613
 614             /**************************
 615              * CALCULATE INTERACTIONS *
 616              **************************/
 617
 618             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 619             {
 620
 621             r10              = _mm256_mul_ps(rsq10,rinv10);
 622             r10              = _mm256_andnot_ps(dummy_mask,r10);
 623
 624             /* Compute parameters for interactions between i and j atoms */
 625             qq10             = _mm256_mul_ps(iq1,jq0);
 626
 627             /* EWALD ELECTROSTATICS */
 628
 629             /* Analytical PME correction */
 630             zeta2            = _mm256_mul_ps(beta2,rsq10);
 631             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 632             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 633             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 634             felec            = _mm256_mul_ps(qq10,felec);
 635             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 636             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 637             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 638             velec            = _mm256_mul_ps(qq10,velec);
 639
 640             d                = _mm256_sub_ps(r10,rswitch);
 641             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 642             d2               = _mm256_mul_ps(d,d);
 643             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)))))));
 644
 645             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 646
 647             /* Evaluate switch function */
 648             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 649             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv10,_mm256_mul_ps(velec,dsw)) );
 650             velec            = _mm256_mul_ps(velec,sw);
 651             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 652
 653             /* Update potential sum for this i atom from the interaction with this j atom. */
 654             velec            = _mm256_and_ps(velec,cutoff_mask);
 655             velec            = _mm256_andnot_ps(dummy_mask,velec);
 656             velecsum         = _mm256_add_ps(velecsum,velec);
 657
 658             fscal            = felec;
 659
 660             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 661
 662             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 663
 664             /* Calculate temporary vectorial force */
 665             tx               = _mm256_mul_ps(fscal,dx10);
 666             ty               = _mm256_mul_ps(fscal,dy10);
 667             tz               = _mm256_mul_ps(fscal,dz10);
 668
 669             /* Update vectorial force */
 670             fix1             = _mm256_add_ps(fix1,tx);
 671             fiy1             = _mm256_add_ps(fiy1,ty);
 672             fiz1             = _mm256_add_ps(fiz1,tz);
 673
 674             fjx0             = _mm256_add_ps(fjx0,tx);
 675             fjy0             = _mm256_add_ps(fjy0,ty);
 676             fjz0             = _mm256_add_ps(fjz0,tz);
 677
 678             }
 679
 680             /**************************
 681              * CALCULATE INTERACTIONS *
 682              **************************/
 683
 684             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 685             {
 686
 687             r20              = _mm256_mul_ps(rsq20,rinv20);
 688             r20              = _mm256_andnot_ps(dummy_mask,r20);
 689
 690             /* Compute parameters for interactions between i and j atoms */
 691             qq20             = _mm256_mul_ps(iq2,jq0);
 692
 693             /* EWALD ELECTROSTATICS */
 694
 695             /* Analytical PME correction */
 696             zeta2            = _mm256_mul_ps(beta2,rsq20);
 697             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 698             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 699             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 700             felec            = _mm256_mul_ps(qq20,felec);
 701             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 702             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 703             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 704             velec            = _mm256_mul_ps(qq20,velec);
 705
 706             d                = _mm256_sub_ps(r20,rswitch);
 707             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 708             d2               = _mm256_mul_ps(d,d);
 709             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)))))));
 710
 711             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 712
 713             /* Evaluate switch function */
 714             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 715             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv20,_mm256_mul_ps(velec,dsw)) );
 716             velec            = _mm256_mul_ps(velec,sw);
 717             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 718
 719             /* Update potential sum for this i atom from the interaction with this j atom. */
 720             velec            = _mm256_and_ps(velec,cutoff_mask);
 721             velec            = _mm256_andnot_ps(dummy_mask,velec);
 722             velecsum         = _mm256_add_ps(velecsum,velec);
 723
 724             fscal            = felec;
 725
 726             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 727
 728             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 729
 730             /* Calculate temporary vectorial force */
 731             tx               = _mm256_mul_ps(fscal,dx20);
 732             ty               = _mm256_mul_ps(fscal,dy20);
 733             tz               = _mm256_mul_ps(fscal,dz20);
 734
 735             /* Update vectorial force */
 736             fix2             = _mm256_add_ps(fix2,tx);
 737             fiy2             = _mm256_add_ps(fiy2,ty);
 738             fiz2             = _mm256_add_ps(fiz2,tz);
 739
 740             fjx0             = _mm256_add_ps(fjx0,tx);
 741             fjy0             = _mm256_add_ps(fjy0,ty);
 742             fjz0             = _mm256_add_ps(fjz0,tz);
 743
 744             }
 745
 746             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 747             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 748             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 749             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 750             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 751             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 752             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 753             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 754
 755             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 756
 757             /* Inner loop uses 330 flops */
 758         }
 759
 760         /* End of innermost loop */
 761
 762         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 763                                                  f+i_coord_offset,fshift+i_shift_offset);
 764
 765         ggid                        = gid[iidx];
 766         /* Update potential energies */
 767         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 768
 769         /* Increment number of inner iterations */
 770         inneriter                  += j_index_end - j_index_start;
 771
 772         /* Outer loop uses 19 flops */
 773     }
 774
 775     /* Increment number of outer iterations */
 776     outeriter        += nri;
 777
 778     /* Update outer/inner flops */
 779
 780     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*330);
 781 }
 782 /*
 783  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomW3P1_F_avx_256_single
 784  * Electrostatics interaction: Ewald
 785  * VdW interaction:            None
 786  * Geometry:                   Water3-Particle
 787  * Calculate force/pot:        Force
 788  */
 789 void
 790 nb_kernel_ElecEwSw_VdwNone_GeomW3P1_F_avx_256_single
 791                     (t_nblist                    * gmx_restrict       nlist,
 792                      rvec                        * gmx_restrict          xx,
 793                      rvec                        * gmx_restrict          ff,
 794                      t_forcerec                  * gmx_restrict          fr,
 795                      t_mdatoms                   * gmx_restrict     mdatoms,
 796                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 797                      t_nrnb                      * gmx_restrict        nrnb)
 798 {
 799     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 800      * just 0 for non-waters.
 801      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 802      * jnr indices corresponding to data put in the four positions in the SIMD register.
 803      */
 804     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 805     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 806     int              jnrA,jnrB,jnrC,jnrD;
 807     int              jnrE,jnrF,jnrG,jnrH;
 808     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 809     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 810     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 811     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 812     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 813     real             rcutoff_scalar;
 814     real             *shiftvec,*fshift,*x,*f;
 815     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 816     real             scratch[4*DIM];
 817     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 818     real *           vdwioffsetptr0;
 819     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 820     real *           vdwioffsetptr1;
 821     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 822     real *           vdwioffsetptr2;
 823     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 824     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 825     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 826     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 827     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 828     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 829     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 830     real             *charge;
 831     __m256i          ewitab;
 832     __m128i          ewitab_lo,ewitab_hi;
 833     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 834     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 835     real             *ewtab;
 836     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 837     real             rswitch_scalar,d_scalar;
 838     __m256           dummy_mask,cutoff_mask;
 839     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 840     __m256           one     = _mm256_set1_ps(1.0);
 841     __m256           two     = _mm256_set1_ps(2.0);
 842     x                = xx[0];
 843     f                = ff[0];
 844
 845     nri              = nlist->nri;
 846     iinr             = nlist->iinr;
 847     jindex           = nlist->jindex;
 848     jjnr             = nlist->jjnr;
 849     shiftidx         = nlist->shift;
 850     gid              = nlist->gid;
 851     shiftvec         = fr->shift_vec[0];
 852     fshift           = fr->fshift[0];
 853     facel            = _mm256_set1_ps(fr->epsfac);
 854     charge           = mdatoms->chargeA;
 855
 856     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 857     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 858     beta2            = _mm256_mul_ps(beta,beta);
 859     beta3            = _mm256_mul_ps(beta,beta2);
 860
 861     ewtab            = fr->ic->tabq_coul_FDV0;
 862     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 863     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 864
 865     /* Setup water-specific parameters */
 866     inr              = nlist->iinr[0];
 867     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 868     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 869     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 870
 871     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 872     rcutoff_scalar   = fr->rcoulomb;
 873     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 874     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 875
 876     rswitch_scalar   = fr->rcoulomb_switch;
 877     rswitch          = _mm256_set1_ps(rswitch_scalar);
 878     /* Setup switch parameters */
 879     d_scalar         = rcutoff_scalar-rswitch_scalar;
 880     d                = _mm256_set1_ps(d_scalar);
 881     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 882     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 883     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 884     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 885     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 886     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 887
 888     /* Avoid stupid compiler warnings */
 889     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 890     j_coord_offsetA = 0;
 891     j_coord_offsetB = 0;
 892     j_coord_offsetC = 0;
 893     j_coord_offsetD = 0;
 894     j_coord_offsetE = 0;
 895     j_coord_offsetF = 0;
 896     j_coord_offsetG = 0;
 897     j_coord_offsetH = 0;
 898
 899     outeriter        = 0;
 900     inneriter        = 0;
 901
 902     for(iidx=0;iidx<4*DIM;iidx++)
 903     {
 904         scratch[iidx] = 0.0;
 905     }
 906
 907     /* Start outer loop over neighborlists */
 908     for(iidx=0; iidx<nri; iidx++)
 909     {
 910         /* Load shift vector for this list */
 911         i_shift_offset   = DIM*shiftidx[iidx];
 912
 913         /* Load limits for loop over neighbors */
 914         j_index_start    = jindex[iidx];
 915         j_index_end      = jindex[iidx+1];
 916
 917         /* Get outer coordinate index */
 918         inr              = iinr[iidx];
 919         i_coord_offset   = DIM*inr;
 920
 921         /* Load i particle coords and add shift vector */
 922         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 923                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 924
 925         fix0             = _mm256_setzero_ps();
 926         fiy0             = _mm256_setzero_ps();
 927         fiz0             = _mm256_setzero_ps();
 928         fix1             = _mm256_setzero_ps();
 929         fiy1             = _mm256_setzero_ps();
 930         fiz1             = _mm256_setzero_ps();
 931         fix2             = _mm256_setzero_ps();
 932         fiy2             = _mm256_setzero_ps();
 933         fiz2             = _mm256_setzero_ps();
 934
 935         /* Start inner kernel loop */
 936         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 937         {
 938
 939             /* Get j neighbor index, and coordinate index */
 940             jnrA             = jjnr[jidx];
 941             jnrB             = jjnr[jidx+1];
 942             jnrC             = jjnr[jidx+2];
 943             jnrD             = jjnr[jidx+3];
 944             jnrE             = jjnr[jidx+4];
 945             jnrF             = jjnr[jidx+5];
 946             jnrG             = jjnr[jidx+6];
 947             jnrH             = jjnr[jidx+7];
 948             j_coord_offsetA  = DIM*jnrA;
 949             j_coord_offsetB  = DIM*jnrB;
 950             j_coord_offsetC  = DIM*jnrC;
 951             j_coord_offsetD  = DIM*jnrD;
 952             j_coord_offsetE  = DIM*jnrE;
 953             j_coord_offsetF  = DIM*jnrF;
 954             j_coord_offsetG  = DIM*jnrG;
 955             j_coord_offsetH  = DIM*jnrH;
 956
 957             /* load j atom coordinates */
 958             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 959                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 960                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 961                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 962                                                  &jx0,&jy0,&jz0);
 963
 964             /* Calculate displacement vector */
 965             dx00             = _mm256_sub_ps(ix0,jx0);
 966             dy00             = _mm256_sub_ps(iy0,jy0);
 967             dz00             = _mm256_sub_ps(iz0,jz0);
 968             dx10             = _mm256_sub_ps(ix1,jx0);
 969             dy10             = _mm256_sub_ps(iy1,jy0);
 970             dz10             = _mm256_sub_ps(iz1,jz0);
 971             dx20             = _mm256_sub_ps(ix2,jx0);
 972             dy20             = _mm256_sub_ps(iy2,jy0);
 973             dz20             = _mm256_sub_ps(iz2,jz0);
 974
 975             /* Calculate squared distance and things based on it */
 976             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 977             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 978             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 979
 980             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 981             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 982             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 983
 984             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 985             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 986             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 987
 988             /* Load parameters for j particles */
 989             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 990                                                                  charge+jnrC+0,charge+jnrD+0,
 991                                                                  charge+jnrE+0,charge+jnrF+0,
 992                                                                  charge+jnrG+0,charge+jnrH+0);
 993
 994             fjx0             = _mm256_setzero_ps();
 995             fjy0             = _mm256_setzero_ps();
 996             fjz0             = _mm256_setzero_ps();
 997
 998             /**************************
 999              * CALCULATE INTERACTIONS *
1000              **************************/
1001
1002             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1003             {
1004
1005             r00              = _mm256_mul_ps(rsq00,rinv00);
1006
1007             /* Compute parameters for interactions between i and j atoms */
1008             qq00             = _mm256_mul_ps(iq0,jq0);
1009
1010             /* EWALD ELECTROSTATICS */
1011
1012             /* Analytical PME correction */
1013             zeta2            = _mm256_mul_ps(beta2,rsq00);
1014             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1015             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1016             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1017             felec            = _mm256_mul_ps(qq00,felec);
1018             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1019             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1020             velec            = _mm256_sub_ps(rinv00,pmecorrV);
1021             velec            = _mm256_mul_ps(qq00,velec);
1022
1023             d                = _mm256_sub_ps(r00,rswitch);
1024             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1025             d2               = _mm256_mul_ps(d,d);
1026             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)))))));
1027
1028             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1029
1030             /* Evaluate switch function */
1031             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1032             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
1033             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1034
1035             fscal            = felec;
1036
1037             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1038
1039             /* Calculate temporary vectorial force */
1040             tx               = _mm256_mul_ps(fscal,dx00);
1041             ty               = _mm256_mul_ps(fscal,dy00);
1042             tz               = _mm256_mul_ps(fscal,dz00);
1043
1044             /* Update vectorial force */
1045             fix0             = _mm256_add_ps(fix0,tx);
1046             fiy0             = _mm256_add_ps(fiy0,ty);
1047             fiz0             = _mm256_add_ps(fiz0,tz);
1048
1049             fjx0             = _mm256_add_ps(fjx0,tx);
1050             fjy0             = _mm256_add_ps(fjy0,ty);
1051             fjz0             = _mm256_add_ps(fjz0,tz);
1052
1053             }
1054
1055             /**************************
1056              * CALCULATE INTERACTIONS *
1057              **************************/
1058
1059             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1060             {
1061
1062             r10              = _mm256_mul_ps(rsq10,rinv10);
1063
1064             /* Compute parameters for interactions between i and j atoms */
1065             qq10             = _mm256_mul_ps(iq1,jq0);
1066
1067             /* EWALD ELECTROSTATICS */
1068
1069             /* Analytical PME correction */
1070             zeta2            = _mm256_mul_ps(beta2,rsq10);
1071             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1072             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1073             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1074             felec            = _mm256_mul_ps(qq10,felec);
1075             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1076             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1077             velec            = _mm256_sub_ps(rinv10,pmecorrV);
1078             velec            = _mm256_mul_ps(qq10,velec);
1079
1080             d                = _mm256_sub_ps(r10,rswitch);
1081             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1082             d2               = _mm256_mul_ps(d,d);
1083             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)))))));
1084
1085             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1086
1087             /* Evaluate switch function */
1088             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1089             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv10,_mm256_mul_ps(velec,dsw)) );
1090             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1091
1092             fscal            = felec;
1093
1094             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1095
1096             /* Calculate temporary vectorial force */
1097             tx               = _mm256_mul_ps(fscal,dx10);
1098             ty               = _mm256_mul_ps(fscal,dy10);
1099             tz               = _mm256_mul_ps(fscal,dz10);
1100
1101             /* Update vectorial force */
1102             fix1             = _mm256_add_ps(fix1,tx);
1103             fiy1             = _mm256_add_ps(fiy1,ty);
1104             fiz1             = _mm256_add_ps(fiz1,tz);
1105
1106             fjx0             = _mm256_add_ps(fjx0,tx);
1107             fjy0             = _mm256_add_ps(fjy0,ty);
1108             fjz0             = _mm256_add_ps(fjz0,tz);
1109
1110             }
1111
1112             /**************************
1113              * CALCULATE INTERACTIONS *
1114              **************************/
1115
1116             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1117             {
1118
1119             r20              = _mm256_mul_ps(rsq20,rinv20);
1120
1121             /* Compute parameters for interactions between i and j atoms */
1122             qq20             = _mm256_mul_ps(iq2,jq0);
1123
1124             /* EWALD ELECTROSTATICS */
1125
1126             /* Analytical PME correction */
1127             zeta2            = _mm256_mul_ps(beta2,rsq20);
1128             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1129             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1130             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1131             felec            = _mm256_mul_ps(qq20,felec);
1132             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1133             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1134             velec            = _mm256_sub_ps(rinv20,pmecorrV);
1135             velec            = _mm256_mul_ps(qq20,velec);
1136
1137             d                = _mm256_sub_ps(r20,rswitch);
1138             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1139             d2               = _mm256_mul_ps(d,d);
1140             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)))))));
1141
1142             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1143
1144             /* Evaluate switch function */
1145             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1146             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv20,_mm256_mul_ps(velec,dsw)) );
1147             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1148
1149             fscal            = felec;
1150
1151             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1152
1153             /* Calculate temporary vectorial force */
1154             tx               = _mm256_mul_ps(fscal,dx20);
1155             ty               = _mm256_mul_ps(fscal,dy20);
1156             tz               = _mm256_mul_ps(fscal,dz20);
1157
1158             /* Update vectorial force */
1159             fix2             = _mm256_add_ps(fix2,tx);
1160             fiy2             = _mm256_add_ps(fiy2,ty);
1161             fiz2             = _mm256_add_ps(fiz2,tz);
1162
1163             fjx0             = _mm256_add_ps(fjx0,tx);
1164             fjy0             = _mm256_add_ps(fjy0,ty);
1165             fjz0             = _mm256_add_ps(fjz0,tz);
1166
1167             }
1168
1169             fjptrA             = f+j_coord_offsetA;
1170             fjptrB             = f+j_coord_offsetB;
1171             fjptrC             = f+j_coord_offsetC;
1172             fjptrD             = f+j_coord_offsetD;
1173             fjptrE             = f+j_coord_offsetE;
1174             fjptrF             = f+j_coord_offsetF;
1175             fjptrG             = f+j_coord_offsetG;
1176             fjptrH             = f+j_coord_offsetH;
1177
1178             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1179
1180             /* Inner loop uses 318 flops */
1181         }
1182
1183         if(jidx<j_index_end)
1184         {
1185
1186             /* Get j neighbor index, and coordinate index */
1187             jnrlistA         = jjnr[jidx];
1188             jnrlistB         = jjnr[jidx+1];
1189             jnrlistC         = jjnr[jidx+2];
1190             jnrlistD         = jjnr[jidx+3];
1191             jnrlistE         = jjnr[jidx+4];
1192             jnrlistF         = jjnr[jidx+5];
1193             jnrlistG         = jjnr[jidx+6];
1194             jnrlistH         = jjnr[jidx+7];
1195             /* Sign of each element will be negative for non-real atoms.
1196              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1197              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1198              */
1199             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1200                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1201
1202             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1203             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1204             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1205             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1206             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1207             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1208             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1209             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1210             j_coord_offsetA  = DIM*jnrA;
1211             j_coord_offsetB  = DIM*jnrB;
1212             j_coord_offsetC  = DIM*jnrC;
1213             j_coord_offsetD  = DIM*jnrD;
1214             j_coord_offsetE  = DIM*jnrE;
1215             j_coord_offsetF  = DIM*jnrF;
1216             j_coord_offsetG  = DIM*jnrG;
1217             j_coord_offsetH  = DIM*jnrH;
1218
1219             /* load j atom coordinates */
1220             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1221                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1222                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1223                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1224                                                  &jx0,&jy0,&jz0);
1225
1226             /* Calculate displacement vector */
1227             dx00             = _mm256_sub_ps(ix0,jx0);
1228             dy00             = _mm256_sub_ps(iy0,jy0);
1229             dz00             = _mm256_sub_ps(iz0,jz0);
1230             dx10             = _mm256_sub_ps(ix1,jx0);
1231             dy10             = _mm256_sub_ps(iy1,jy0);
1232             dz10             = _mm256_sub_ps(iz1,jz0);
1233             dx20             = _mm256_sub_ps(ix2,jx0);
1234             dy20             = _mm256_sub_ps(iy2,jy0);
1235             dz20             = _mm256_sub_ps(iz2,jz0);
1236
1237             /* Calculate squared distance and things based on it */
1238             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1239             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1240             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1241
1242             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1243             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1244             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1245
1246             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1247             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1248             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1249
1250             /* Load parameters for j particles */
1251             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1252                                                                  charge+jnrC+0,charge+jnrD+0,
1253                                                                  charge+jnrE+0,charge+jnrF+0,
1254                                                                  charge+jnrG+0,charge+jnrH+0);
1255
1256             fjx0             = _mm256_setzero_ps();
1257             fjy0             = _mm256_setzero_ps();
1258             fjz0             = _mm256_setzero_ps();
1259
1260             /**************************
1261              * CALCULATE INTERACTIONS *
1262              **************************/
1263
1264             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1265             {
1266
1267             r00              = _mm256_mul_ps(rsq00,rinv00);
1268             r00              = _mm256_andnot_ps(dummy_mask,r00);
1269
1270             /* Compute parameters for interactions between i and j atoms */
1271             qq00             = _mm256_mul_ps(iq0,jq0);
1272
1273             /* EWALD ELECTROSTATICS */
1274
1275             /* Analytical PME correction */
1276             zeta2            = _mm256_mul_ps(beta2,rsq00);
1277             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1278             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1279             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1280             felec            = _mm256_mul_ps(qq00,felec);
1281             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1282             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1283             velec            = _mm256_sub_ps(rinv00,pmecorrV);
1284             velec            = _mm256_mul_ps(qq00,velec);
1285
1286             d                = _mm256_sub_ps(r00,rswitch);
1287             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1288             d2               = _mm256_mul_ps(d,d);
1289             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)))))));
1290
1291             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1292
1293             /* Evaluate switch function */
1294             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1295             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
1296             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1297
1298             fscal            = felec;
1299
1300             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1301
1302             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1303
1304             /* Calculate temporary vectorial force */
1305             tx               = _mm256_mul_ps(fscal,dx00);
1306             ty               = _mm256_mul_ps(fscal,dy00);
1307             tz               = _mm256_mul_ps(fscal,dz00);
1308
1309             /* Update vectorial force */
1310             fix0             = _mm256_add_ps(fix0,tx);
1311             fiy0             = _mm256_add_ps(fiy0,ty);
1312             fiz0             = _mm256_add_ps(fiz0,tz);
1313
1314             fjx0             = _mm256_add_ps(fjx0,tx);
1315             fjy0             = _mm256_add_ps(fjy0,ty);
1316             fjz0             = _mm256_add_ps(fjz0,tz);
1317
1318             }
1319
1320             /**************************
1321              * CALCULATE INTERACTIONS *
1322              **************************/
1323
1324             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1325             {
1326
1327             r10              = _mm256_mul_ps(rsq10,rinv10);
1328             r10              = _mm256_andnot_ps(dummy_mask,r10);
1329
1330             /* Compute parameters for interactions between i and j atoms */
1331             qq10             = _mm256_mul_ps(iq1,jq0);
1332
1333             /* EWALD ELECTROSTATICS */
1334
1335             /* Analytical PME correction */
1336             zeta2            = _mm256_mul_ps(beta2,rsq10);
1337             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1338             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1339             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1340             felec            = _mm256_mul_ps(qq10,felec);
1341             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1342             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1343             velec            = _mm256_sub_ps(rinv10,pmecorrV);
1344             velec            = _mm256_mul_ps(qq10,velec);
1345
1346             d                = _mm256_sub_ps(r10,rswitch);
1347             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1348             d2               = _mm256_mul_ps(d,d);
1349             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)))))));
1350
1351             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1352
1353             /* Evaluate switch function */
1354             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1355             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv10,_mm256_mul_ps(velec,dsw)) );
1356             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1357
1358             fscal            = felec;
1359
1360             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1361
1362             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1363
1364             /* Calculate temporary vectorial force */
1365             tx               = _mm256_mul_ps(fscal,dx10);
1366             ty               = _mm256_mul_ps(fscal,dy10);
1367             tz               = _mm256_mul_ps(fscal,dz10);
1368
1369             /* Update vectorial force */
1370             fix1             = _mm256_add_ps(fix1,tx);
1371             fiy1             = _mm256_add_ps(fiy1,ty);
1372             fiz1             = _mm256_add_ps(fiz1,tz);
1373
1374             fjx0             = _mm256_add_ps(fjx0,tx);
1375             fjy0             = _mm256_add_ps(fjy0,ty);
1376             fjz0             = _mm256_add_ps(fjz0,tz);
1377
1378             }
1379
1380             /**************************
1381              * CALCULATE INTERACTIONS *
1382              **************************/
1383
1384             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1385             {
1386
1387             r20              = _mm256_mul_ps(rsq20,rinv20);
1388             r20              = _mm256_andnot_ps(dummy_mask,r20);
1389
1390             /* Compute parameters for interactions between i and j atoms */
1391             qq20             = _mm256_mul_ps(iq2,jq0);
1392
1393             /* EWALD ELECTROSTATICS */
1394
1395             /* Analytical PME correction */
1396             zeta2            = _mm256_mul_ps(beta2,rsq20);
1397             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1398             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1399             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1400             felec            = _mm256_mul_ps(qq20,felec);
1401             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1402             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1403             velec            = _mm256_sub_ps(rinv20,pmecorrV);
1404             velec            = _mm256_mul_ps(qq20,velec);
1405
1406             d                = _mm256_sub_ps(r20,rswitch);
1407             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1408             d2               = _mm256_mul_ps(d,d);
1409             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)))))));
1410
1411             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1412
1413             /* Evaluate switch function */
1414             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1415             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv20,_mm256_mul_ps(velec,dsw)) );
1416             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1417
1418             fscal            = felec;
1419
1420             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1421
1422             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1423
1424             /* Calculate temporary vectorial force */
1425             tx               = _mm256_mul_ps(fscal,dx20);
1426             ty               = _mm256_mul_ps(fscal,dy20);
1427             tz               = _mm256_mul_ps(fscal,dz20);
1428
1429             /* Update vectorial force */
1430             fix2             = _mm256_add_ps(fix2,tx);
1431             fiy2             = _mm256_add_ps(fiy2,ty);
1432             fiz2             = _mm256_add_ps(fiz2,tz);
1433
1434             fjx0             = _mm256_add_ps(fjx0,tx);
1435             fjy0             = _mm256_add_ps(fjy0,ty);
1436             fjz0             = _mm256_add_ps(fjz0,tz);
1437
1438             }
1439
1440             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1441             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1442             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1443             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1444             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1445             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1446             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1447             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1448
1449             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1450
1451             /* Inner loop uses 321 flops */
1452         }
1453
1454         /* End of innermost loop */
1455
1456         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1457                                                  f+i_coord_offset,fshift+i_shift_offset);
1458
1459         /* Increment number of inner iterations */
1460         inneriter                  += j_index_end - j_index_start;
1461
1462         /* Outer loop uses 18 flops */
1463     }
1464
1465     /* Increment number of outer iterations */
1466     outeriter        += nri;
1467
1468     /* Update outer/inner flops */
1469
1470     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*321);
1471 }