src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEwSw_VdwLJSw_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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_VF_avx_256_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSw_VdwLJSw_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     int              nvdwtype;
 102     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 106     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 107     __m256i          ewitab;
 108     __m128i          ewitab_lo,ewitab_hi;
 109     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 110     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 111     real             *ewtab;
 112     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 113     real             rswitch_scalar,d_scalar;
 114     __m256           dummy_mask,cutoff_mask;
 115     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 116     __m256           one     = _mm256_set1_ps(1.0);
 117     __m256           two     = _mm256_set1_ps(2.0);
 118     x                = xx[0];
 119     f                = ff[0];
 120
 121     nri              = nlist->nri;
 122     iinr             = nlist->iinr;
 123     jindex           = nlist->jindex;
 124     jjnr             = nlist->jjnr;
 125     shiftidx         = nlist->shift;
 126     gid              = nlist->gid;
 127     shiftvec         = fr->shift_vec[0];
 128     fshift           = fr->fshift[0];
 129     facel            = _mm256_set1_ps(fr->epsfac);
 130     charge           = mdatoms->chargeA;
 131     nvdwtype         = fr->ntype;
 132     vdwparam         = fr->nbfp;
 133     vdwtype          = mdatoms->typeA;
 134
 135     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 136     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 137     beta2            = _mm256_mul_ps(beta,beta);
 138     beta3            = _mm256_mul_ps(beta,beta2);
 139
 140     ewtab            = fr->ic->tabq_coul_FDV0;
 141     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 142     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 143
 144     /* Setup water-specific parameters */
 145     inr              = nlist->iinr[0];
 146     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 147     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 148     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 149     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 150
 151     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 152     rcutoff_scalar   = fr->rcoulomb;
 153     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 154     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 155
 156     rswitch_scalar   = fr->rcoulomb_switch;
 157     rswitch          = _mm256_set1_ps(rswitch_scalar);
 158     /* Setup switch parameters */
 159     d_scalar         = rcutoff_scalar-rswitch_scalar;
 160     d                = _mm256_set1_ps(d_scalar);
 161     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 162     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 163     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 164     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 165     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 166     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 167
 168     /* Avoid stupid compiler warnings */
 169     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 170     j_coord_offsetA = 0;
 171     j_coord_offsetB = 0;
 172     j_coord_offsetC = 0;
 173     j_coord_offsetD = 0;
 174     j_coord_offsetE = 0;
 175     j_coord_offsetF = 0;
 176     j_coord_offsetG = 0;
 177     j_coord_offsetH = 0;
 178
 179     outeriter        = 0;
 180     inneriter        = 0;
 181
 182     for(iidx=0;iidx<4*DIM;iidx++)
 183     {
 184         scratch[iidx] = 0.0;
 185     }
 186
 187     /* Start outer loop over neighborlists */
 188     for(iidx=0; iidx<nri; iidx++)
 189     {
 190         /* Load shift vector for this list */
 191         i_shift_offset   = DIM*shiftidx[iidx];
 192
 193         /* Load limits for loop over neighbors */
 194         j_index_start    = jindex[iidx];
 195         j_index_end      = jindex[iidx+1];
 196
 197         /* Get outer coordinate index */
 198         inr              = iinr[iidx];
 199         i_coord_offset   = DIM*inr;
 200
 201         /* Load i particle coords and add shift vector */
 202         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 203                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 204
 205         fix0             = _mm256_setzero_ps();
 206         fiy0             = _mm256_setzero_ps();
 207         fiz0             = _mm256_setzero_ps();
 208         fix1             = _mm256_setzero_ps();
 209         fiy1             = _mm256_setzero_ps();
 210         fiz1             = _mm256_setzero_ps();
 211         fix2             = _mm256_setzero_ps();
 212         fiy2             = _mm256_setzero_ps();
 213         fiz2             = _mm256_setzero_ps();
 214
 215         /* Reset potential sums */
 216         velecsum         = _mm256_setzero_ps();
 217         vvdwsum          = _mm256_setzero_ps();
 218
 219         /* Start inner kernel loop */
 220         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 221         {
 222
 223             /* Get j neighbor index, and coordinate index */
 224             jnrA             = jjnr[jidx];
 225             jnrB             = jjnr[jidx+1];
 226             jnrC             = jjnr[jidx+2];
 227             jnrD             = jjnr[jidx+3];
 228             jnrE             = jjnr[jidx+4];
 229             jnrF             = jjnr[jidx+5];
 230             jnrG             = jjnr[jidx+6];
 231             jnrH             = jjnr[jidx+7];
 232             j_coord_offsetA  = DIM*jnrA;
 233             j_coord_offsetB  = DIM*jnrB;
 234             j_coord_offsetC  = DIM*jnrC;
 235             j_coord_offsetD  = DIM*jnrD;
 236             j_coord_offsetE  = DIM*jnrE;
 237             j_coord_offsetF  = DIM*jnrF;
 238             j_coord_offsetG  = DIM*jnrG;
 239             j_coord_offsetH  = DIM*jnrH;
 240
 241             /* load j atom coordinates */
 242             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 243                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 244                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 245                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 246                                                  &jx0,&jy0,&jz0);
 247
 248             /* Calculate displacement vector */
 249             dx00             = _mm256_sub_ps(ix0,jx0);
 250             dy00             = _mm256_sub_ps(iy0,jy0);
 251             dz00             = _mm256_sub_ps(iz0,jz0);
 252             dx10             = _mm256_sub_ps(ix1,jx0);
 253             dy10             = _mm256_sub_ps(iy1,jy0);
 254             dz10             = _mm256_sub_ps(iz1,jz0);
 255             dx20             = _mm256_sub_ps(ix2,jx0);
 256             dy20             = _mm256_sub_ps(iy2,jy0);
 257             dz20             = _mm256_sub_ps(iz2,jz0);
 258
 259             /* Calculate squared distance and things based on it */
 260             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 261             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 262             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 263
 264             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 265             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 266             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 267
 268             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 269             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 270             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 271
 272             /* Load parameters for j particles */
 273             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 274                                                                  charge+jnrC+0,charge+jnrD+0,
 275                                                                  charge+jnrE+0,charge+jnrF+0,
 276                                                                  charge+jnrG+0,charge+jnrH+0);
 277             vdwjidx0A        = 2*vdwtype[jnrA+0];
 278             vdwjidx0B        = 2*vdwtype[jnrB+0];
 279             vdwjidx0C        = 2*vdwtype[jnrC+0];
 280             vdwjidx0D        = 2*vdwtype[jnrD+0];
 281             vdwjidx0E        = 2*vdwtype[jnrE+0];
 282             vdwjidx0F        = 2*vdwtype[jnrF+0];
 283             vdwjidx0G        = 2*vdwtype[jnrG+0];
 284             vdwjidx0H        = 2*vdwtype[jnrH+0];
 285
 286             fjx0             = _mm256_setzero_ps();
 287             fjy0             = _mm256_setzero_ps();
 288             fjz0             = _mm256_setzero_ps();
 289
 290             /**************************
 291              * CALCULATE INTERACTIONS *
 292              **************************/
 293
 294             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 295             {
 296
 297             r00              = _mm256_mul_ps(rsq00,rinv00);
 298
 299             /* Compute parameters for interactions between i and j atoms */
 300             qq00             = _mm256_mul_ps(iq0,jq0);
 301             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 302                                             vdwioffsetptr0+vdwjidx0B,
 303                                             vdwioffsetptr0+vdwjidx0C,
 304                                             vdwioffsetptr0+vdwjidx0D,
 305                                             vdwioffsetptr0+vdwjidx0E,
 306                                             vdwioffsetptr0+vdwjidx0F,
 307                                             vdwioffsetptr0+vdwjidx0G,
 308                                             vdwioffsetptr0+vdwjidx0H,
 309                                             &c6_00,&c12_00);
 310
 311             /* EWALD ELECTROSTATICS */
 312
 313             /* Analytical PME correction */
 314             zeta2            = _mm256_mul_ps(beta2,rsq00);
 315             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 316             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 317             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 318             felec            = _mm256_mul_ps(qq00,felec);
 319             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 320             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 321             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 322             velec            = _mm256_mul_ps(qq00,velec);
 323
 324             /* LENNARD-JONES DISPERSION/REPULSION */
 325
 326             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 327             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 328             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 329             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 330             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 331
 332             d                = _mm256_sub_ps(r00,rswitch);
 333             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 334             d2               = _mm256_mul_ps(d,d);
 335             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)))))));
 336
 337             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 338
 339             /* Evaluate switch function */
 340             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 341             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
 342             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 343             velec            = _mm256_mul_ps(velec,sw);
 344             vvdw             = _mm256_mul_ps(vvdw,sw);
 345             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 346
 347             /* Update potential sum for this i atom from the interaction with this j atom. */
 348             velec            = _mm256_and_ps(velec,cutoff_mask);
 349             velecsum         = _mm256_add_ps(velecsum,velec);
 350             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 351             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 352
 353             fscal            = _mm256_add_ps(felec,fvdw);
 354
 355             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 356
 357             /* Calculate temporary vectorial force */
 358             tx               = _mm256_mul_ps(fscal,dx00);
 359             ty               = _mm256_mul_ps(fscal,dy00);
 360             tz               = _mm256_mul_ps(fscal,dz00);
 361
 362             /* Update vectorial force */
 363             fix0             = _mm256_add_ps(fix0,tx);
 364             fiy0             = _mm256_add_ps(fiy0,ty);
 365             fiz0             = _mm256_add_ps(fiz0,tz);
 366
 367             fjx0             = _mm256_add_ps(fjx0,tx);
 368             fjy0             = _mm256_add_ps(fjy0,ty);
 369             fjz0             = _mm256_add_ps(fjz0,tz);
 370
 371             }
 372
 373             /**************************
 374              * CALCULATE INTERACTIONS *
 375              **************************/
 376
 377             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 378             {
 379
 380             r10              = _mm256_mul_ps(rsq10,rinv10);
 381
 382             /* Compute parameters for interactions between i and j atoms */
 383             qq10             = _mm256_mul_ps(iq1,jq0);
 384
 385             /* EWALD ELECTROSTATICS */
 386
 387             /* Analytical PME correction */
 388             zeta2            = _mm256_mul_ps(beta2,rsq10);
 389             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 390             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 391             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 392             felec            = _mm256_mul_ps(qq10,felec);
 393             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 394             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 395             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 396             velec            = _mm256_mul_ps(qq10,velec);
 397
 398             d                = _mm256_sub_ps(r10,rswitch);
 399             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 400             d2               = _mm256_mul_ps(d,d);
 401             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)))))));
 402
 403             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 404
 405             /* Evaluate switch function */
 406             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 407             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv10,_mm256_mul_ps(velec,dsw)) );
 408             velec            = _mm256_mul_ps(velec,sw);
 409             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 410
 411             /* Update potential sum for this i atom from the interaction with this j atom. */
 412             velec            = _mm256_and_ps(velec,cutoff_mask);
 413             velecsum         = _mm256_add_ps(velecsum,velec);
 414
 415             fscal            = felec;
 416
 417             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 418
 419             /* Calculate temporary vectorial force */
 420             tx               = _mm256_mul_ps(fscal,dx10);
 421             ty               = _mm256_mul_ps(fscal,dy10);
 422             tz               = _mm256_mul_ps(fscal,dz10);
 423
 424             /* Update vectorial force */
 425             fix1             = _mm256_add_ps(fix1,tx);
 426             fiy1             = _mm256_add_ps(fiy1,ty);
 427             fiz1             = _mm256_add_ps(fiz1,tz);
 428
 429             fjx0             = _mm256_add_ps(fjx0,tx);
 430             fjy0             = _mm256_add_ps(fjy0,ty);
 431             fjz0             = _mm256_add_ps(fjz0,tz);
 432
 433             }
 434
 435             /**************************
 436              * CALCULATE INTERACTIONS *
 437              **************************/
 438
 439             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 440             {
 441
 442             r20              = _mm256_mul_ps(rsq20,rinv20);
 443
 444             /* Compute parameters for interactions between i and j atoms */
 445             qq20             = _mm256_mul_ps(iq2,jq0);
 446
 447             /* EWALD ELECTROSTATICS */
 448
 449             /* Analytical PME correction */
 450             zeta2            = _mm256_mul_ps(beta2,rsq20);
 451             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 452             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 453             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 454             felec            = _mm256_mul_ps(qq20,felec);
 455             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 456             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 457             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 458             velec            = _mm256_mul_ps(qq20,velec);
 459
 460             d                = _mm256_sub_ps(r20,rswitch);
 461             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 462             d2               = _mm256_mul_ps(d,d);
 463             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)))))));
 464
 465             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 466
 467             /* Evaluate switch function */
 468             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 469             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv20,_mm256_mul_ps(velec,dsw)) );
 470             velec            = _mm256_mul_ps(velec,sw);
 471             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 472
 473             /* Update potential sum for this i atom from the interaction with this j atom. */
 474             velec            = _mm256_and_ps(velec,cutoff_mask);
 475             velecsum         = _mm256_add_ps(velecsum,velec);
 476
 477             fscal            = felec;
 478
 479             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 480
 481             /* Calculate temporary vectorial force */
 482             tx               = _mm256_mul_ps(fscal,dx20);
 483             ty               = _mm256_mul_ps(fscal,dy20);
 484             tz               = _mm256_mul_ps(fscal,dz20);
 485
 486             /* Update vectorial force */
 487             fix2             = _mm256_add_ps(fix2,tx);
 488             fiy2             = _mm256_add_ps(fiy2,ty);
 489             fiz2             = _mm256_add_ps(fiz2,tz);
 490
 491             fjx0             = _mm256_add_ps(fjx0,tx);
 492             fjy0             = _mm256_add_ps(fjy0,ty);
 493             fjz0             = _mm256_add_ps(fjz0,tz);
 494
 495             }
 496
 497             fjptrA             = f+j_coord_offsetA;
 498             fjptrB             = f+j_coord_offsetB;
 499             fjptrC             = f+j_coord_offsetC;
 500             fjptrD             = f+j_coord_offsetD;
 501             fjptrE             = f+j_coord_offsetE;
 502             fjptrF             = f+j_coord_offsetF;
 503             fjptrG             = f+j_coord_offsetG;
 504             fjptrH             = f+j_coord_offsetH;
 505
 506             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 507
 508             /* Inner loop uses 345 flops */
 509         }
 510
 511         if(jidx<j_index_end)
 512         {
 513
 514             /* Get j neighbor index, and coordinate index */
 515             jnrlistA         = jjnr[jidx];
 516             jnrlistB         = jjnr[jidx+1];
 517             jnrlistC         = jjnr[jidx+2];
 518             jnrlistD         = jjnr[jidx+3];
 519             jnrlistE         = jjnr[jidx+4];
 520             jnrlistF         = jjnr[jidx+5];
 521             jnrlistG         = jjnr[jidx+6];
 522             jnrlistH         = jjnr[jidx+7];
 523             /* Sign of each element will be negative for non-real atoms.
 524              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 525              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 526              */
 527             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 528                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 529
 530             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 531             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 532             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 533             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 534             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 535             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 536             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 537             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 538             j_coord_offsetA  = DIM*jnrA;
 539             j_coord_offsetB  = DIM*jnrB;
 540             j_coord_offsetC  = DIM*jnrC;
 541             j_coord_offsetD  = DIM*jnrD;
 542             j_coord_offsetE  = DIM*jnrE;
 543             j_coord_offsetF  = DIM*jnrF;
 544             j_coord_offsetG  = DIM*jnrG;
 545             j_coord_offsetH  = DIM*jnrH;
 546
 547             /* load j atom coordinates */
 548             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 549                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 550                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 551                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 552                                                  &jx0,&jy0,&jz0);
 553
 554             /* Calculate displacement vector */
 555             dx00             = _mm256_sub_ps(ix0,jx0);
 556             dy00             = _mm256_sub_ps(iy0,jy0);
 557             dz00             = _mm256_sub_ps(iz0,jz0);
 558             dx10             = _mm256_sub_ps(ix1,jx0);
 559             dy10             = _mm256_sub_ps(iy1,jy0);
 560             dz10             = _mm256_sub_ps(iz1,jz0);
 561             dx20             = _mm256_sub_ps(ix2,jx0);
 562             dy20             = _mm256_sub_ps(iy2,jy0);
 563             dz20             = _mm256_sub_ps(iz2,jz0);
 564
 565             /* Calculate squared distance and things based on it */
 566             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 567             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 568             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 569
 570             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 571             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 572             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 573
 574             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 575             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 576             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 577
 578             /* Load parameters for j particles */
 579             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 580                                                                  charge+jnrC+0,charge+jnrD+0,
 581                                                                  charge+jnrE+0,charge+jnrF+0,
 582                                                                  charge+jnrG+0,charge+jnrH+0);
 583             vdwjidx0A        = 2*vdwtype[jnrA+0];
 584             vdwjidx0B        = 2*vdwtype[jnrB+0];
 585             vdwjidx0C        = 2*vdwtype[jnrC+0];
 586             vdwjidx0D        = 2*vdwtype[jnrD+0];
 587             vdwjidx0E        = 2*vdwtype[jnrE+0];
 588             vdwjidx0F        = 2*vdwtype[jnrF+0];
 589             vdwjidx0G        = 2*vdwtype[jnrG+0];
 590             vdwjidx0H        = 2*vdwtype[jnrH+0];
 591
 592             fjx0             = _mm256_setzero_ps();
 593             fjy0             = _mm256_setzero_ps();
 594             fjz0             = _mm256_setzero_ps();
 595
 596             /**************************
 597              * CALCULATE INTERACTIONS *
 598              **************************/
 599
 600             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 601             {
 602
 603             r00              = _mm256_mul_ps(rsq00,rinv00);
 604             r00              = _mm256_andnot_ps(dummy_mask,r00);
 605
 606             /* Compute parameters for interactions between i and j atoms */
 607             qq00             = _mm256_mul_ps(iq0,jq0);
 608             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 609                                             vdwioffsetptr0+vdwjidx0B,
 610                                             vdwioffsetptr0+vdwjidx0C,
 611                                             vdwioffsetptr0+vdwjidx0D,
 612                                             vdwioffsetptr0+vdwjidx0E,
 613                                             vdwioffsetptr0+vdwjidx0F,
 614                                             vdwioffsetptr0+vdwjidx0G,
 615                                             vdwioffsetptr0+vdwjidx0H,
 616                                             &c6_00,&c12_00);
 617
 618             /* EWALD ELECTROSTATICS */
 619
 620             /* Analytical PME correction */
 621             zeta2            = _mm256_mul_ps(beta2,rsq00);
 622             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 623             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 624             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 625             felec            = _mm256_mul_ps(qq00,felec);
 626             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 627             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 628             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 629             velec            = _mm256_mul_ps(qq00,velec);
 630
 631             /* LENNARD-JONES DISPERSION/REPULSION */
 632
 633             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 634             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 635             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 636             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 637             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 638
 639             d                = _mm256_sub_ps(r00,rswitch);
 640             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 641             d2               = _mm256_mul_ps(d,d);
 642             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)))))));
 643
 644             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 645
 646             /* Evaluate switch function */
 647             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 648             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
 649             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 650             velec            = _mm256_mul_ps(velec,sw);
 651             vvdw             = _mm256_mul_ps(vvdw,sw);
 652             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 653
 654             /* Update potential sum for this i atom from the interaction with this j atom. */
 655             velec            = _mm256_and_ps(velec,cutoff_mask);
 656             velec            = _mm256_andnot_ps(dummy_mask,velec);
 657             velecsum         = _mm256_add_ps(velecsum,velec);
 658             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 659             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 660             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 661
 662             fscal            = _mm256_add_ps(felec,fvdw);
 663
 664             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 665
 666             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 667
 668             /* Calculate temporary vectorial force */
 669             tx               = _mm256_mul_ps(fscal,dx00);
 670             ty               = _mm256_mul_ps(fscal,dy00);
 671             tz               = _mm256_mul_ps(fscal,dz00);
 672
 673             /* Update vectorial force */
 674             fix0             = _mm256_add_ps(fix0,tx);
 675             fiy0             = _mm256_add_ps(fiy0,ty);
 676             fiz0             = _mm256_add_ps(fiz0,tz);
 677
 678             fjx0             = _mm256_add_ps(fjx0,tx);
 679             fjy0             = _mm256_add_ps(fjy0,ty);
 680             fjz0             = _mm256_add_ps(fjz0,tz);
 681
 682             }
 683
 684             /**************************
 685              * CALCULATE INTERACTIONS *
 686              **************************/
 687
 688             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 689             {
 690
 691             r10              = _mm256_mul_ps(rsq10,rinv10);
 692             r10              = _mm256_andnot_ps(dummy_mask,r10);
 693
 694             /* Compute parameters for interactions between i and j atoms */
 695             qq10             = _mm256_mul_ps(iq1,jq0);
 696
 697             /* EWALD ELECTROSTATICS */
 698
 699             /* Analytical PME correction */
 700             zeta2            = _mm256_mul_ps(beta2,rsq10);
 701             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 702             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 703             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 704             felec            = _mm256_mul_ps(qq10,felec);
 705             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 706             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 707             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 708             velec            = _mm256_mul_ps(qq10,velec);
 709
 710             d                = _mm256_sub_ps(r10,rswitch);
 711             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 712             d2               = _mm256_mul_ps(d,d);
 713             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)))))));
 714
 715             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 716
 717             /* Evaluate switch function */
 718             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 719             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv10,_mm256_mul_ps(velec,dsw)) );
 720             velec            = _mm256_mul_ps(velec,sw);
 721             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 722
 723             /* Update potential sum for this i atom from the interaction with this j atom. */
 724             velec            = _mm256_and_ps(velec,cutoff_mask);
 725             velec            = _mm256_andnot_ps(dummy_mask,velec);
 726             velecsum         = _mm256_add_ps(velecsum,velec);
 727
 728             fscal            = felec;
 729
 730             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 731
 732             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 733
 734             /* Calculate temporary vectorial force */
 735             tx               = _mm256_mul_ps(fscal,dx10);
 736             ty               = _mm256_mul_ps(fscal,dy10);
 737             tz               = _mm256_mul_ps(fscal,dz10);
 738
 739             /* Update vectorial force */
 740             fix1             = _mm256_add_ps(fix1,tx);
 741             fiy1             = _mm256_add_ps(fiy1,ty);
 742             fiz1             = _mm256_add_ps(fiz1,tz);
 743
 744             fjx0             = _mm256_add_ps(fjx0,tx);
 745             fjy0             = _mm256_add_ps(fjy0,ty);
 746             fjz0             = _mm256_add_ps(fjz0,tz);
 747
 748             }
 749
 750             /**************************
 751              * CALCULATE INTERACTIONS *
 752              **************************/
 753
 754             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 755             {
 756
 757             r20              = _mm256_mul_ps(rsq20,rinv20);
 758             r20              = _mm256_andnot_ps(dummy_mask,r20);
 759
 760             /* Compute parameters for interactions between i and j atoms */
 761             qq20             = _mm256_mul_ps(iq2,jq0);
 762
 763             /* EWALD ELECTROSTATICS */
 764
 765             /* Analytical PME correction */
 766             zeta2            = _mm256_mul_ps(beta2,rsq20);
 767             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 768             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 769             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 770             felec            = _mm256_mul_ps(qq20,felec);
 771             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 772             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 773             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 774             velec            = _mm256_mul_ps(qq20,velec);
 775
 776             d                = _mm256_sub_ps(r20,rswitch);
 777             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 778             d2               = _mm256_mul_ps(d,d);
 779             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)))))));
 780
 781             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 782
 783             /* Evaluate switch function */
 784             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 785             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv20,_mm256_mul_ps(velec,dsw)) );
 786             velec            = _mm256_mul_ps(velec,sw);
 787             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 788
 789             /* Update potential sum for this i atom from the interaction with this j atom. */
 790             velec            = _mm256_and_ps(velec,cutoff_mask);
 791             velec            = _mm256_andnot_ps(dummy_mask,velec);
 792             velecsum         = _mm256_add_ps(velecsum,velec);
 793
 794             fscal            = felec;
 795
 796             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 797
 798             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 799
 800             /* Calculate temporary vectorial force */
 801             tx               = _mm256_mul_ps(fscal,dx20);
 802             ty               = _mm256_mul_ps(fscal,dy20);
 803             tz               = _mm256_mul_ps(fscal,dz20);
 804
 805             /* Update vectorial force */
 806             fix2             = _mm256_add_ps(fix2,tx);
 807             fiy2             = _mm256_add_ps(fiy2,ty);
 808             fiz2             = _mm256_add_ps(fiz2,tz);
 809
 810             fjx0             = _mm256_add_ps(fjx0,tx);
 811             fjy0             = _mm256_add_ps(fjy0,ty);
 812             fjz0             = _mm256_add_ps(fjz0,tz);
 813
 814             }
 815
 816             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 817             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 818             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 819             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 820             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 821             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 822             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 823             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 824
 825             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 826
 827             /* Inner loop uses 348 flops */
 828         }
 829
 830         /* End of innermost loop */
 831
 832         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 833                                                  f+i_coord_offset,fshift+i_shift_offset);
 834
 835         ggid                        = gid[iidx];
 836         /* Update potential energies */
 837         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 838         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 839
 840         /* Increment number of inner iterations */
 841         inneriter                  += j_index_end - j_index_start;
 842
 843         /* Outer loop uses 20 flops */
 844     }
 845
 846     /* Increment number of outer iterations */
 847     outeriter        += nri;
 848
 849     /* Update outer/inner flops */
 850
 851     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*348);
 852 }
 853 /*
 854  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_avx_256_single
 855  * Electrostatics interaction: Ewald
 856  * VdW interaction:            LennardJones
 857  * Geometry:                   Water3-Particle
 858  * Calculate force/pot:        Force
 859  */
 860 void
 861 nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_avx_256_single
 862                     (t_nblist                    * gmx_restrict       nlist,
 863                      rvec                        * gmx_restrict          xx,
 864                      rvec                        * gmx_restrict          ff,
 865                      t_forcerec                  * gmx_restrict          fr,
 866                      t_mdatoms                   * gmx_restrict     mdatoms,
 867                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 868                      t_nrnb                      * gmx_restrict        nrnb)
 869 {
 870     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 871      * just 0 for non-waters.
 872      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 873      * jnr indices corresponding to data put in the four positions in the SIMD register.
 874      */
 875     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 876     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 877     int              jnrA,jnrB,jnrC,jnrD;
 878     int              jnrE,jnrF,jnrG,jnrH;
 879     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 880     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 881     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 882     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 883     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 884     real             rcutoff_scalar;
 885     real             *shiftvec,*fshift,*x,*f;
 886     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 887     real             scratch[4*DIM];
 888     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 889     real *           vdwioffsetptr0;
 890     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 891     real *           vdwioffsetptr1;
 892     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 893     real *           vdwioffsetptr2;
 894     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 895     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 896     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 897     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 898     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 899     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 900     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 901     real             *charge;
 902     int              nvdwtype;
 903     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 904     int              *vdwtype;
 905     real             *vdwparam;
 906     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 907     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 908     __m256i          ewitab;
 909     __m128i          ewitab_lo,ewitab_hi;
 910     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 911     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 912     real             *ewtab;
 913     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 914     real             rswitch_scalar,d_scalar;
 915     __m256           dummy_mask,cutoff_mask;
 916     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 917     __m256           one     = _mm256_set1_ps(1.0);
 918     __m256           two     = _mm256_set1_ps(2.0);
 919     x                = xx[0];
 920     f                = ff[0];
 921
 922     nri              = nlist->nri;
 923     iinr             = nlist->iinr;
 924     jindex           = nlist->jindex;
 925     jjnr             = nlist->jjnr;
 926     shiftidx         = nlist->shift;
 927     gid              = nlist->gid;
 928     shiftvec         = fr->shift_vec[0];
 929     fshift           = fr->fshift[0];
 930     facel            = _mm256_set1_ps(fr->epsfac);
 931     charge           = mdatoms->chargeA;
 932     nvdwtype         = fr->ntype;
 933     vdwparam         = fr->nbfp;
 934     vdwtype          = mdatoms->typeA;
 935
 936     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 937     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 938     beta2            = _mm256_mul_ps(beta,beta);
 939     beta3            = _mm256_mul_ps(beta,beta2);
 940
 941     ewtab            = fr->ic->tabq_coul_FDV0;
 942     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 943     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 944
 945     /* Setup water-specific parameters */
 946     inr              = nlist->iinr[0];
 947     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 948     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 949     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 950     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 951
 952     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 953     rcutoff_scalar   = fr->rcoulomb;
 954     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 955     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 956
 957     rswitch_scalar   = fr->rcoulomb_switch;
 958     rswitch          = _mm256_set1_ps(rswitch_scalar);
 959     /* Setup switch parameters */
 960     d_scalar         = rcutoff_scalar-rswitch_scalar;
 961     d                = _mm256_set1_ps(d_scalar);
 962     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 963     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 964     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 965     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 966     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 967     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 968
 969     /* Avoid stupid compiler warnings */
 970     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 971     j_coord_offsetA = 0;
 972     j_coord_offsetB = 0;
 973     j_coord_offsetC = 0;
 974     j_coord_offsetD = 0;
 975     j_coord_offsetE = 0;
 976     j_coord_offsetF = 0;
 977     j_coord_offsetG = 0;
 978     j_coord_offsetH = 0;
 979
 980     outeriter        = 0;
 981     inneriter        = 0;
 982
 983     for(iidx=0;iidx<4*DIM;iidx++)
 984     {
 985         scratch[iidx] = 0.0;
 986     }
 987
 988     /* Start outer loop over neighborlists */
 989     for(iidx=0; iidx<nri; iidx++)
 990     {
 991         /* Load shift vector for this list */
 992         i_shift_offset   = DIM*shiftidx[iidx];
 993
 994         /* Load limits for loop over neighbors */
 995         j_index_start    = jindex[iidx];
 996         j_index_end      = jindex[iidx+1];
 997
 998         /* Get outer coordinate index */
 999         inr              = iinr[iidx];
1000         i_coord_offset   = DIM*inr;
1001
1002         /* Load i particle coords and add shift vector */
1003         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1004                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1005
1006         fix0             = _mm256_setzero_ps();
1007         fiy0             = _mm256_setzero_ps();
1008         fiz0             = _mm256_setzero_ps();
1009         fix1             = _mm256_setzero_ps();
1010         fiy1             = _mm256_setzero_ps();
1011         fiz1             = _mm256_setzero_ps();
1012         fix2             = _mm256_setzero_ps();
1013         fiy2             = _mm256_setzero_ps();
1014         fiz2             = _mm256_setzero_ps();
1015
1016         /* Start inner kernel loop */
1017         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1018         {
1019
1020             /* Get j neighbor index, and coordinate index */
1021             jnrA             = jjnr[jidx];
1022             jnrB             = jjnr[jidx+1];
1023             jnrC             = jjnr[jidx+2];
1024             jnrD             = jjnr[jidx+3];
1025             jnrE             = jjnr[jidx+4];
1026             jnrF             = jjnr[jidx+5];
1027             jnrG             = jjnr[jidx+6];
1028             jnrH             = jjnr[jidx+7];
1029             j_coord_offsetA  = DIM*jnrA;
1030             j_coord_offsetB  = DIM*jnrB;
1031             j_coord_offsetC  = DIM*jnrC;
1032             j_coord_offsetD  = DIM*jnrD;
1033             j_coord_offsetE  = DIM*jnrE;
1034             j_coord_offsetF  = DIM*jnrF;
1035             j_coord_offsetG  = DIM*jnrG;
1036             j_coord_offsetH  = DIM*jnrH;
1037
1038             /* load j atom coordinates */
1039             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1040                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1041                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1042                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1043                                                  &jx0,&jy0,&jz0);
1044
1045             /* Calculate displacement vector */
1046             dx00             = _mm256_sub_ps(ix0,jx0);
1047             dy00             = _mm256_sub_ps(iy0,jy0);
1048             dz00             = _mm256_sub_ps(iz0,jz0);
1049             dx10             = _mm256_sub_ps(ix1,jx0);
1050             dy10             = _mm256_sub_ps(iy1,jy0);
1051             dz10             = _mm256_sub_ps(iz1,jz0);
1052             dx20             = _mm256_sub_ps(ix2,jx0);
1053             dy20             = _mm256_sub_ps(iy2,jy0);
1054             dz20             = _mm256_sub_ps(iz2,jz0);
1055
1056             /* Calculate squared distance and things based on it */
1057             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1058             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1059             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1060
1061             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1062             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1063             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1064
1065             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1066             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1067             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1068
1069             /* Load parameters for j particles */
1070             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1071                                                                  charge+jnrC+0,charge+jnrD+0,
1072                                                                  charge+jnrE+0,charge+jnrF+0,
1073                                                                  charge+jnrG+0,charge+jnrH+0);
1074             vdwjidx0A        = 2*vdwtype[jnrA+0];
1075             vdwjidx0B        = 2*vdwtype[jnrB+0];
1076             vdwjidx0C        = 2*vdwtype[jnrC+0];
1077             vdwjidx0D        = 2*vdwtype[jnrD+0];
1078             vdwjidx0E        = 2*vdwtype[jnrE+0];
1079             vdwjidx0F        = 2*vdwtype[jnrF+0];
1080             vdwjidx0G        = 2*vdwtype[jnrG+0];
1081             vdwjidx0H        = 2*vdwtype[jnrH+0];
1082
1083             fjx0             = _mm256_setzero_ps();
1084             fjy0             = _mm256_setzero_ps();
1085             fjz0             = _mm256_setzero_ps();
1086
1087             /**************************
1088              * CALCULATE INTERACTIONS *
1089              **************************/
1090
1091             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1092             {
1093
1094             r00              = _mm256_mul_ps(rsq00,rinv00);
1095
1096             /* Compute parameters for interactions between i and j atoms */
1097             qq00             = _mm256_mul_ps(iq0,jq0);
1098             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1099                                             vdwioffsetptr0+vdwjidx0B,
1100                                             vdwioffsetptr0+vdwjidx0C,
1101                                             vdwioffsetptr0+vdwjidx0D,
1102                                             vdwioffsetptr0+vdwjidx0E,
1103                                             vdwioffsetptr0+vdwjidx0F,
1104                                             vdwioffsetptr0+vdwjidx0G,
1105                                             vdwioffsetptr0+vdwjidx0H,
1106                                             &c6_00,&c12_00);
1107
1108             /* EWALD ELECTROSTATICS */
1109
1110             /* Analytical PME correction */
1111             zeta2            = _mm256_mul_ps(beta2,rsq00);
1112             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1113             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1114             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1115             felec            = _mm256_mul_ps(qq00,felec);
1116             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1117             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1118             velec            = _mm256_sub_ps(rinv00,pmecorrV);
1119             velec            = _mm256_mul_ps(qq00,velec);
1120
1121             /* LENNARD-JONES DISPERSION/REPULSION */
1122
1123             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1124             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
1125             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
1126             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
1127             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
1128
1129             d                = _mm256_sub_ps(r00,rswitch);
1130             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1131             d2               = _mm256_mul_ps(d,d);
1132             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)))))));
1133
1134             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1135
1136             /* Evaluate switch function */
1137             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1138             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
1139             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
1140             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1141
1142             fscal            = _mm256_add_ps(felec,fvdw);
1143
1144             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1145
1146             /* Calculate temporary vectorial force */
1147             tx               = _mm256_mul_ps(fscal,dx00);
1148             ty               = _mm256_mul_ps(fscal,dy00);
1149             tz               = _mm256_mul_ps(fscal,dz00);
1150
1151             /* Update vectorial force */
1152             fix0             = _mm256_add_ps(fix0,tx);
1153             fiy0             = _mm256_add_ps(fiy0,ty);
1154             fiz0             = _mm256_add_ps(fiz0,tz);
1155
1156             fjx0             = _mm256_add_ps(fjx0,tx);
1157             fjy0             = _mm256_add_ps(fjy0,ty);
1158             fjz0             = _mm256_add_ps(fjz0,tz);
1159
1160             }
1161
1162             /**************************
1163              * CALCULATE INTERACTIONS *
1164              **************************/
1165
1166             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1167             {
1168
1169             r10              = _mm256_mul_ps(rsq10,rinv10);
1170
1171             /* Compute parameters for interactions between i and j atoms */
1172             qq10             = _mm256_mul_ps(iq1,jq0);
1173
1174             /* EWALD ELECTROSTATICS */
1175
1176             /* Analytical PME correction */
1177             zeta2            = _mm256_mul_ps(beta2,rsq10);
1178             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1179             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1180             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1181             felec            = _mm256_mul_ps(qq10,felec);
1182             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1183             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1184             velec            = _mm256_sub_ps(rinv10,pmecorrV);
1185             velec            = _mm256_mul_ps(qq10,velec);
1186
1187             d                = _mm256_sub_ps(r10,rswitch);
1188             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1189             d2               = _mm256_mul_ps(d,d);
1190             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)))))));
1191
1192             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1193
1194             /* Evaluate switch function */
1195             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1196             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv10,_mm256_mul_ps(velec,dsw)) );
1197             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1198
1199             fscal            = felec;
1200
1201             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1202
1203             /* Calculate temporary vectorial force */
1204             tx               = _mm256_mul_ps(fscal,dx10);
1205             ty               = _mm256_mul_ps(fscal,dy10);
1206             tz               = _mm256_mul_ps(fscal,dz10);
1207
1208             /* Update vectorial force */
1209             fix1             = _mm256_add_ps(fix1,tx);
1210             fiy1             = _mm256_add_ps(fiy1,ty);
1211             fiz1             = _mm256_add_ps(fiz1,tz);
1212
1213             fjx0             = _mm256_add_ps(fjx0,tx);
1214             fjy0             = _mm256_add_ps(fjy0,ty);
1215             fjz0             = _mm256_add_ps(fjz0,tz);
1216
1217             }
1218
1219             /**************************
1220              * CALCULATE INTERACTIONS *
1221              **************************/
1222
1223             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1224             {
1225
1226             r20              = _mm256_mul_ps(rsq20,rinv20);
1227
1228             /* Compute parameters for interactions between i and j atoms */
1229             qq20             = _mm256_mul_ps(iq2,jq0);
1230
1231             /* EWALD ELECTROSTATICS */
1232
1233             /* Analytical PME correction */
1234             zeta2            = _mm256_mul_ps(beta2,rsq20);
1235             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1236             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1237             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1238             felec            = _mm256_mul_ps(qq20,felec);
1239             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1240             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1241             velec            = _mm256_sub_ps(rinv20,pmecorrV);
1242             velec            = _mm256_mul_ps(qq20,velec);
1243
1244             d                = _mm256_sub_ps(r20,rswitch);
1245             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1246             d2               = _mm256_mul_ps(d,d);
1247             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)))))));
1248
1249             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1250
1251             /* Evaluate switch function */
1252             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1253             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv20,_mm256_mul_ps(velec,dsw)) );
1254             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1255
1256             fscal            = felec;
1257
1258             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1259
1260             /* Calculate temporary vectorial force */
1261             tx               = _mm256_mul_ps(fscal,dx20);
1262             ty               = _mm256_mul_ps(fscal,dy20);
1263             tz               = _mm256_mul_ps(fscal,dz20);
1264
1265             /* Update vectorial force */
1266             fix2             = _mm256_add_ps(fix2,tx);
1267             fiy2             = _mm256_add_ps(fiy2,ty);
1268             fiz2             = _mm256_add_ps(fiz2,tz);
1269
1270             fjx0             = _mm256_add_ps(fjx0,tx);
1271             fjy0             = _mm256_add_ps(fjy0,ty);
1272             fjz0             = _mm256_add_ps(fjz0,tz);
1273
1274             }
1275
1276             fjptrA             = f+j_coord_offsetA;
1277             fjptrB             = f+j_coord_offsetB;
1278             fjptrC             = f+j_coord_offsetC;
1279             fjptrD             = f+j_coord_offsetD;
1280             fjptrE             = f+j_coord_offsetE;
1281             fjptrF             = f+j_coord_offsetF;
1282             fjptrG             = f+j_coord_offsetG;
1283             fjptrH             = f+j_coord_offsetH;
1284
1285             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1286
1287             /* Inner loop uses 333 flops */
1288         }
1289
1290         if(jidx<j_index_end)
1291         {
1292
1293             /* Get j neighbor index, and coordinate index */
1294             jnrlistA         = jjnr[jidx];
1295             jnrlistB         = jjnr[jidx+1];
1296             jnrlistC         = jjnr[jidx+2];
1297             jnrlistD         = jjnr[jidx+3];
1298             jnrlistE         = jjnr[jidx+4];
1299             jnrlistF         = jjnr[jidx+5];
1300             jnrlistG         = jjnr[jidx+6];
1301             jnrlistH         = jjnr[jidx+7];
1302             /* Sign of each element will be negative for non-real atoms.
1303              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1304              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1305              */
1306             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1307                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1308
1309             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1310             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1311             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1312             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1313             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1314             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1315             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1316             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1317             j_coord_offsetA  = DIM*jnrA;
1318             j_coord_offsetB  = DIM*jnrB;
1319             j_coord_offsetC  = DIM*jnrC;
1320             j_coord_offsetD  = DIM*jnrD;
1321             j_coord_offsetE  = DIM*jnrE;
1322             j_coord_offsetF  = DIM*jnrF;
1323             j_coord_offsetG  = DIM*jnrG;
1324             j_coord_offsetH  = DIM*jnrH;
1325
1326             /* load j atom coordinates */
1327             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1328                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1329                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1330                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1331                                                  &jx0,&jy0,&jz0);
1332
1333             /* Calculate displacement vector */
1334             dx00             = _mm256_sub_ps(ix0,jx0);
1335             dy00             = _mm256_sub_ps(iy0,jy0);
1336             dz00             = _mm256_sub_ps(iz0,jz0);
1337             dx10             = _mm256_sub_ps(ix1,jx0);
1338             dy10             = _mm256_sub_ps(iy1,jy0);
1339             dz10             = _mm256_sub_ps(iz1,jz0);
1340             dx20             = _mm256_sub_ps(ix2,jx0);
1341             dy20             = _mm256_sub_ps(iy2,jy0);
1342             dz20             = _mm256_sub_ps(iz2,jz0);
1343
1344             /* Calculate squared distance and things based on it */
1345             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1346             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1347             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1348
1349             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1350             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1351             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1352
1353             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1354             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1355             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1356
1357             /* Load parameters for j particles */
1358             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1359                                                                  charge+jnrC+0,charge+jnrD+0,
1360                                                                  charge+jnrE+0,charge+jnrF+0,
1361                                                                  charge+jnrG+0,charge+jnrH+0);
1362             vdwjidx0A        = 2*vdwtype[jnrA+0];
1363             vdwjidx0B        = 2*vdwtype[jnrB+0];
1364             vdwjidx0C        = 2*vdwtype[jnrC+0];
1365             vdwjidx0D        = 2*vdwtype[jnrD+0];
1366             vdwjidx0E        = 2*vdwtype[jnrE+0];
1367             vdwjidx0F        = 2*vdwtype[jnrF+0];
1368             vdwjidx0G        = 2*vdwtype[jnrG+0];
1369             vdwjidx0H        = 2*vdwtype[jnrH+0];
1370
1371             fjx0             = _mm256_setzero_ps();
1372             fjy0             = _mm256_setzero_ps();
1373             fjz0             = _mm256_setzero_ps();
1374
1375             /**************************
1376              * CALCULATE INTERACTIONS *
1377              **************************/
1378
1379             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1380             {
1381
1382             r00              = _mm256_mul_ps(rsq00,rinv00);
1383             r00              = _mm256_andnot_ps(dummy_mask,r00);
1384
1385             /* Compute parameters for interactions between i and j atoms */
1386             qq00             = _mm256_mul_ps(iq0,jq0);
1387             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1388                                             vdwioffsetptr0+vdwjidx0B,
1389                                             vdwioffsetptr0+vdwjidx0C,
1390                                             vdwioffsetptr0+vdwjidx0D,
1391                                             vdwioffsetptr0+vdwjidx0E,
1392                                             vdwioffsetptr0+vdwjidx0F,
1393                                             vdwioffsetptr0+vdwjidx0G,
1394                                             vdwioffsetptr0+vdwjidx0H,
1395                                             &c6_00,&c12_00);
1396
1397             /* EWALD ELECTROSTATICS */
1398
1399             /* Analytical PME correction */
1400             zeta2            = _mm256_mul_ps(beta2,rsq00);
1401             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1402             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1403             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1404             felec            = _mm256_mul_ps(qq00,felec);
1405             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1406             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1407             velec            = _mm256_sub_ps(rinv00,pmecorrV);
1408             velec            = _mm256_mul_ps(qq00,velec);
1409
1410             /* LENNARD-JONES DISPERSION/REPULSION */
1411
1412             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1413             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
1414             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
1415             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
1416             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
1417
1418             d                = _mm256_sub_ps(r00,rswitch);
1419             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1420             d2               = _mm256_mul_ps(d,d);
1421             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)))))));
1422
1423             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1424
1425             /* Evaluate switch function */
1426             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1427             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
1428             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
1429             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1430
1431             fscal            = _mm256_add_ps(felec,fvdw);
1432
1433             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1434
1435             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1436
1437             /* Calculate temporary vectorial force */
1438             tx               = _mm256_mul_ps(fscal,dx00);
1439             ty               = _mm256_mul_ps(fscal,dy00);
1440             tz               = _mm256_mul_ps(fscal,dz00);
1441
1442             /* Update vectorial force */
1443             fix0             = _mm256_add_ps(fix0,tx);
1444             fiy0             = _mm256_add_ps(fiy0,ty);
1445             fiz0             = _mm256_add_ps(fiz0,tz);
1446
1447             fjx0             = _mm256_add_ps(fjx0,tx);
1448             fjy0             = _mm256_add_ps(fjy0,ty);
1449             fjz0             = _mm256_add_ps(fjz0,tz);
1450
1451             }
1452
1453             /**************************
1454              * CALCULATE INTERACTIONS *
1455              **************************/
1456
1457             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1458             {
1459
1460             r10              = _mm256_mul_ps(rsq10,rinv10);
1461             r10              = _mm256_andnot_ps(dummy_mask,r10);
1462
1463             /* Compute parameters for interactions between i and j atoms */
1464             qq10             = _mm256_mul_ps(iq1,jq0);
1465
1466             /* EWALD ELECTROSTATICS */
1467
1468             /* Analytical PME correction */
1469             zeta2            = _mm256_mul_ps(beta2,rsq10);
1470             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1471             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1472             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1473             felec            = _mm256_mul_ps(qq10,felec);
1474             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1475             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1476             velec            = _mm256_sub_ps(rinv10,pmecorrV);
1477             velec            = _mm256_mul_ps(qq10,velec);
1478
1479             d                = _mm256_sub_ps(r10,rswitch);
1480             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1481             d2               = _mm256_mul_ps(d,d);
1482             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)))))));
1483
1484             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1485
1486             /* Evaluate switch function */
1487             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1488             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv10,_mm256_mul_ps(velec,dsw)) );
1489             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1490
1491             fscal            = felec;
1492
1493             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1494
1495             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1496
1497             /* Calculate temporary vectorial force */
1498             tx               = _mm256_mul_ps(fscal,dx10);
1499             ty               = _mm256_mul_ps(fscal,dy10);
1500             tz               = _mm256_mul_ps(fscal,dz10);
1501
1502             /* Update vectorial force */
1503             fix1             = _mm256_add_ps(fix1,tx);
1504             fiy1             = _mm256_add_ps(fiy1,ty);
1505             fiz1             = _mm256_add_ps(fiz1,tz);
1506
1507             fjx0             = _mm256_add_ps(fjx0,tx);
1508             fjy0             = _mm256_add_ps(fjy0,ty);
1509             fjz0             = _mm256_add_ps(fjz0,tz);
1510
1511             }
1512
1513             /**************************
1514              * CALCULATE INTERACTIONS *
1515              **************************/
1516
1517             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1518             {
1519
1520             r20              = _mm256_mul_ps(rsq20,rinv20);
1521             r20              = _mm256_andnot_ps(dummy_mask,r20);
1522
1523             /* Compute parameters for interactions between i and j atoms */
1524             qq20             = _mm256_mul_ps(iq2,jq0);
1525
1526             /* EWALD ELECTROSTATICS */
1527
1528             /* Analytical PME correction */
1529             zeta2            = _mm256_mul_ps(beta2,rsq20);
1530             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1531             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1532             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1533             felec            = _mm256_mul_ps(qq20,felec);
1534             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1535             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1536             velec            = _mm256_sub_ps(rinv20,pmecorrV);
1537             velec            = _mm256_mul_ps(qq20,velec);
1538
1539             d                = _mm256_sub_ps(r20,rswitch);
1540             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1541             d2               = _mm256_mul_ps(d,d);
1542             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)))))));
1543
1544             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1545
1546             /* Evaluate switch function */
1547             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1548             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv20,_mm256_mul_ps(velec,dsw)) );
1549             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1550
1551             fscal            = felec;
1552
1553             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1554
1555             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1556
1557             /* Calculate temporary vectorial force */
1558             tx               = _mm256_mul_ps(fscal,dx20);
1559             ty               = _mm256_mul_ps(fscal,dy20);
1560             tz               = _mm256_mul_ps(fscal,dz20);
1561
1562             /* Update vectorial force */
1563             fix2             = _mm256_add_ps(fix2,tx);
1564             fiy2             = _mm256_add_ps(fiy2,ty);
1565             fiz2             = _mm256_add_ps(fiz2,tz);
1566
1567             fjx0             = _mm256_add_ps(fjx0,tx);
1568             fjy0             = _mm256_add_ps(fjy0,ty);
1569             fjz0             = _mm256_add_ps(fjz0,tz);
1570
1571             }
1572
1573             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1574             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1575             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1576             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1577             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1578             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1579             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1580             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1581
1582             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1583
1584             /* Inner loop uses 336 flops */
1585         }
1586
1587         /* End of innermost loop */
1588
1589         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1590                                                  f+i_coord_offset,fshift+i_shift_offset);
1591
1592         /* Increment number of inner iterations */
1593         inneriter                  += j_index_end - j_index_start;
1594
1595         /* Outer loop uses 18 flops */
1596     }
1597
1598     /* Increment number of outer iterations */
1599     outeriter        += nri;
1600
1601     /* Update outer/inner flops */
1602
1603     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*336);
1604 }