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) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_double.h"
  48 #include "kernelutil_x86_avx_256_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_avx_256_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_avx_256_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  84     real *           vdwioffsetptr0;
  85     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  86     real *           vdwioffsetptr1;
  87     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  88     real *           vdwioffsetptr2;
  89     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  91     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  96     real             *charge;
  97     int              nvdwtype;
  98     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  99     int              *vdwtype;
 100     real             *vdwparam;
 101     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 102     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 103     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 104     real             rswitch_scalar,d_scalar;
 105     __m256d          dummy_mask,cutoff_mask;
 106     __m128           tmpmask0,tmpmask1;
 107     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 108     __m256d          one     = _mm256_set1_pd(1.0);
 109     __m256d          two     = _mm256_set1_pd(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm256_set1_pd(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     krf              = _mm256_set1_pd(fr->ic->k_rf);
 124     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 125     crf              = _mm256_set1_pd(fr->ic->c_rf);
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     /* Setup water-specific parameters */
 131     inr              = nlist->iinr[0];
 132     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 133     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 134     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 135     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 136
 137     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 138     rcutoff_scalar   = fr->rcoulomb;
 139     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 140     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 141
 142     rswitch_scalar   = fr->rvdw_switch;
 143     rswitch          = _mm256_set1_pd(rswitch_scalar);
 144     /* Setup switch parameters */
 145     d_scalar         = rcutoff_scalar-rswitch_scalar;
 146     d                = _mm256_set1_pd(d_scalar);
 147     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 148     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 149     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 150     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 151     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 152     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 153
 154     /* Avoid stupid compiler warnings */
 155     jnrA = jnrB = jnrC = jnrD = 0;
 156     j_coord_offsetA = 0;
 157     j_coord_offsetB = 0;
 158     j_coord_offsetC = 0;
 159     j_coord_offsetD = 0;
 160
 161     outeriter        = 0;
 162     inneriter        = 0;
 163
 164     for(iidx=0;iidx<4*DIM;iidx++)
 165     {
 166         scratch[iidx] = 0.0;
 167     }
 168
 169     /* Start outer loop over neighborlists */
 170     for(iidx=0; iidx<nri; iidx++)
 171     {
 172         /* Load shift vector for this list */
 173         i_shift_offset   = DIM*shiftidx[iidx];
 174
 175         /* Load limits for loop over neighbors */
 176         j_index_start    = jindex[iidx];
 177         j_index_end      = jindex[iidx+1];
 178
 179         /* Get outer coordinate index */
 180         inr              = iinr[iidx];
 181         i_coord_offset   = DIM*inr;
 182
 183         /* Load i particle coords and add shift vector */
 184         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 185                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 186
 187         fix0             = _mm256_setzero_pd();
 188         fiy0             = _mm256_setzero_pd();
 189         fiz0             = _mm256_setzero_pd();
 190         fix1             = _mm256_setzero_pd();
 191         fiy1             = _mm256_setzero_pd();
 192         fiz1             = _mm256_setzero_pd();
 193         fix2             = _mm256_setzero_pd();
 194         fiy2             = _mm256_setzero_pd();
 195         fiz2             = _mm256_setzero_pd();
 196
 197         /* Reset potential sums */
 198         velecsum         = _mm256_setzero_pd();
 199         vvdwsum          = _mm256_setzero_pd();
 200
 201         /* Start inner kernel loop */
 202         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 203         {
 204
 205             /* Get j neighbor index, and coordinate index */
 206             jnrA             = jjnr[jidx];
 207             jnrB             = jjnr[jidx+1];
 208             jnrC             = jjnr[jidx+2];
 209             jnrD             = jjnr[jidx+3];
 210             j_coord_offsetA  = DIM*jnrA;
 211             j_coord_offsetB  = DIM*jnrB;
 212             j_coord_offsetC  = DIM*jnrC;
 213             j_coord_offsetD  = DIM*jnrD;
 214
 215             /* load j atom coordinates */
 216             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 217                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 218                                                  &jx0,&jy0,&jz0);
 219
 220             /* Calculate displacement vector */
 221             dx00             = _mm256_sub_pd(ix0,jx0);
 222             dy00             = _mm256_sub_pd(iy0,jy0);
 223             dz00             = _mm256_sub_pd(iz0,jz0);
 224             dx10             = _mm256_sub_pd(ix1,jx0);
 225             dy10             = _mm256_sub_pd(iy1,jy0);
 226             dz10             = _mm256_sub_pd(iz1,jz0);
 227             dx20             = _mm256_sub_pd(ix2,jx0);
 228             dy20             = _mm256_sub_pd(iy2,jy0);
 229             dz20             = _mm256_sub_pd(iz2,jz0);
 230
 231             /* Calculate squared distance and things based on it */
 232             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 233             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 234             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 235
 236             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 237             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 238             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 239
 240             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 241             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 242             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 243
 244             /* Load parameters for j particles */
 245             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 246                                                                  charge+jnrC+0,charge+jnrD+0);
 247             vdwjidx0A        = 2*vdwtype[jnrA+0];
 248             vdwjidx0B        = 2*vdwtype[jnrB+0];
 249             vdwjidx0C        = 2*vdwtype[jnrC+0];
 250             vdwjidx0D        = 2*vdwtype[jnrD+0];
 251
 252             fjx0             = _mm256_setzero_pd();
 253             fjy0             = _mm256_setzero_pd();
 254             fjz0             = _mm256_setzero_pd();
 255
 256             /**************************
 257              * CALCULATE INTERACTIONS *
 258              **************************/
 259
 260             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 261             {
 262
 263             r00              = _mm256_mul_pd(rsq00,rinv00);
 264
 265             /* Compute parameters for interactions between i and j atoms */
 266             qq00             = _mm256_mul_pd(iq0,jq0);
 267             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 268                                             vdwioffsetptr0+vdwjidx0B,
 269                                             vdwioffsetptr0+vdwjidx0C,
 270                                             vdwioffsetptr0+vdwjidx0D,
 271                                             &c6_00,&c12_00);
 272
 273             /* REACTION-FIELD ELECTROSTATICS */
 274             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 275             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 276
 277             /* LENNARD-JONES DISPERSION/REPULSION */
 278
 279             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 280             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 281             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 282             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 283             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 284
 285             d                = _mm256_sub_pd(r00,rswitch);
 286             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 287             d2               = _mm256_mul_pd(d,d);
 288             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)))))));
 289
 290             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 291
 292             /* Evaluate switch function */
 293             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 294             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
 295             vvdw             = _mm256_mul_pd(vvdw,sw);
 296             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 297
 298             /* Update potential sum for this i atom from the interaction with this j atom. */
 299             velec            = _mm256_and_pd(velec,cutoff_mask);
 300             velecsum         = _mm256_add_pd(velecsum,velec);
 301             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 302             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 303
 304             fscal            = _mm256_add_pd(felec,fvdw);
 305
 306             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 307
 308             /* Calculate temporary vectorial force */
 309             tx               = _mm256_mul_pd(fscal,dx00);
 310             ty               = _mm256_mul_pd(fscal,dy00);
 311             tz               = _mm256_mul_pd(fscal,dz00);
 312
 313             /* Update vectorial force */
 314             fix0             = _mm256_add_pd(fix0,tx);
 315             fiy0             = _mm256_add_pd(fiy0,ty);
 316             fiz0             = _mm256_add_pd(fiz0,tz);
 317
 318             fjx0             = _mm256_add_pd(fjx0,tx);
 319             fjy0             = _mm256_add_pd(fjy0,ty);
 320             fjz0             = _mm256_add_pd(fjz0,tz);
 321
 322             }
 323
 324             /**************************
 325              * CALCULATE INTERACTIONS *
 326              **************************/
 327
 328             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 329             {
 330
 331             /* Compute parameters for interactions between i and j atoms */
 332             qq10             = _mm256_mul_pd(iq1,jq0);
 333
 334             /* REACTION-FIELD ELECTROSTATICS */
 335             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 336             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 337
 338             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 339
 340             /* Update potential sum for this i atom from the interaction with this j atom. */
 341             velec            = _mm256_and_pd(velec,cutoff_mask);
 342             velecsum         = _mm256_add_pd(velecsum,velec);
 343
 344             fscal            = felec;
 345
 346             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 347
 348             /* Calculate temporary vectorial force */
 349             tx               = _mm256_mul_pd(fscal,dx10);
 350             ty               = _mm256_mul_pd(fscal,dy10);
 351             tz               = _mm256_mul_pd(fscal,dz10);
 352
 353             /* Update vectorial force */
 354             fix1             = _mm256_add_pd(fix1,tx);
 355             fiy1             = _mm256_add_pd(fiy1,ty);
 356             fiz1             = _mm256_add_pd(fiz1,tz);
 357
 358             fjx0             = _mm256_add_pd(fjx0,tx);
 359             fjy0             = _mm256_add_pd(fjy0,ty);
 360             fjz0             = _mm256_add_pd(fjz0,tz);
 361
 362             }
 363
 364             /**************************
 365              * CALCULATE INTERACTIONS *
 366              **************************/
 367
 368             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 369             {
 370
 371             /* Compute parameters for interactions between i and j atoms */
 372             qq20             = _mm256_mul_pd(iq2,jq0);
 373
 374             /* REACTION-FIELD ELECTROSTATICS */
 375             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 376             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 377
 378             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 379
 380             /* Update potential sum for this i atom from the interaction with this j atom. */
 381             velec            = _mm256_and_pd(velec,cutoff_mask);
 382             velecsum         = _mm256_add_pd(velecsum,velec);
 383
 384             fscal            = felec;
 385
 386             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 387
 388             /* Calculate temporary vectorial force */
 389             tx               = _mm256_mul_pd(fscal,dx20);
 390             ty               = _mm256_mul_pd(fscal,dy20);
 391             tz               = _mm256_mul_pd(fscal,dz20);
 392
 393             /* Update vectorial force */
 394             fix2             = _mm256_add_pd(fix2,tx);
 395             fiy2             = _mm256_add_pd(fiy2,ty);
 396             fiz2             = _mm256_add_pd(fiz2,tz);
 397
 398             fjx0             = _mm256_add_pd(fjx0,tx);
 399             fjy0             = _mm256_add_pd(fjy0,ty);
 400             fjz0             = _mm256_add_pd(fjz0,tz);
 401
 402             }
 403
 404             fjptrA             = f+j_coord_offsetA;
 405             fjptrB             = f+j_coord_offsetB;
 406             fjptrC             = f+j_coord_offsetC;
 407             fjptrD             = f+j_coord_offsetD;
 408
 409             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 410
 411             /* Inner loop uses 145 flops */
 412         }
 413
 414         if(jidx<j_index_end)
 415         {
 416
 417             /* Get j neighbor index, and coordinate index */
 418             jnrlistA         = jjnr[jidx];
 419             jnrlistB         = jjnr[jidx+1];
 420             jnrlistC         = jjnr[jidx+2];
 421             jnrlistD         = jjnr[jidx+3];
 422             /* Sign of each element will be negative for non-real atoms.
 423              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 424              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 425              */
 426             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 427
 428             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 429             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 430             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 431
 432             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 433             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 434             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 435             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 436             j_coord_offsetA  = DIM*jnrA;
 437             j_coord_offsetB  = DIM*jnrB;
 438             j_coord_offsetC  = DIM*jnrC;
 439             j_coord_offsetD  = DIM*jnrD;
 440
 441             /* load j atom coordinates */
 442             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 443                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 444                                                  &jx0,&jy0,&jz0);
 445
 446             /* Calculate displacement vector */
 447             dx00             = _mm256_sub_pd(ix0,jx0);
 448             dy00             = _mm256_sub_pd(iy0,jy0);
 449             dz00             = _mm256_sub_pd(iz0,jz0);
 450             dx10             = _mm256_sub_pd(ix1,jx0);
 451             dy10             = _mm256_sub_pd(iy1,jy0);
 452             dz10             = _mm256_sub_pd(iz1,jz0);
 453             dx20             = _mm256_sub_pd(ix2,jx0);
 454             dy20             = _mm256_sub_pd(iy2,jy0);
 455             dz20             = _mm256_sub_pd(iz2,jz0);
 456
 457             /* Calculate squared distance and things based on it */
 458             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 459             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 460             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 461
 462             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 463             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 464             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 465
 466             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 467             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 468             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 469
 470             /* Load parameters for j particles */
 471             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 472                                                                  charge+jnrC+0,charge+jnrD+0);
 473             vdwjidx0A        = 2*vdwtype[jnrA+0];
 474             vdwjidx0B        = 2*vdwtype[jnrB+0];
 475             vdwjidx0C        = 2*vdwtype[jnrC+0];
 476             vdwjidx0D        = 2*vdwtype[jnrD+0];
 477
 478             fjx0             = _mm256_setzero_pd();
 479             fjy0             = _mm256_setzero_pd();
 480             fjz0             = _mm256_setzero_pd();
 481
 482             /**************************
 483              * CALCULATE INTERACTIONS *
 484              **************************/
 485
 486             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 487             {
 488
 489             r00              = _mm256_mul_pd(rsq00,rinv00);
 490             r00              = _mm256_andnot_pd(dummy_mask,r00);
 491
 492             /* Compute parameters for interactions between i and j atoms */
 493             qq00             = _mm256_mul_pd(iq0,jq0);
 494             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 495                                             vdwioffsetptr0+vdwjidx0B,
 496                                             vdwioffsetptr0+vdwjidx0C,
 497                                             vdwioffsetptr0+vdwjidx0D,
 498                                             &c6_00,&c12_00);
 499
 500             /* REACTION-FIELD ELECTROSTATICS */
 501             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 502             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 503
 504             /* LENNARD-JONES DISPERSION/REPULSION */
 505
 506             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 507             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 508             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 509             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 510             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 511
 512             d                = _mm256_sub_pd(r00,rswitch);
 513             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 514             d2               = _mm256_mul_pd(d,d);
 515             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)))))));
 516
 517             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 518
 519             /* Evaluate switch function */
 520             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 521             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
 522             vvdw             = _mm256_mul_pd(vvdw,sw);
 523             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 524
 525             /* Update potential sum for this i atom from the interaction with this j atom. */
 526             velec            = _mm256_and_pd(velec,cutoff_mask);
 527             velec            = _mm256_andnot_pd(dummy_mask,velec);
 528             velecsum         = _mm256_add_pd(velecsum,velec);
 529             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 530             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 531             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 532
 533             fscal            = _mm256_add_pd(felec,fvdw);
 534
 535             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 536
 537             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 538
 539             /* Calculate temporary vectorial force */
 540             tx               = _mm256_mul_pd(fscal,dx00);
 541             ty               = _mm256_mul_pd(fscal,dy00);
 542             tz               = _mm256_mul_pd(fscal,dz00);
 543
 544             /* Update vectorial force */
 545             fix0             = _mm256_add_pd(fix0,tx);
 546             fiy0             = _mm256_add_pd(fiy0,ty);
 547             fiz0             = _mm256_add_pd(fiz0,tz);
 548
 549             fjx0             = _mm256_add_pd(fjx0,tx);
 550             fjy0             = _mm256_add_pd(fjy0,ty);
 551             fjz0             = _mm256_add_pd(fjz0,tz);
 552
 553             }
 554
 555             /**************************
 556              * CALCULATE INTERACTIONS *
 557              **************************/
 558
 559             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 560             {
 561
 562             /* Compute parameters for interactions between i and j atoms */
 563             qq10             = _mm256_mul_pd(iq1,jq0);
 564
 565             /* REACTION-FIELD ELECTROSTATICS */
 566             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 567             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 568
 569             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 570
 571             /* Update potential sum for this i atom from the interaction with this j atom. */
 572             velec            = _mm256_and_pd(velec,cutoff_mask);
 573             velec            = _mm256_andnot_pd(dummy_mask,velec);
 574             velecsum         = _mm256_add_pd(velecsum,velec);
 575
 576             fscal            = felec;
 577
 578             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 579
 580             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 581
 582             /* Calculate temporary vectorial force */
 583             tx               = _mm256_mul_pd(fscal,dx10);
 584             ty               = _mm256_mul_pd(fscal,dy10);
 585             tz               = _mm256_mul_pd(fscal,dz10);
 586
 587             /* Update vectorial force */
 588             fix1             = _mm256_add_pd(fix1,tx);
 589             fiy1             = _mm256_add_pd(fiy1,ty);
 590             fiz1             = _mm256_add_pd(fiz1,tz);
 591
 592             fjx0             = _mm256_add_pd(fjx0,tx);
 593             fjy0             = _mm256_add_pd(fjy0,ty);
 594             fjz0             = _mm256_add_pd(fjz0,tz);
 595
 596             }
 597
 598             /**************************
 599              * CALCULATE INTERACTIONS *
 600              **************************/
 601
 602             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 603             {
 604
 605             /* Compute parameters for interactions between i and j atoms */
 606             qq20             = _mm256_mul_pd(iq2,jq0);
 607
 608             /* REACTION-FIELD ELECTROSTATICS */
 609             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 610             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 611
 612             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 613
 614             /* Update potential sum for this i atom from the interaction with this j atom. */
 615             velec            = _mm256_and_pd(velec,cutoff_mask);
 616             velec            = _mm256_andnot_pd(dummy_mask,velec);
 617             velecsum         = _mm256_add_pd(velecsum,velec);
 618
 619             fscal            = felec;
 620
 621             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 622
 623             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 624
 625             /* Calculate temporary vectorial force */
 626             tx               = _mm256_mul_pd(fscal,dx20);
 627             ty               = _mm256_mul_pd(fscal,dy20);
 628             tz               = _mm256_mul_pd(fscal,dz20);
 629
 630             /* Update vectorial force */
 631             fix2             = _mm256_add_pd(fix2,tx);
 632             fiy2             = _mm256_add_pd(fiy2,ty);
 633             fiz2             = _mm256_add_pd(fiz2,tz);
 634
 635             fjx0             = _mm256_add_pd(fjx0,tx);
 636             fjy0             = _mm256_add_pd(fjy0,ty);
 637             fjz0             = _mm256_add_pd(fjz0,tz);
 638
 639             }
 640
 641             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 642             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 643             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 644             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 645
 646             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 647
 648             /* Inner loop uses 146 flops */
 649         }
 650
 651         /* End of innermost loop */
 652
 653         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 654                                                  f+i_coord_offset,fshift+i_shift_offset);
 655
 656         ggid                        = gid[iidx];
 657         /* Update potential energies */
 658         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 659         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 660
 661         /* Increment number of inner iterations */
 662         inneriter                  += j_index_end - j_index_start;
 663
 664         /* Outer loop uses 20 flops */
 665     }
 666
 667     /* Increment number of outer iterations */
 668     outeriter        += nri;
 669
 670     /* Update outer/inner flops */
 671
 672     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*146);
 673 }
 674 /*
 675  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_avx_256_double
 676  * Electrostatics interaction: ReactionField
 677  * VdW interaction:            LennardJones
 678  * Geometry:                   Water3-Particle
 679  * Calculate force/pot:        Force
 680  */
 681 void
 682 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_avx_256_double
 683                     (t_nblist                    * gmx_restrict       nlist,
 684                      rvec                        * gmx_restrict          xx,
 685                      rvec                        * gmx_restrict          ff,
 686                      t_forcerec                  * gmx_restrict          fr,
 687                      t_mdatoms                   * gmx_restrict     mdatoms,
 688                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 689                      t_nrnb                      * gmx_restrict        nrnb)
 690 {
 691     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 692      * just 0 for non-waters.
 693      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 694      * jnr indices corresponding to data put in the four positions in the SIMD register.
 695      */
 696     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 697     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 698     int              jnrA,jnrB,jnrC,jnrD;
 699     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 700     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 701     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 702     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 703     real             rcutoff_scalar;
 704     real             *shiftvec,*fshift,*x,*f;
 705     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 706     real             scratch[4*DIM];
 707     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 708     real *           vdwioffsetptr0;
 709     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 710     real *           vdwioffsetptr1;
 711     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 712     real *           vdwioffsetptr2;
 713     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 714     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 715     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 716     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 717     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 718     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 719     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 720     real             *charge;
 721     int              nvdwtype;
 722     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 723     int              *vdwtype;
 724     real             *vdwparam;
 725     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 726     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 727     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 728     real             rswitch_scalar,d_scalar;
 729     __m256d          dummy_mask,cutoff_mask;
 730     __m128           tmpmask0,tmpmask1;
 731     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 732     __m256d          one     = _mm256_set1_pd(1.0);
 733     __m256d          two     = _mm256_set1_pd(2.0);
 734     x                = xx[0];
 735     f                = ff[0];
 736
 737     nri              = nlist->nri;
 738     iinr             = nlist->iinr;
 739     jindex           = nlist->jindex;
 740     jjnr             = nlist->jjnr;
 741     shiftidx         = nlist->shift;
 742     gid              = nlist->gid;
 743     shiftvec         = fr->shift_vec[0];
 744     fshift           = fr->fshift[0];
 745     facel            = _mm256_set1_pd(fr->epsfac);
 746     charge           = mdatoms->chargeA;
 747     krf              = _mm256_set1_pd(fr->ic->k_rf);
 748     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 749     crf              = _mm256_set1_pd(fr->ic->c_rf);
 750     nvdwtype         = fr->ntype;
 751     vdwparam         = fr->nbfp;
 752     vdwtype          = mdatoms->typeA;
 753
 754     /* Setup water-specific parameters */
 755     inr              = nlist->iinr[0];
 756     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 757     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 758     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 759     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 760
 761     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 762     rcutoff_scalar   = fr->rcoulomb;
 763     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 764     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 765
 766     rswitch_scalar   = fr->rvdw_switch;
 767     rswitch          = _mm256_set1_pd(rswitch_scalar);
 768     /* Setup switch parameters */
 769     d_scalar         = rcutoff_scalar-rswitch_scalar;
 770     d                = _mm256_set1_pd(d_scalar);
 771     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 772     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 773     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 774     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 775     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 776     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 777
 778     /* Avoid stupid compiler warnings */
 779     jnrA = jnrB = jnrC = jnrD = 0;
 780     j_coord_offsetA = 0;
 781     j_coord_offsetB = 0;
 782     j_coord_offsetC = 0;
 783     j_coord_offsetD = 0;
 784
 785     outeriter        = 0;
 786     inneriter        = 0;
 787
 788     for(iidx=0;iidx<4*DIM;iidx++)
 789     {
 790         scratch[iidx] = 0.0;
 791     }
 792
 793     /* Start outer loop over neighborlists */
 794     for(iidx=0; iidx<nri; iidx++)
 795     {
 796         /* Load shift vector for this list */
 797         i_shift_offset   = DIM*shiftidx[iidx];
 798
 799         /* Load limits for loop over neighbors */
 800         j_index_start    = jindex[iidx];
 801         j_index_end      = jindex[iidx+1];
 802
 803         /* Get outer coordinate index */
 804         inr              = iinr[iidx];
 805         i_coord_offset   = DIM*inr;
 806
 807         /* Load i particle coords and add shift vector */
 808         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 809                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 810
 811         fix0             = _mm256_setzero_pd();
 812         fiy0             = _mm256_setzero_pd();
 813         fiz0             = _mm256_setzero_pd();
 814         fix1             = _mm256_setzero_pd();
 815         fiy1             = _mm256_setzero_pd();
 816         fiz1             = _mm256_setzero_pd();
 817         fix2             = _mm256_setzero_pd();
 818         fiy2             = _mm256_setzero_pd();
 819         fiz2             = _mm256_setzero_pd();
 820
 821         /* Start inner kernel loop */
 822         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 823         {
 824
 825             /* Get j neighbor index, and coordinate index */
 826             jnrA             = jjnr[jidx];
 827             jnrB             = jjnr[jidx+1];
 828             jnrC             = jjnr[jidx+2];
 829             jnrD             = jjnr[jidx+3];
 830             j_coord_offsetA  = DIM*jnrA;
 831             j_coord_offsetB  = DIM*jnrB;
 832             j_coord_offsetC  = DIM*jnrC;
 833             j_coord_offsetD  = DIM*jnrD;
 834
 835             /* load j atom coordinates */
 836             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 837                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 838                                                  &jx0,&jy0,&jz0);
 839
 840             /* Calculate displacement vector */
 841             dx00             = _mm256_sub_pd(ix0,jx0);
 842             dy00             = _mm256_sub_pd(iy0,jy0);
 843             dz00             = _mm256_sub_pd(iz0,jz0);
 844             dx10             = _mm256_sub_pd(ix1,jx0);
 845             dy10             = _mm256_sub_pd(iy1,jy0);
 846             dz10             = _mm256_sub_pd(iz1,jz0);
 847             dx20             = _mm256_sub_pd(ix2,jx0);
 848             dy20             = _mm256_sub_pd(iy2,jy0);
 849             dz20             = _mm256_sub_pd(iz2,jz0);
 850
 851             /* Calculate squared distance and things based on it */
 852             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 853             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 854             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 855
 856             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 857             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 858             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 859
 860             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 861             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 862             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 863
 864             /* Load parameters for j particles */
 865             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 866                                                                  charge+jnrC+0,charge+jnrD+0);
 867             vdwjidx0A        = 2*vdwtype[jnrA+0];
 868             vdwjidx0B        = 2*vdwtype[jnrB+0];
 869             vdwjidx0C        = 2*vdwtype[jnrC+0];
 870             vdwjidx0D        = 2*vdwtype[jnrD+0];
 871
 872             fjx0             = _mm256_setzero_pd();
 873             fjy0             = _mm256_setzero_pd();
 874             fjz0             = _mm256_setzero_pd();
 875
 876             /**************************
 877              * CALCULATE INTERACTIONS *
 878              **************************/
 879
 880             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 881             {
 882
 883             r00              = _mm256_mul_pd(rsq00,rinv00);
 884
 885             /* Compute parameters for interactions between i and j atoms */
 886             qq00             = _mm256_mul_pd(iq0,jq0);
 887             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 888                                             vdwioffsetptr0+vdwjidx0B,
 889                                             vdwioffsetptr0+vdwjidx0C,
 890                                             vdwioffsetptr0+vdwjidx0D,
 891                                             &c6_00,&c12_00);
 892
 893             /* REACTION-FIELD ELECTROSTATICS */
 894             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 895
 896             /* LENNARD-JONES DISPERSION/REPULSION */
 897
 898             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 899             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 900             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 901             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 902             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 903
 904             d                = _mm256_sub_pd(r00,rswitch);
 905             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 906             d2               = _mm256_mul_pd(d,d);
 907             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)))))));
 908
 909             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 910
 911             /* Evaluate switch function */
 912             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 913             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
 914             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 915
 916             fscal            = _mm256_add_pd(felec,fvdw);
 917
 918             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 919
 920             /* Calculate temporary vectorial force */
 921             tx               = _mm256_mul_pd(fscal,dx00);
 922             ty               = _mm256_mul_pd(fscal,dy00);
 923             tz               = _mm256_mul_pd(fscal,dz00);
 924
 925             /* Update vectorial force */
 926             fix0             = _mm256_add_pd(fix0,tx);
 927             fiy0             = _mm256_add_pd(fiy0,ty);
 928             fiz0             = _mm256_add_pd(fiz0,tz);
 929
 930             fjx0             = _mm256_add_pd(fjx0,tx);
 931             fjy0             = _mm256_add_pd(fjy0,ty);
 932             fjz0             = _mm256_add_pd(fjz0,tz);
 933
 934             }
 935
 936             /**************************
 937              * CALCULATE INTERACTIONS *
 938              **************************/
 939
 940             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 941             {
 942
 943             /* Compute parameters for interactions between i and j atoms */
 944             qq10             = _mm256_mul_pd(iq1,jq0);
 945
 946             /* REACTION-FIELD ELECTROSTATICS */
 947             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 948
 949             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 950
 951             fscal            = felec;
 952
 953             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 954
 955             /* Calculate temporary vectorial force */
 956             tx               = _mm256_mul_pd(fscal,dx10);
 957             ty               = _mm256_mul_pd(fscal,dy10);
 958             tz               = _mm256_mul_pd(fscal,dz10);
 959
 960             /* Update vectorial force */
 961             fix1             = _mm256_add_pd(fix1,tx);
 962             fiy1             = _mm256_add_pd(fiy1,ty);
 963             fiz1             = _mm256_add_pd(fiz1,tz);
 964
 965             fjx0             = _mm256_add_pd(fjx0,tx);
 966             fjy0             = _mm256_add_pd(fjy0,ty);
 967             fjz0             = _mm256_add_pd(fjz0,tz);
 968
 969             }
 970
 971             /**************************
 972              * CALCULATE INTERACTIONS *
 973              **************************/
 974
 975             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 976             {
 977
 978             /* Compute parameters for interactions between i and j atoms */
 979             qq20             = _mm256_mul_pd(iq2,jq0);
 980
 981             /* REACTION-FIELD ELECTROSTATICS */
 982             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 983
 984             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 985
 986             fscal            = felec;
 987
 988             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 989
 990             /* Calculate temporary vectorial force */
 991             tx               = _mm256_mul_pd(fscal,dx20);
 992             ty               = _mm256_mul_pd(fscal,dy20);
 993             tz               = _mm256_mul_pd(fscal,dz20);
 994
 995             /* Update vectorial force */
 996             fix2             = _mm256_add_pd(fix2,tx);
 997             fiy2             = _mm256_add_pd(fiy2,ty);
 998             fiz2             = _mm256_add_pd(fiz2,tz);
 999
1000             fjx0             = _mm256_add_pd(fjx0,tx);
1001             fjy0             = _mm256_add_pd(fjy0,ty);
1002             fjz0             = _mm256_add_pd(fjz0,tz);
1003
1004             }
1005
1006             fjptrA             = f+j_coord_offsetA;
1007             fjptrB             = f+j_coord_offsetB;
1008             fjptrC             = f+j_coord_offsetC;
1009             fjptrD             = f+j_coord_offsetD;
1010
1011             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1012
1013             /* Inner loop uses 124 flops */
1014         }
1015
1016         if(jidx<j_index_end)
1017         {
1018
1019             /* Get j neighbor index, and coordinate index */
1020             jnrlistA         = jjnr[jidx];
1021             jnrlistB         = jjnr[jidx+1];
1022             jnrlistC         = jjnr[jidx+2];
1023             jnrlistD         = jjnr[jidx+3];
1024             /* Sign of each element will be negative for non-real atoms.
1025              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1026              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1027              */
1028             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1029
1030             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1031             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1032             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1033
1034             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1035             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1036             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1037             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1038             j_coord_offsetA  = DIM*jnrA;
1039             j_coord_offsetB  = DIM*jnrB;
1040             j_coord_offsetC  = DIM*jnrC;
1041             j_coord_offsetD  = DIM*jnrD;
1042
1043             /* load j atom coordinates */
1044             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1045                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1046                                                  &jx0,&jy0,&jz0);
1047
1048             /* Calculate displacement vector */
1049             dx00             = _mm256_sub_pd(ix0,jx0);
1050             dy00             = _mm256_sub_pd(iy0,jy0);
1051             dz00             = _mm256_sub_pd(iz0,jz0);
1052             dx10             = _mm256_sub_pd(ix1,jx0);
1053             dy10             = _mm256_sub_pd(iy1,jy0);
1054             dz10             = _mm256_sub_pd(iz1,jz0);
1055             dx20             = _mm256_sub_pd(ix2,jx0);
1056             dy20             = _mm256_sub_pd(iy2,jy0);
1057             dz20             = _mm256_sub_pd(iz2,jz0);
1058
1059             /* Calculate squared distance and things based on it */
1060             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1061             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1062             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1063
1064             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1065             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1066             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1067
1068             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1069             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1070             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1071
1072             /* Load parameters for j particles */
1073             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1074                                                                  charge+jnrC+0,charge+jnrD+0);
1075             vdwjidx0A        = 2*vdwtype[jnrA+0];
1076             vdwjidx0B        = 2*vdwtype[jnrB+0];
1077             vdwjidx0C        = 2*vdwtype[jnrC+0];
1078             vdwjidx0D        = 2*vdwtype[jnrD+0];
1079
1080             fjx0             = _mm256_setzero_pd();
1081             fjy0             = _mm256_setzero_pd();
1082             fjz0             = _mm256_setzero_pd();
1083
1084             /**************************
1085              * CALCULATE INTERACTIONS *
1086              **************************/
1087
1088             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1089             {
1090
1091             r00              = _mm256_mul_pd(rsq00,rinv00);
1092             r00              = _mm256_andnot_pd(dummy_mask,r00);
1093
1094             /* Compute parameters for interactions between i and j atoms */
1095             qq00             = _mm256_mul_pd(iq0,jq0);
1096             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1097                                             vdwioffsetptr0+vdwjidx0B,
1098                                             vdwioffsetptr0+vdwjidx0C,
1099                                             vdwioffsetptr0+vdwjidx0D,
1100                                             &c6_00,&c12_00);
1101
1102             /* REACTION-FIELD ELECTROSTATICS */
1103             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
1104
1105             /* LENNARD-JONES DISPERSION/REPULSION */
1106
1107             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1108             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
1109             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
1110             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
1111             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
1112
1113             d                = _mm256_sub_pd(r00,rswitch);
1114             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1115             d2               = _mm256_mul_pd(d,d);
1116             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)))))));
1117
1118             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1119
1120             /* Evaluate switch function */
1121             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1122             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
1123             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1124
1125             fscal            = _mm256_add_pd(felec,fvdw);
1126
1127             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1128
1129             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1130
1131             /* Calculate temporary vectorial force */
1132             tx               = _mm256_mul_pd(fscal,dx00);
1133             ty               = _mm256_mul_pd(fscal,dy00);
1134             tz               = _mm256_mul_pd(fscal,dz00);
1135
1136             /* Update vectorial force */
1137             fix0             = _mm256_add_pd(fix0,tx);
1138             fiy0             = _mm256_add_pd(fiy0,ty);
1139             fiz0             = _mm256_add_pd(fiz0,tz);
1140
1141             fjx0             = _mm256_add_pd(fjx0,tx);
1142             fjy0             = _mm256_add_pd(fjy0,ty);
1143             fjz0             = _mm256_add_pd(fjz0,tz);
1144
1145             }
1146
1147             /**************************
1148              * CALCULATE INTERACTIONS *
1149              **************************/
1150
1151             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1152             {
1153
1154             /* Compute parameters for interactions between i and j atoms */
1155             qq10             = _mm256_mul_pd(iq1,jq0);
1156
1157             /* REACTION-FIELD ELECTROSTATICS */
1158             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1159
1160             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1161
1162             fscal            = felec;
1163
1164             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1165
1166             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1167
1168             /* Calculate temporary vectorial force */
1169             tx               = _mm256_mul_pd(fscal,dx10);
1170             ty               = _mm256_mul_pd(fscal,dy10);
1171             tz               = _mm256_mul_pd(fscal,dz10);
1172
1173             /* Update vectorial force */
1174             fix1             = _mm256_add_pd(fix1,tx);
1175             fiy1             = _mm256_add_pd(fiy1,ty);
1176             fiz1             = _mm256_add_pd(fiz1,tz);
1177
1178             fjx0             = _mm256_add_pd(fjx0,tx);
1179             fjy0             = _mm256_add_pd(fjy0,ty);
1180             fjz0             = _mm256_add_pd(fjz0,tz);
1181
1182             }
1183
1184             /**************************
1185              * CALCULATE INTERACTIONS *
1186              **************************/
1187
1188             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1189             {
1190
1191             /* Compute parameters for interactions between i and j atoms */
1192             qq20             = _mm256_mul_pd(iq2,jq0);
1193
1194             /* REACTION-FIELD ELECTROSTATICS */
1195             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1196
1197             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1198
1199             fscal            = felec;
1200
1201             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1202
1203             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1204
1205             /* Calculate temporary vectorial force */
1206             tx               = _mm256_mul_pd(fscal,dx20);
1207             ty               = _mm256_mul_pd(fscal,dy20);
1208             tz               = _mm256_mul_pd(fscal,dz20);
1209
1210             /* Update vectorial force */
1211             fix2             = _mm256_add_pd(fix2,tx);
1212             fiy2             = _mm256_add_pd(fiy2,ty);
1213             fiz2             = _mm256_add_pd(fiz2,tz);
1214
1215             fjx0             = _mm256_add_pd(fjx0,tx);
1216             fjy0             = _mm256_add_pd(fjy0,ty);
1217             fjz0             = _mm256_add_pd(fjz0,tz);
1218
1219             }
1220
1221             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1222             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1223             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1224             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1225
1226             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1227
1228             /* Inner loop uses 125 flops */
1229         }
1230
1231         /* End of innermost loop */
1232
1233         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1234                                                  f+i_coord_offset,fshift+i_shift_offset);
1235
1236         /* Increment number of inner iterations */
1237         inneriter                  += j_index_end - j_index_start;
1238
1239         /* Outer loop uses 18 flops */
1240     }
1241
1242     /* Increment number of outer iterations */
1243     outeriter        += nri;
1244
1245     /* Update outer/inner flops */
1246
1247     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*125);
1248 }