src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_avx_256_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_avx_256_double
  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 refer to j loop unrolling done with AVX, e.g. for the four 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              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  93     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  96     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  97     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 104     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 105     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 106     real             rswitch_scalar,d_scalar;
 107     __m256d          dummy_mask,cutoff_mask;
 108     __m128           tmpmask0,tmpmask1;
 109     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 110     __m256d          one     = _mm256_set1_pd(1.0);
 111     __m256d          two     = _mm256_set1_pd(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_pd(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125     krf              = _mm256_set1_pd(fr->ic->k_rf);
 126     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 127     crf              = _mm256_set1_pd(fr->ic->c_rf);
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131
 132     /* Setup water-specific parameters */
 133     inr              = nlist->iinr[0];
 134     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 135     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 136     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 137     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 138
 139     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 140     rcutoff_scalar   = fr->rcoulomb;
 141     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 142     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 143
 144     rswitch_scalar   = fr->rvdw_switch;
 145     rswitch          = _mm256_set1_pd(rswitch_scalar);
 146     /* Setup switch parameters */
 147     d_scalar         = rcutoff_scalar-rswitch_scalar;
 148     d                = _mm256_set1_pd(d_scalar);
 149     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 150     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 151     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 152     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 153     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 154     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 155
 156     /* Avoid stupid compiler warnings */
 157     jnrA = jnrB = jnrC = jnrD = 0;
 158     j_coord_offsetA = 0;
 159     j_coord_offsetB = 0;
 160     j_coord_offsetC = 0;
 161     j_coord_offsetD = 0;
 162
 163     outeriter        = 0;
 164     inneriter        = 0;
 165
 166     for(iidx=0;iidx<4*DIM;iidx++)
 167     {
 168         scratch[iidx] = 0.0;
 169     }
 170
 171     /* Start outer loop over neighborlists */
 172     for(iidx=0; iidx<nri; iidx++)
 173     {
 174         /* Load shift vector for this list */
 175         i_shift_offset   = DIM*shiftidx[iidx];
 176
 177         /* Load limits for loop over neighbors */
 178         j_index_start    = jindex[iidx];
 179         j_index_end      = jindex[iidx+1];
 180
 181         /* Get outer coordinate index */
 182         inr              = iinr[iidx];
 183         i_coord_offset   = DIM*inr;
 184
 185         /* Load i particle coords and add shift vector */
 186         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 187                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 188
 189         fix0             = _mm256_setzero_pd();
 190         fiy0             = _mm256_setzero_pd();
 191         fiz0             = _mm256_setzero_pd();
 192         fix1             = _mm256_setzero_pd();
 193         fiy1             = _mm256_setzero_pd();
 194         fiz1             = _mm256_setzero_pd();
 195         fix2             = _mm256_setzero_pd();
 196         fiy2             = _mm256_setzero_pd();
 197         fiz2             = _mm256_setzero_pd();
 198
 199         /* Reset potential sums */
 200         velecsum         = _mm256_setzero_pd();
 201         vvdwsum          = _mm256_setzero_pd();
 202
 203         /* Start inner kernel loop */
 204         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 205         {
 206
 207             /* Get j neighbor index, and coordinate index */
 208             jnrA             = jjnr[jidx];
 209             jnrB             = jjnr[jidx+1];
 210             jnrC             = jjnr[jidx+2];
 211             jnrD             = jjnr[jidx+3];
 212             j_coord_offsetA  = DIM*jnrA;
 213             j_coord_offsetB  = DIM*jnrB;
 214             j_coord_offsetC  = DIM*jnrC;
 215             j_coord_offsetD  = DIM*jnrD;
 216
 217             /* load j atom coordinates */
 218             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 219                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 220                                                  &jx0,&jy0,&jz0);
 221
 222             /* Calculate displacement vector */
 223             dx00             = _mm256_sub_pd(ix0,jx0);
 224             dy00             = _mm256_sub_pd(iy0,jy0);
 225             dz00             = _mm256_sub_pd(iz0,jz0);
 226             dx10             = _mm256_sub_pd(ix1,jx0);
 227             dy10             = _mm256_sub_pd(iy1,jy0);
 228             dz10             = _mm256_sub_pd(iz1,jz0);
 229             dx20             = _mm256_sub_pd(ix2,jx0);
 230             dy20             = _mm256_sub_pd(iy2,jy0);
 231             dz20             = _mm256_sub_pd(iz2,jz0);
 232
 233             /* Calculate squared distance and things based on it */
 234             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 235             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 236             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 237
 238             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 239             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 240             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 241
 242             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 243             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 244             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 245
 246             /* Load parameters for j particles */
 247             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 248                                                                  charge+jnrC+0,charge+jnrD+0);
 249             vdwjidx0A        = 2*vdwtype[jnrA+0];
 250             vdwjidx0B        = 2*vdwtype[jnrB+0];
 251             vdwjidx0C        = 2*vdwtype[jnrC+0];
 252             vdwjidx0D        = 2*vdwtype[jnrD+0];
 253
 254             fjx0             = _mm256_setzero_pd();
 255             fjy0             = _mm256_setzero_pd();
 256             fjz0             = _mm256_setzero_pd();
 257
 258             /**************************
 259              * CALCULATE INTERACTIONS *
 260              **************************/
 261
 262             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 263             {
 264
 265             r00              = _mm256_mul_pd(rsq00,rinv00);
 266
 267             /* Compute parameters for interactions between i and j atoms */
 268             qq00             = _mm256_mul_pd(iq0,jq0);
 269             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 270                                             vdwioffsetptr0+vdwjidx0B,
 271                                             vdwioffsetptr0+vdwjidx0C,
 272                                             vdwioffsetptr0+vdwjidx0D,
 273                                             &c6_00,&c12_00);
 274
 275             /* REACTION-FIELD ELECTROSTATICS */
 276             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 277             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 278
 279             /* LENNARD-JONES DISPERSION/REPULSION */
 280
 281             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 282             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 283             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 284             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 285             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 286
 287             d                = _mm256_sub_pd(r00,rswitch);
 288             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 289             d2               = _mm256_mul_pd(d,d);
 290             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 291
 292             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 293
 294             /* Evaluate switch function */
 295             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 296             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
 297             vvdw             = _mm256_mul_pd(vvdw,sw);
 298             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 299
 300             /* Update potential sum for this i atom from the interaction with this j atom. */
 301             velec            = _mm256_and_pd(velec,cutoff_mask);
 302             velecsum         = _mm256_add_pd(velecsum,velec);
 303             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 304             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 305
 306             fscal            = _mm256_add_pd(felec,fvdw);
 307
 308             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 309
 310             /* Calculate temporary vectorial force */
 311             tx               = _mm256_mul_pd(fscal,dx00);
 312             ty               = _mm256_mul_pd(fscal,dy00);
 313             tz               = _mm256_mul_pd(fscal,dz00);
 314
 315             /* Update vectorial force */
 316             fix0             = _mm256_add_pd(fix0,tx);
 317             fiy0             = _mm256_add_pd(fiy0,ty);
 318             fiz0             = _mm256_add_pd(fiz0,tz);
 319
 320             fjx0             = _mm256_add_pd(fjx0,tx);
 321             fjy0             = _mm256_add_pd(fjy0,ty);
 322             fjz0             = _mm256_add_pd(fjz0,tz);
 323
 324             }
 325
 326             /**************************
 327              * CALCULATE INTERACTIONS *
 328              **************************/
 329
 330             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 331             {
 332
 333             /* Compute parameters for interactions between i and j atoms */
 334             qq10             = _mm256_mul_pd(iq1,jq0);
 335
 336             /* REACTION-FIELD ELECTROSTATICS */
 337             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 338             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 339
 340             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 341
 342             /* Update potential sum for this i atom from the interaction with this j atom. */
 343             velec            = _mm256_and_pd(velec,cutoff_mask);
 344             velecsum         = _mm256_add_pd(velecsum,velec);
 345
 346             fscal            = felec;
 347
 348             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 349
 350             /* Calculate temporary vectorial force */
 351             tx               = _mm256_mul_pd(fscal,dx10);
 352             ty               = _mm256_mul_pd(fscal,dy10);
 353             tz               = _mm256_mul_pd(fscal,dz10);
 354
 355             /* Update vectorial force */
 356             fix1             = _mm256_add_pd(fix1,tx);
 357             fiy1             = _mm256_add_pd(fiy1,ty);
 358             fiz1             = _mm256_add_pd(fiz1,tz);
 359
 360             fjx0             = _mm256_add_pd(fjx0,tx);
 361             fjy0             = _mm256_add_pd(fjy0,ty);
 362             fjz0             = _mm256_add_pd(fjz0,tz);
 363
 364             }
 365
 366             /**************************
 367              * CALCULATE INTERACTIONS *
 368              **************************/
 369
 370             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 371             {
 372
 373             /* Compute parameters for interactions between i and j atoms */
 374             qq20             = _mm256_mul_pd(iq2,jq0);
 375
 376             /* REACTION-FIELD ELECTROSTATICS */
 377             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 378             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 379
 380             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 381
 382             /* Update potential sum for this i atom from the interaction with this j atom. */
 383             velec            = _mm256_and_pd(velec,cutoff_mask);
 384             velecsum         = _mm256_add_pd(velecsum,velec);
 385
 386             fscal            = felec;
 387
 388             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 389
 390             /* Calculate temporary vectorial force */
 391             tx               = _mm256_mul_pd(fscal,dx20);
 392             ty               = _mm256_mul_pd(fscal,dy20);
 393             tz               = _mm256_mul_pd(fscal,dz20);
 394
 395             /* Update vectorial force */
 396             fix2             = _mm256_add_pd(fix2,tx);
 397             fiy2             = _mm256_add_pd(fiy2,ty);
 398             fiz2             = _mm256_add_pd(fiz2,tz);
 399
 400             fjx0             = _mm256_add_pd(fjx0,tx);
 401             fjy0             = _mm256_add_pd(fjy0,ty);
 402             fjz0             = _mm256_add_pd(fjz0,tz);
 403
 404             }
 405
 406             fjptrA             = f+j_coord_offsetA;
 407             fjptrB             = f+j_coord_offsetB;
 408             fjptrC             = f+j_coord_offsetC;
 409             fjptrD             = f+j_coord_offsetD;
 410
 411             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 412
 413             /* Inner loop uses 145 flops */
 414         }
 415
 416         if(jidx<j_index_end)
 417         {
 418
 419             /* Get j neighbor index, and coordinate index */
 420             jnrlistA         = jjnr[jidx];
 421             jnrlistB         = jjnr[jidx+1];
 422             jnrlistC         = jjnr[jidx+2];
 423             jnrlistD         = jjnr[jidx+3];
 424             /* Sign of each element will be negative for non-real atoms.
 425              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 426              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 427              */
 428             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 429
 430             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 431             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 432             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 433
 434             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 435             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 436             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 437             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 438             j_coord_offsetA  = DIM*jnrA;
 439             j_coord_offsetB  = DIM*jnrB;
 440             j_coord_offsetC  = DIM*jnrC;
 441             j_coord_offsetD  = DIM*jnrD;
 442
 443             /* load j atom coordinates */
 444             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 445                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 446                                                  &jx0,&jy0,&jz0);
 447
 448             /* Calculate displacement vector */
 449             dx00             = _mm256_sub_pd(ix0,jx0);
 450             dy00             = _mm256_sub_pd(iy0,jy0);
 451             dz00             = _mm256_sub_pd(iz0,jz0);
 452             dx10             = _mm256_sub_pd(ix1,jx0);
 453             dy10             = _mm256_sub_pd(iy1,jy0);
 454             dz10             = _mm256_sub_pd(iz1,jz0);
 455             dx20             = _mm256_sub_pd(ix2,jx0);
 456             dy20             = _mm256_sub_pd(iy2,jy0);
 457             dz20             = _mm256_sub_pd(iz2,jz0);
 458
 459             /* Calculate squared distance and things based on it */
 460             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 461             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 462             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 463
 464             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 465             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 466             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 467
 468             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 469             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 470             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 471
 472             /* Load parameters for j particles */
 473             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 474                                                                  charge+jnrC+0,charge+jnrD+0);
 475             vdwjidx0A        = 2*vdwtype[jnrA+0];
 476             vdwjidx0B        = 2*vdwtype[jnrB+0];
 477             vdwjidx0C        = 2*vdwtype[jnrC+0];
 478             vdwjidx0D        = 2*vdwtype[jnrD+0];
 479
 480             fjx0             = _mm256_setzero_pd();
 481             fjy0             = _mm256_setzero_pd();
 482             fjz0             = _mm256_setzero_pd();
 483
 484             /**************************
 485              * CALCULATE INTERACTIONS *
 486              **************************/
 487
 488             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 489             {
 490
 491             r00              = _mm256_mul_pd(rsq00,rinv00);
 492             r00              = _mm256_andnot_pd(dummy_mask,r00);
 493
 494             /* Compute parameters for interactions between i and j atoms */
 495             qq00             = _mm256_mul_pd(iq0,jq0);
 496             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 497                                             vdwioffsetptr0+vdwjidx0B,
 498                                             vdwioffsetptr0+vdwjidx0C,
 499                                             vdwioffsetptr0+vdwjidx0D,
 500                                             &c6_00,&c12_00);
 501
 502             /* REACTION-FIELD ELECTROSTATICS */
 503             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 504             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 505
 506             /* LENNARD-JONES DISPERSION/REPULSION */
 507
 508             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 509             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 510             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 511             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 512             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 513
 514             d                = _mm256_sub_pd(r00,rswitch);
 515             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 516             d2               = _mm256_mul_pd(d,d);
 517             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 518
 519             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 520
 521             /* Evaluate switch function */
 522             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 523             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
 524             vvdw             = _mm256_mul_pd(vvdw,sw);
 525             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 526
 527             /* Update potential sum for this i atom from the interaction with this j atom. */
 528             velec            = _mm256_and_pd(velec,cutoff_mask);
 529             velec            = _mm256_andnot_pd(dummy_mask,velec);
 530             velecsum         = _mm256_add_pd(velecsum,velec);
 531             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 532             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 533             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 534
 535             fscal            = _mm256_add_pd(felec,fvdw);
 536
 537             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 538
 539             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 540
 541             /* Calculate temporary vectorial force */
 542             tx               = _mm256_mul_pd(fscal,dx00);
 543             ty               = _mm256_mul_pd(fscal,dy00);
 544             tz               = _mm256_mul_pd(fscal,dz00);
 545
 546             /* Update vectorial force */
 547             fix0             = _mm256_add_pd(fix0,tx);
 548             fiy0             = _mm256_add_pd(fiy0,ty);
 549             fiz0             = _mm256_add_pd(fiz0,tz);
 550
 551             fjx0             = _mm256_add_pd(fjx0,tx);
 552             fjy0             = _mm256_add_pd(fjy0,ty);
 553             fjz0             = _mm256_add_pd(fjz0,tz);
 554
 555             }
 556
 557             /**************************
 558              * CALCULATE INTERACTIONS *
 559              **************************/
 560
 561             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 562             {
 563
 564             /* Compute parameters for interactions between i and j atoms */
 565             qq10             = _mm256_mul_pd(iq1,jq0);
 566
 567             /* REACTION-FIELD ELECTROSTATICS */
 568             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 569             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 570
 571             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 572
 573             /* Update potential sum for this i atom from the interaction with this j atom. */
 574             velec            = _mm256_and_pd(velec,cutoff_mask);
 575             velec            = _mm256_andnot_pd(dummy_mask,velec);
 576             velecsum         = _mm256_add_pd(velecsum,velec);
 577
 578             fscal            = felec;
 579
 580             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 581
 582             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 583
 584             /* Calculate temporary vectorial force */
 585             tx               = _mm256_mul_pd(fscal,dx10);
 586             ty               = _mm256_mul_pd(fscal,dy10);
 587             tz               = _mm256_mul_pd(fscal,dz10);
 588
 589             /* Update vectorial force */
 590             fix1             = _mm256_add_pd(fix1,tx);
 591             fiy1             = _mm256_add_pd(fiy1,ty);
 592             fiz1             = _mm256_add_pd(fiz1,tz);
 593
 594             fjx0             = _mm256_add_pd(fjx0,tx);
 595             fjy0             = _mm256_add_pd(fjy0,ty);
 596             fjz0             = _mm256_add_pd(fjz0,tz);
 597
 598             }
 599
 600             /**************************
 601              * CALCULATE INTERACTIONS *
 602              **************************/
 603
 604             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 605             {
 606
 607             /* Compute parameters for interactions between i and j atoms */
 608             qq20             = _mm256_mul_pd(iq2,jq0);
 609
 610             /* REACTION-FIELD ELECTROSTATICS */
 611             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 612             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 613
 614             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 615
 616             /* Update potential sum for this i atom from the interaction with this j atom. */
 617             velec            = _mm256_and_pd(velec,cutoff_mask);
 618             velec            = _mm256_andnot_pd(dummy_mask,velec);
 619             velecsum         = _mm256_add_pd(velecsum,velec);
 620
 621             fscal            = felec;
 622
 623             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 624
 625             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 626
 627             /* Calculate temporary vectorial force */
 628             tx               = _mm256_mul_pd(fscal,dx20);
 629             ty               = _mm256_mul_pd(fscal,dy20);
 630             tz               = _mm256_mul_pd(fscal,dz20);
 631
 632             /* Update vectorial force */
 633             fix2             = _mm256_add_pd(fix2,tx);
 634             fiy2             = _mm256_add_pd(fiy2,ty);
 635             fiz2             = _mm256_add_pd(fiz2,tz);
 636
 637             fjx0             = _mm256_add_pd(fjx0,tx);
 638             fjy0             = _mm256_add_pd(fjy0,ty);
 639             fjz0             = _mm256_add_pd(fjz0,tz);
 640
 641             }
 642
 643             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 644             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 645             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 646             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 647
 648             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 649
 650             /* Inner loop uses 146 flops */
 651         }
 652
 653         /* End of innermost loop */
 654
 655         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 656                                                  f+i_coord_offset,fshift+i_shift_offset);
 657
 658         ggid                        = gid[iidx];
 659         /* Update potential energies */
 660         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 661         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 662
 663         /* Increment number of inner iterations */
 664         inneriter                  += j_index_end - j_index_start;
 665
 666         /* Outer loop uses 20 flops */
 667     }
 668
 669     /* Increment number of outer iterations */
 670     outeriter        += nri;
 671
 672     /* Update outer/inner flops */
 673
 674     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*146);
 675 }
 676 /*
 677  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_avx_256_double
 678  * Electrostatics interaction: ReactionField
 679  * VdW interaction:            LennardJones
 680  * Geometry:                   Water3-Particle
 681  * Calculate force/pot:        Force
 682  */
 683 void
 684 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_avx_256_double
 685                     (t_nblist                    * gmx_restrict       nlist,
 686                      rvec                        * gmx_restrict          xx,
 687                      rvec                        * gmx_restrict          ff,
 688                      t_forcerec                  * gmx_restrict          fr,
 689                      t_mdatoms                   * gmx_restrict     mdatoms,
 690                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 691                      t_nrnb                      * gmx_restrict        nrnb)
 692 {
 693     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 694      * just 0 for non-waters.
 695      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 696      * jnr indices corresponding to data put in the four positions in the SIMD register.
 697      */
 698     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 699     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 700     int              jnrA,jnrB,jnrC,jnrD;
 701     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 702     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 703     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 704     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 705     real             rcutoff_scalar;
 706     real             *shiftvec,*fshift,*x,*f;
 707     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 708     real             scratch[4*DIM];
 709     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 710     real *           vdwioffsetptr0;
 711     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 712     real *           vdwioffsetptr1;
 713     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 714     real *           vdwioffsetptr2;
 715     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 716     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 717     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 718     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 719     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 720     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 721     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 722     real             *charge;
 723     int              nvdwtype;
 724     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 725     int              *vdwtype;
 726     real             *vdwparam;
 727     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 728     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 729     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 730     real             rswitch_scalar,d_scalar;
 731     __m256d          dummy_mask,cutoff_mask;
 732     __m128           tmpmask0,tmpmask1;
 733     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 734     __m256d          one     = _mm256_set1_pd(1.0);
 735     __m256d          two     = _mm256_set1_pd(2.0);
 736     x                = xx[0];
 737     f                = ff[0];
 738
 739     nri              = nlist->nri;
 740     iinr             = nlist->iinr;
 741     jindex           = nlist->jindex;
 742     jjnr             = nlist->jjnr;
 743     shiftidx         = nlist->shift;
 744     gid              = nlist->gid;
 745     shiftvec         = fr->shift_vec[0];
 746     fshift           = fr->fshift[0];
 747     facel            = _mm256_set1_pd(fr->epsfac);
 748     charge           = mdatoms->chargeA;
 749     krf              = _mm256_set1_pd(fr->ic->k_rf);
 750     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 751     crf              = _mm256_set1_pd(fr->ic->c_rf);
 752     nvdwtype         = fr->ntype;
 753     vdwparam         = fr->nbfp;
 754     vdwtype          = mdatoms->typeA;
 755
 756     /* Setup water-specific parameters */
 757     inr              = nlist->iinr[0];
 758     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 759     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 760     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 761     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 762
 763     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 764     rcutoff_scalar   = fr->rcoulomb;
 765     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 766     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 767
 768     rswitch_scalar   = fr->rvdw_switch;
 769     rswitch          = _mm256_set1_pd(rswitch_scalar);
 770     /* Setup switch parameters */
 771     d_scalar         = rcutoff_scalar-rswitch_scalar;
 772     d                = _mm256_set1_pd(d_scalar);
 773     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 774     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 775     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 776     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 777     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 778     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 779
 780     /* Avoid stupid compiler warnings */
 781     jnrA = jnrB = jnrC = jnrD = 0;
 782     j_coord_offsetA = 0;
 783     j_coord_offsetB = 0;
 784     j_coord_offsetC = 0;
 785     j_coord_offsetD = 0;
 786
 787     outeriter        = 0;
 788     inneriter        = 0;
 789
 790     for(iidx=0;iidx<4*DIM;iidx++)
 791     {
 792         scratch[iidx] = 0.0;
 793     }
 794
 795     /* Start outer loop over neighborlists */
 796     for(iidx=0; iidx<nri; iidx++)
 797     {
 798         /* Load shift vector for this list */
 799         i_shift_offset   = DIM*shiftidx[iidx];
 800
 801         /* Load limits for loop over neighbors */
 802         j_index_start    = jindex[iidx];
 803         j_index_end      = jindex[iidx+1];
 804
 805         /* Get outer coordinate index */
 806         inr              = iinr[iidx];
 807         i_coord_offset   = DIM*inr;
 808
 809         /* Load i particle coords and add shift vector */
 810         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 811                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 812
 813         fix0             = _mm256_setzero_pd();
 814         fiy0             = _mm256_setzero_pd();
 815         fiz0             = _mm256_setzero_pd();
 816         fix1             = _mm256_setzero_pd();
 817         fiy1             = _mm256_setzero_pd();
 818         fiz1             = _mm256_setzero_pd();
 819         fix2             = _mm256_setzero_pd();
 820         fiy2             = _mm256_setzero_pd();
 821         fiz2             = _mm256_setzero_pd();
 822
 823         /* Start inner kernel loop */
 824         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 825         {
 826
 827             /* Get j neighbor index, and coordinate index */
 828             jnrA             = jjnr[jidx];
 829             jnrB             = jjnr[jidx+1];
 830             jnrC             = jjnr[jidx+2];
 831             jnrD             = jjnr[jidx+3];
 832             j_coord_offsetA  = DIM*jnrA;
 833             j_coord_offsetB  = DIM*jnrB;
 834             j_coord_offsetC  = DIM*jnrC;
 835             j_coord_offsetD  = DIM*jnrD;
 836
 837             /* load j atom coordinates */
 838             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 839                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 840                                                  &jx0,&jy0,&jz0);
 841
 842             /* Calculate displacement vector */
 843             dx00             = _mm256_sub_pd(ix0,jx0);
 844             dy00             = _mm256_sub_pd(iy0,jy0);
 845             dz00             = _mm256_sub_pd(iz0,jz0);
 846             dx10             = _mm256_sub_pd(ix1,jx0);
 847             dy10             = _mm256_sub_pd(iy1,jy0);
 848             dz10             = _mm256_sub_pd(iz1,jz0);
 849             dx20             = _mm256_sub_pd(ix2,jx0);
 850             dy20             = _mm256_sub_pd(iy2,jy0);
 851             dz20             = _mm256_sub_pd(iz2,jz0);
 852
 853             /* Calculate squared distance and things based on it */
 854             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 855             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 856             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 857
 858             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 859             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 860             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 861
 862             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 863             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 864             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 865
 866             /* Load parameters for j particles */
 867             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 868                                                                  charge+jnrC+0,charge+jnrD+0);
 869             vdwjidx0A        = 2*vdwtype[jnrA+0];
 870             vdwjidx0B        = 2*vdwtype[jnrB+0];
 871             vdwjidx0C        = 2*vdwtype[jnrC+0];
 872             vdwjidx0D        = 2*vdwtype[jnrD+0];
 873
 874             fjx0             = _mm256_setzero_pd();
 875             fjy0             = _mm256_setzero_pd();
 876             fjz0             = _mm256_setzero_pd();
 877
 878             /**************************
 879              * CALCULATE INTERACTIONS *
 880              **************************/
 881
 882             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 883             {
 884
 885             r00              = _mm256_mul_pd(rsq00,rinv00);
 886
 887             /* Compute parameters for interactions between i and j atoms */
 888             qq00             = _mm256_mul_pd(iq0,jq0);
 889             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 890                                             vdwioffsetptr0+vdwjidx0B,
 891                                             vdwioffsetptr0+vdwjidx0C,
 892                                             vdwioffsetptr0+vdwjidx0D,
 893                                             &c6_00,&c12_00);
 894
 895             /* REACTION-FIELD ELECTROSTATICS */
 896             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 897
 898             /* LENNARD-JONES DISPERSION/REPULSION */
 899
 900             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 901             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 902             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 903             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 904             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 905
 906             d                = _mm256_sub_pd(r00,rswitch);
 907             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 908             d2               = _mm256_mul_pd(d,d);
 909             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 910
 911             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 912
 913             /* Evaluate switch function */
 914             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 915             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
 916             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 917
 918             fscal            = _mm256_add_pd(felec,fvdw);
 919
 920             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 921
 922             /* Calculate temporary vectorial force */
 923             tx               = _mm256_mul_pd(fscal,dx00);
 924             ty               = _mm256_mul_pd(fscal,dy00);
 925             tz               = _mm256_mul_pd(fscal,dz00);
 926
 927             /* Update vectorial force */
 928             fix0             = _mm256_add_pd(fix0,tx);
 929             fiy0             = _mm256_add_pd(fiy0,ty);
 930             fiz0             = _mm256_add_pd(fiz0,tz);
 931
 932             fjx0             = _mm256_add_pd(fjx0,tx);
 933             fjy0             = _mm256_add_pd(fjy0,ty);
 934             fjz0             = _mm256_add_pd(fjz0,tz);
 935
 936             }
 937
 938             /**************************
 939              * CALCULATE INTERACTIONS *
 940              **************************/
 941
 942             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 943             {
 944
 945             /* Compute parameters for interactions between i and j atoms */
 946             qq10             = _mm256_mul_pd(iq1,jq0);
 947
 948             /* REACTION-FIELD ELECTROSTATICS */
 949             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 950
 951             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 952
 953             fscal            = felec;
 954
 955             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 956
 957             /* Calculate temporary vectorial force */
 958             tx               = _mm256_mul_pd(fscal,dx10);
 959             ty               = _mm256_mul_pd(fscal,dy10);
 960             tz               = _mm256_mul_pd(fscal,dz10);
 961
 962             /* Update vectorial force */
 963             fix1             = _mm256_add_pd(fix1,tx);
 964             fiy1             = _mm256_add_pd(fiy1,ty);
 965             fiz1             = _mm256_add_pd(fiz1,tz);
 966
 967             fjx0             = _mm256_add_pd(fjx0,tx);
 968             fjy0             = _mm256_add_pd(fjy0,ty);
 969             fjz0             = _mm256_add_pd(fjz0,tz);
 970
 971             }
 972
 973             /**************************
 974              * CALCULATE INTERACTIONS *
 975              **************************/
 976
 977             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 978             {
 979
 980             /* Compute parameters for interactions between i and j atoms */
 981             qq20             = _mm256_mul_pd(iq2,jq0);
 982
 983             /* REACTION-FIELD ELECTROSTATICS */
 984             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 985
 986             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 987
 988             fscal            = felec;
 989
 990             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 991
 992             /* Calculate temporary vectorial force */
 993             tx               = _mm256_mul_pd(fscal,dx20);
 994             ty               = _mm256_mul_pd(fscal,dy20);
 995             tz               = _mm256_mul_pd(fscal,dz20);
 996
 997             /* Update vectorial force */
 998             fix2             = _mm256_add_pd(fix2,tx);
 999             fiy2             = _mm256_add_pd(fiy2,ty);
1000             fiz2             = _mm256_add_pd(fiz2,tz);
1001
1002             fjx0             = _mm256_add_pd(fjx0,tx);
1003             fjy0             = _mm256_add_pd(fjy0,ty);
1004             fjz0             = _mm256_add_pd(fjz0,tz);
1005
1006             }
1007
1008             fjptrA             = f+j_coord_offsetA;
1009             fjptrB             = f+j_coord_offsetB;
1010             fjptrC             = f+j_coord_offsetC;
1011             fjptrD             = f+j_coord_offsetD;
1012
1013             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1014
1015             /* Inner loop uses 124 flops */
1016         }
1017
1018         if(jidx<j_index_end)
1019         {
1020
1021             /* Get j neighbor index, and coordinate index */
1022             jnrlistA         = jjnr[jidx];
1023             jnrlistB         = jjnr[jidx+1];
1024             jnrlistC         = jjnr[jidx+2];
1025             jnrlistD         = jjnr[jidx+3];
1026             /* Sign of each element will be negative for non-real atoms.
1027              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1028              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1029              */
1030             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1031
1032             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1033             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1034             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1035
1036             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1037             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1038             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1039             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1040             j_coord_offsetA  = DIM*jnrA;
1041             j_coord_offsetB  = DIM*jnrB;
1042             j_coord_offsetC  = DIM*jnrC;
1043             j_coord_offsetD  = DIM*jnrD;
1044
1045             /* load j atom coordinates */
1046             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1047                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1048                                                  &jx0,&jy0,&jz0);
1049
1050             /* Calculate displacement vector */
1051             dx00             = _mm256_sub_pd(ix0,jx0);
1052             dy00             = _mm256_sub_pd(iy0,jy0);
1053             dz00             = _mm256_sub_pd(iz0,jz0);
1054             dx10             = _mm256_sub_pd(ix1,jx0);
1055             dy10             = _mm256_sub_pd(iy1,jy0);
1056             dz10             = _mm256_sub_pd(iz1,jz0);
1057             dx20             = _mm256_sub_pd(ix2,jx0);
1058             dy20             = _mm256_sub_pd(iy2,jy0);
1059             dz20             = _mm256_sub_pd(iz2,jz0);
1060
1061             /* Calculate squared distance and things based on it */
1062             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1063             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1064             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1065
1066             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1067             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1068             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1069
1070             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1071             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1072             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1073
1074             /* Load parameters for j particles */
1075             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1076                                                                  charge+jnrC+0,charge+jnrD+0);
1077             vdwjidx0A        = 2*vdwtype[jnrA+0];
1078             vdwjidx0B        = 2*vdwtype[jnrB+0];
1079             vdwjidx0C        = 2*vdwtype[jnrC+0];
1080             vdwjidx0D        = 2*vdwtype[jnrD+0];
1081
1082             fjx0             = _mm256_setzero_pd();
1083             fjy0             = _mm256_setzero_pd();
1084             fjz0             = _mm256_setzero_pd();
1085
1086             /**************************
1087              * CALCULATE INTERACTIONS *
1088              **************************/
1089
1090             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1091             {
1092
1093             r00              = _mm256_mul_pd(rsq00,rinv00);
1094             r00              = _mm256_andnot_pd(dummy_mask,r00);
1095
1096             /* Compute parameters for interactions between i and j atoms */
1097             qq00             = _mm256_mul_pd(iq0,jq0);
1098             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1099                                             vdwioffsetptr0+vdwjidx0B,
1100                                             vdwioffsetptr0+vdwjidx0C,
1101                                             vdwioffsetptr0+vdwjidx0D,
1102                                             &c6_00,&c12_00);
1103
1104             /* REACTION-FIELD ELECTROSTATICS */
1105             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
1106
1107             /* LENNARD-JONES DISPERSION/REPULSION */
1108
1109             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1110             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
1111             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
1112             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
1113             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
1114
1115             d                = _mm256_sub_pd(r00,rswitch);
1116             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1117             d2               = _mm256_mul_pd(d,d);
1118             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1119
1120             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1121
1122             /* Evaluate switch function */
1123             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1124             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
1125             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1126
1127             fscal            = _mm256_add_pd(felec,fvdw);
1128
1129             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1130
1131             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1132
1133             /* Calculate temporary vectorial force */
1134             tx               = _mm256_mul_pd(fscal,dx00);
1135             ty               = _mm256_mul_pd(fscal,dy00);
1136             tz               = _mm256_mul_pd(fscal,dz00);
1137
1138             /* Update vectorial force */
1139             fix0             = _mm256_add_pd(fix0,tx);
1140             fiy0             = _mm256_add_pd(fiy0,ty);
1141             fiz0             = _mm256_add_pd(fiz0,tz);
1142
1143             fjx0             = _mm256_add_pd(fjx0,tx);
1144             fjy0             = _mm256_add_pd(fjy0,ty);
1145             fjz0             = _mm256_add_pd(fjz0,tz);
1146
1147             }
1148
1149             /**************************
1150              * CALCULATE INTERACTIONS *
1151              **************************/
1152
1153             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1154             {
1155
1156             /* Compute parameters for interactions between i and j atoms */
1157             qq10             = _mm256_mul_pd(iq1,jq0);
1158
1159             /* REACTION-FIELD ELECTROSTATICS */
1160             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1161
1162             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1163
1164             fscal            = felec;
1165
1166             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1167
1168             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1169
1170             /* Calculate temporary vectorial force */
1171             tx               = _mm256_mul_pd(fscal,dx10);
1172             ty               = _mm256_mul_pd(fscal,dy10);
1173             tz               = _mm256_mul_pd(fscal,dz10);
1174
1175             /* Update vectorial force */
1176             fix1             = _mm256_add_pd(fix1,tx);
1177             fiy1             = _mm256_add_pd(fiy1,ty);
1178             fiz1             = _mm256_add_pd(fiz1,tz);
1179
1180             fjx0             = _mm256_add_pd(fjx0,tx);
1181             fjy0             = _mm256_add_pd(fjy0,ty);
1182             fjz0             = _mm256_add_pd(fjz0,tz);
1183
1184             }
1185
1186             /**************************
1187              * CALCULATE INTERACTIONS *
1188              **************************/
1189
1190             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1191             {
1192
1193             /* Compute parameters for interactions between i and j atoms */
1194             qq20             = _mm256_mul_pd(iq2,jq0);
1195
1196             /* REACTION-FIELD ELECTROSTATICS */
1197             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1198
1199             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1200
1201             fscal            = felec;
1202
1203             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1204
1205             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1206
1207             /* Calculate temporary vectorial force */
1208             tx               = _mm256_mul_pd(fscal,dx20);
1209             ty               = _mm256_mul_pd(fscal,dy20);
1210             tz               = _mm256_mul_pd(fscal,dz20);
1211
1212             /* Update vectorial force */
1213             fix2             = _mm256_add_pd(fix2,tx);
1214             fiy2             = _mm256_add_pd(fiy2,ty);
1215             fiz2             = _mm256_add_pd(fiz2,tz);
1216
1217             fjx0             = _mm256_add_pd(fjx0,tx);
1218             fjy0             = _mm256_add_pd(fjy0,ty);
1219             fjz0             = _mm256_add_pd(fjz0,tz);
1220
1221             }
1222
1223             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1224             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1225             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1226             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1227
1228             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1229
1230             /* Inner loop uses 125 flops */
1231         }
1232
1233         /* End of innermost loop */
1234
1235         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1236                                                  f+i_coord_offset,fshift+i_shift_offset);
1237
1238         /* Increment number of inner iterations */
1239         inneriter                  += j_index_end - j_index_start;
1240
1241         /* Outer loop uses 18 flops */
1242     }
1243
1244     /* Increment number of outer iterations */
1245     outeriter        += nri;
1246
1247     /* Update outer/inner flops */
1248
1249     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*125);
1250 }