src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCoul_VdwLJ_GeomW3P1_avx_256_single.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_avx_256_single
  51  * Electrostatics interaction: Coulomb
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_avx_256_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrE,jnrF,jnrG,jnrH;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  83     real             scratch[4*DIM];
  84     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     real *           vdwioffsetptr0;
  86     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     real *           vdwioffsetptr1;
  88     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     real *           vdwioffsetptr2;
  90     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  92     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 103     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 104     __m256           dummy_mask,cutoff_mask;
 105     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 106     __m256           one     = _mm256_set1_ps(1.0);
 107     __m256           two     = _mm256_set1_ps(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm256_set1_ps(fr->ic->epsfac);
 120     charge           = mdatoms->chargeA;
 121     nvdwtype         = fr->ntype;
 122     vdwparam         = fr->nbfp;
 123     vdwtype          = mdatoms->typeA;
 124
 125     /* Setup water-specific parameters */
 126     inr              = nlist->iinr[0];
 127     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 128     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 129     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 130     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 131
 132     /* Avoid stupid compiler warnings */
 133     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 134     j_coord_offsetA = 0;
 135     j_coord_offsetB = 0;
 136     j_coord_offsetC = 0;
 137     j_coord_offsetD = 0;
 138     j_coord_offsetE = 0;
 139     j_coord_offsetF = 0;
 140     j_coord_offsetG = 0;
 141     j_coord_offsetH = 0;
 142
 143     outeriter        = 0;
 144     inneriter        = 0;
 145
 146     for(iidx=0;iidx<4*DIM;iidx++)
 147     {
 148         scratch[iidx] = 0.0;
 149     }
 150
 151     /* Start outer loop over neighborlists */
 152     for(iidx=0; iidx<nri; iidx++)
 153     {
 154         /* Load shift vector for this list */
 155         i_shift_offset   = DIM*shiftidx[iidx];
 156
 157         /* Load limits for loop over neighbors */
 158         j_index_start    = jindex[iidx];
 159         j_index_end      = jindex[iidx+1];
 160
 161         /* Get outer coordinate index */
 162         inr              = iinr[iidx];
 163         i_coord_offset   = DIM*inr;
 164
 165         /* Load i particle coords and add shift vector */
 166         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 167                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 168
 169         fix0             = _mm256_setzero_ps();
 170         fiy0             = _mm256_setzero_ps();
 171         fiz0             = _mm256_setzero_ps();
 172         fix1             = _mm256_setzero_ps();
 173         fiy1             = _mm256_setzero_ps();
 174         fiz1             = _mm256_setzero_ps();
 175         fix2             = _mm256_setzero_ps();
 176         fiy2             = _mm256_setzero_ps();
 177         fiz2             = _mm256_setzero_ps();
 178
 179         /* Reset potential sums */
 180         velecsum         = _mm256_setzero_ps();
 181         vvdwsum          = _mm256_setzero_ps();
 182
 183         /* Start inner kernel loop */
 184         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 185         {
 186
 187             /* Get j neighbor index, and coordinate index */
 188             jnrA             = jjnr[jidx];
 189             jnrB             = jjnr[jidx+1];
 190             jnrC             = jjnr[jidx+2];
 191             jnrD             = jjnr[jidx+3];
 192             jnrE             = jjnr[jidx+4];
 193             jnrF             = jjnr[jidx+5];
 194             jnrG             = jjnr[jidx+6];
 195             jnrH             = jjnr[jidx+7];
 196             j_coord_offsetA  = DIM*jnrA;
 197             j_coord_offsetB  = DIM*jnrB;
 198             j_coord_offsetC  = DIM*jnrC;
 199             j_coord_offsetD  = DIM*jnrD;
 200             j_coord_offsetE  = DIM*jnrE;
 201             j_coord_offsetF  = DIM*jnrF;
 202             j_coord_offsetG  = DIM*jnrG;
 203             j_coord_offsetH  = DIM*jnrH;
 204
 205             /* load j atom coordinates */
 206             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 207                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 208                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 209                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 210                                                  &jx0,&jy0,&jz0);
 211
 212             /* Calculate displacement vector */
 213             dx00             = _mm256_sub_ps(ix0,jx0);
 214             dy00             = _mm256_sub_ps(iy0,jy0);
 215             dz00             = _mm256_sub_ps(iz0,jz0);
 216             dx10             = _mm256_sub_ps(ix1,jx0);
 217             dy10             = _mm256_sub_ps(iy1,jy0);
 218             dz10             = _mm256_sub_ps(iz1,jz0);
 219             dx20             = _mm256_sub_ps(ix2,jx0);
 220             dy20             = _mm256_sub_ps(iy2,jy0);
 221             dz20             = _mm256_sub_ps(iz2,jz0);
 222
 223             /* Calculate squared distance and things based on it */
 224             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 225             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 226             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 227
 228             rinv00           = avx256_invsqrt_f(rsq00);
 229             rinv10           = avx256_invsqrt_f(rsq10);
 230             rinv20           = avx256_invsqrt_f(rsq20);
 231
 232             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 233             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 234             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 235
 236             /* Load parameters for j particles */
 237             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 238                                                                  charge+jnrC+0,charge+jnrD+0,
 239                                                                  charge+jnrE+0,charge+jnrF+0,
 240                                                                  charge+jnrG+0,charge+jnrH+0);
 241             vdwjidx0A        = 2*vdwtype[jnrA+0];
 242             vdwjidx0B        = 2*vdwtype[jnrB+0];
 243             vdwjidx0C        = 2*vdwtype[jnrC+0];
 244             vdwjidx0D        = 2*vdwtype[jnrD+0];
 245             vdwjidx0E        = 2*vdwtype[jnrE+0];
 246             vdwjidx0F        = 2*vdwtype[jnrF+0];
 247             vdwjidx0G        = 2*vdwtype[jnrG+0];
 248             vdwjidx0H        = 2*vdwtype[jnrH+0];
 249
 250             fjx0             = _mm256_setzero_ps();
 251             fjy0             = _mm256_setzero_ps();
 252             fjz0             = _mm256_setzero_ps();
 253
 254             /**************************
 255              * CALCULATE INTERACTIONS *
 256              **************************/
 257
 258             /* Compute parameters for interactions between i and j atoms */
 259             qq00             = _mm256_mul_ps(iq0,jq0);
 260             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 261                                             vdwioffsetptr0+vdwjidx0B,
 262                                             vdwioffsetptr0+vdwjidx0C,
 263                                             vdwioffsetptr0+vdwjidx0D,
 264                                             vdwioffsetptr0+vdwjidx0E,
 265                                             vdwioffsetptr0+vdwjidx0F,
 266                                             vdwioffsetptr0+vdwjidx0G,
 267                                             vdwioffsetptr0+vdwjidx0H,
 268                                             &c6_00,&c12_00);
 269
 270             /* COULOMB ELECTROSTATICS */
 271             velec            = _mm256_mul_ps(qq00,rinv00);
 272             felec            = _mm256_mul_ps(velec,rinvsq00);
 273
 274             /* LENNARD-JONES DISPERSION/REPULSION */
 275
 276             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 277             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 278             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 279             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 280             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 281
 282             /* Update potential sum for this i atom from the interaction with this j atom. */
 283             velecsum         = _mm256_add_ps(velecsum,velec);
 284             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 285
 286             fscal            = _mm256_add_ps(felec,fvdw);
 287
 288             /* Calculate temporary vectorial force */
 289             tx               = _mm256_mul_ps(fscal,dx00);
 290             ty               = _mm256_mul_ps(fscal,dy00);
 291             tz               = _mm256_mul_ps(fscal,dz00);
 292
 293             /* Update vectorial force */
 294             fix0             = _mm256_add_ps(fix0,tx);
 295             fiy0             = _mm256_add_ps(fiy0,ty);
 296             fiz0             = _mm256_add_ps(fiz0,tz);
 297
 298             fjx0             = _mm256_add_ps(fjx0,tx);
 299             fjy0             = _mm256_add_ps(fjy0,ty);
 300             fjz0             = _mm256_add_ps(fjz0,tz);
 301
 302             /**************************
 303              * CALCULATE INTERACTIONS *
 304              **************************/
 305
 306             /* Compute parameters for interactions between i and j atoms */
 307             qq10             = _mm256_mul_ps(iq1,jq0);
 308
 309             /* COULOMB ELECTROSTATICS */
 310             velec            = _mm256_mul_ps(qq10,rinv10);
 311             felec            = _mm256_mul_ps(velec,rinvsq10);
 312
 313             /* Update potential sum for this i atom from the interaction with this j atom. */
 314             velecsum         = _mm256_add_ps(velecsum,velec);
 315
 316             fscal            = felec;
 317
 318             /* Calculate temporary vectorial force */
 319             tx               = _mm256_mul_ps(fscal,dx10);
 320             ty               = _mm256_mul_ps(fscal,dy10);
 321             tz               = _mm256_mul_ps(fscal,dz10);
 322
 323             /* Update vectorial force */
 324             fix1             = _mm256_add_ps(fix1,tx);
 325             fiy1             = _mm256_add_ps(fiy1,ty);
 326             fiz1             = _mm256_add_ps(fiz1,tz);
 327
 328             fjx0             = _mm256_add_ps(fjx0,tx);
 329             fjy0             = _mm256_add_ps(fjy0,ty);
 330             fjz0             = _mm256_add_ps(fjz0,tz);
 331
 332             /**************************
 333              * CALCULATE INTERACTIONS *
 334              **************************/
 335
 336             /* Compute parameters for interactions between i and j atoms */
 337             qq20             = _mm256_mul_ps(iq2,jq0);
 338
 339             /* COULOMB ELECTROSTATICS */
 340             velec            = _mm256_mul_ps(qq20,rinv20);
 341             felec            = _mm256_mul_ps(velec,rinvsq20);
 342
 343             /* Update potential sum for this i atom from the interaction with this j atom. */
 344             velecsum         = _mm256_add_ps(velecsum,velec);
 345
 346             fscal            = felec;
 347
 348             /* Calculate temporary vectorial force */
 349             tx               = _mm256_mul_ps(fscal,dx20);
 350             ty               = _mm256_mul_ps(fscal,dy20);
 351             tz               = _mm256_mul_ps(fscal,dz20);
 352
 353             /* Update vectorial force */
 354             fix2             = _mm256_add_ps(fix2,tx);
 355             fiy2             = _mm256_add_ps(fiy2,ty);
 356             fiz2             = _mm256_add_ps(fiz2,tz);
 357
 358             fjx0             = _mm256_add_ps(fjx0,tx);
 359             fjy0             = _mm256_add_ps(fjy0,ty);
 360             fjz0             = _mm256_add_ps(fjz0,tz);
 361
 362             fjptrA             = f+j_coord_offsetA;
 363             fjptrB             = f+j_coord_offsetB;
 364             fjptrC             = f+j_coord_offsetC;
 365             fjptrD             = f+j_coord_offsetD;
 366             fjptrE             = f+j_coord_offsetE;
 367             fjptrF             = f+j_coord_offsetF;
 368             fjptrG             = f+j_coord_offsetG;
 369             fjptrH             = f+j_coord_offsetH;
 370
 371             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 372
 373             /* Inner loop uses 96 flops */
 374         }
 375
 376         if(jidx<j_index_end)
 377         {
 378
 379             /* Get j neighbor index, and coordinate index */
 380             jnrlistA         = jjnr[jidx];
 381             jnrlistB         = jjnr[jidx+1];
 382             jnrlistC         = jjnr[jidx+2];
 383             jnrlistD         = jjnr[jidx+3];
 384             jnrlistE         = jjnr[jidx+4];
 385             jnrlistF         = jjnr[jidx+5];
 386             jnrlistG         = jjnr[jidx+6];
 387             jnrlistH         = jjnr[jidx+7];
 388             /* Sign of each element will be negative for non-real atoms.
 389              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 390              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 391              */
 392             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 393                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 394
 395             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 396             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 397             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 398             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 399             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 400             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 401             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 402             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 403             j_coord_offsetA  = DIM*jnrA;
 404             j_coord_offsetB  = DIM*jnrB;
 405             j_coord_offsetC  = DIM*jnrC;
 406             j_coord_offsetD  = DIM*jnrD;
 407             j_coord_offsetE  = DIM*jnrE;
 408             j_coord_offsetF  = DIM*jnrF;
 409             j_coord_offsetG  = DIM*jnrG;
 410             j_coord_offsetH  = DIM*jnrH;
 411
 412             /* load j atom coordinates */
 413             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 414                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 415                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 416                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 417                                                  &jx0,&jy0,&jz0);
 418
 419             /* Calculate displacement vector */
 420             dx00             = _mm256_sub_ps(ix0,jx0);
 421             dy00             = _mm256_sub_ps(iy0,jy0);
 422             dz00             = _mm256_sub_ps(iz0,jz0);
 423             dx10             = _mm256_sub_ps(ix1,jx0);
 424             dy10             = _mm256_sub_ps(iy1,jy0);
 425             dz10             = _mm256_sub_ps(iz1,jz0);
 426             dx20             = _mm256_sub_ps(ix2,jx0);
 427             dy20             = _mm256_sub_ps(iy2,jy0);
 428             dz20             = _mm256_sub_ps(iz2,jz0);
 429
 430             /* Calculate squared distance and things based on it */
 431             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 432             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 433             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 434
 435             rinv00           = avx256_invsqrt_f(rsq00);
 436             rinv10           = avx256_invsqrt_f(rsq10);
 437             rinv20           = avx256_invsqrt_f(rsq20);
 438
 439             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 440             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 441             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 442
 443             /* Load parameters for j particles */
 444             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 445                                                                  charge+jnrC+0,charge+jnrD+0,
 446                                                                  charge+jnrE+0,charge+jnrF+0,
 447                                                                  charge+jnrG+0,charge+jnrH+0);
 448             vdwjidx0A        = 2*vdwtype[jnrA+0];
 449             vdwjidx0B        = 2*vdwtype[jnrB+0];
 450             vdwjidx0C        = 2*vdwtype[jnrC+0];
 451             vdwjidx0D        = 2*vdwtype[jnrD+0];
 452             vdwjidx0E        = 2*vdwtype[jnrE+0];
 453             vdwjidx0F        = 2*vdwtype[jnrF+0];
 454             vdwjidx0G        = 2*vdwtype[jnrG+0];
 455             vdwjidx0H        = 2*vdwtype[jnrH+0];
 456
 457             fjx0             = _mm256_setzero_ps();
 458             fjy0             = _mm256_setzero_ps();
 459             fjz0             = _mm256_setzero_ps();
 460
 461             /**************************
 462              * CALCULATE INTERACTIONS *
 463              **************************/
 464
 465             /* Compute parameters for interactions between i and j atoms */
 466             qq00             = _mm256_mul_ps(iq0,jq0);
 467             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 468                                             vdwioffsetptr0+vdwjidx0B,
 469                                             vdwioffsetptr0+vdwjidx0C,
 470                                             vdwioffsetptr0+vdwjidx0D,
 471                                             vdwioffsetptr0+vdwjidx0E,
 472                                             vdwioffsetptr0+vdwjidx0F,
 473                                             vdwioffsetptr0+vdwjidx0G,
 474                                             vdwioffsetptr0+vdwjidx0H,
 475                                             &c6_00,&c12_00);
 476
 477             /* COULOMB ELECTROSTATICS */
 478             velec            = _mm256_mul_ps(qq00,rinv00);
 479             felec            = _mm256_mul_ps(velec,rinvsq00);
 480
 481             /* LENNARD-JONES DISPERSION/REPULSION */
 482
 483             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 484             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 485             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 486             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 487             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 488
 489             /* Update potential sum for this i atom from the interaction with this j atom. */
 490             velec            = _mm256_andnot_ps(dummy_mask,velec);
 491             velecsum         = _mm256_add_ps(velecsum,velec);
 492             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 493             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 494
 495             fscal            = _mm256_add_ps(felec,fvdw);
 496
 497             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 498
 499             /* Calculate temporary vectorial force */
 500             tx               = _mm256_mul_ps(fscal,dx00);
 501             ty               = _mm256_mul_ps(fscal,dy00);
 502             tz               = _mm256_mul_ps(fscal,dz00);
 503
 504             /* Update vectorial force */
 505             fix0             = _mm256_add_ps(fix0,tx);
 506             fiy0             = _mm256_add_ps(fiy0,ty);
 507             fiz0             = _mm256_add_ps(fiz0,tz);
 508
 509             fjx0             = _mm256_add_ps(fjx0,tx);
 510             fjy0             = _mm256_add_ps(fjy0,ty);
 511             fjz0             = _mm256_add_ps(fjz0,tz);
 512
 513             /**************************
 514              * CALCULATE INTERACTIONS *
 515              **************************/
 516
 517             /* Compute parameters for interactions between i and j atoms */
 518             qq10             = _mm256_mul_ps(iq1,jq0);
 519
 520             /* COULOMB ELECTROSTATICS */
 521             velec            = _mm256_mul_ps(qq10,rinv10);
 522             felec            = _mm256_mul_ps(velec,rinvsq10);
 523
 524             /* Update potential sum for this i atom from the interaction with this j atom. */
 525             velec            = _mm256_andnot_ps(dummy_mask,velec);
 526             velecsum         = _mm256_add_ps(velecsum,velec);
 527
 528             fscal            = felec;
 529
 530             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 531
 532             /* Calculate temporary vectorial force */
 533             tx               = _mm256_mul_ps(fscal,dx10);
 534             ty               = _mm256_mul_ps(fscal,dy10);
 535             tz               = _mm256_mul_ps(fscal,dz10);
 536
 537             /* Update vectorial force */
 538             fix1             = _mm256_add_ps(fix1,tx);
 539             fiy1             = _mm256_add_ps(fiy1,ty);
 540             fiz1             = _mm256_add_ps(fiz1,tz);
 541
 542             fjx0             = _mm256_add_ps(fjx0,tx);
 543             fjy0             = _mm256_add_ps(fjy0,ty);
 544             fjz0             = _mm256_add_ps(fjz0,tz);
 545
 546             /**************************
 547              * CALCULATE INTERACTIONS *
 548              **************************/
 549
 550             /* Compute parameters for interactions between i and j atoms */
 551             qq20             = _mm256_mul_ps(iq2,jq0);
 552
 553             /* COULOMB ELECTROSTATICS */
 554             velec            = _mm256_mul_ps(qq20,rinv20);
 555             felec            = _mm256_mul_ps(velec,rinvsq20);
 556
 557             /* Update potential sum for this i atom from the interaction with this j atom. */
 558             velec            = _mm256_andnot_ps(dummy_mask,velec);
 559             velecsum         = _mm256_add_ps(velecsum,velec);
 560
 561             fscal            = felec;
 562
 563             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 564
 565             /* Calculate temporary vectorial force */
 566             tx               = _mm256_mul_ps(fscal,dx20);
 567             ty               = _mm256_mul_ps(fscal,dy20);
 568             tz               = _mm256_mul_ps(fscal,dz20);
 569
 570             /* Update vectorial force */
 571             fix2             = _mm256_add_ps(fix2,tx);
 572             fiy2             = _mm256_add_ps(fiy2,ty);
 573             fiz2             = _mm256_add_ps(fiz2,tz);
 574
 575             fjx0             = _mm256_add_ps(fjx0,tx);
 576             fjy0             = _mm256_add_ps(fjy0,ty);
 577             fjz0             = _mm256_add_ps(fjz0,tz);
 578
 579             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 580             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 581             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 582             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 583             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 584             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 585             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 586             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 587
 588             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 589
 590             /* Inner loop uses 96 flops */
 591         }
 592
 593         /* End of innermost loop */
 594
 595         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 596                                                  f+i_coord_offset,fshift+i_shift_offset);
 597
 598         ggid                        = gid[iidx];
 599         /* Update potential energies */
 600         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 601         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 602
 603         /* Increment number of inner iterations */
 604         inneriter                  += j_index_end - j_index_start;
 605
 606         /* Outer loop uses 20 flops */
 607     }
 608
 609     /* Increment number of outer iterations */
 610     outeriter        += nri;
 611
 612     /* Update outer/inner flops */
 613
 614     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*96);
 615 }
 616 /*
 617  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_avx_256_single
 618  * Electrostatics interaction: Coulomb
 619  * VdW interaction:            LennardJones
 620  * Geometry:                   Water3-Particle
 621  * Calculate force/pot:        Force
 622  */
 623 void
 624 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_avx_256_single
 625                     (t_nblist                    * gmx_restrict       nlist,
 626                      rvec                        * gmx_restrict          xx,
 627                      rvec                        * gmx_restrict          ff,
 628                      struct t_forcerec           * gmx_restrict          fr,
 629                      t_mdatoms                   * gmx_restrict     mdatoms,
 630                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 631                      t_nrnb                      * gmx_restrict        nrnb)
 632 {
 633     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 634      * just 0 for non-waters.
 635      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 636      * jnr indices corresponding to data put in the four positions in the SIMD register.
 637      */
 638     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 639     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 640     int              jnrA,jnrB,jnrC,jnrD;
 641     int              jnrE,jnrF,jnrG,jnrH;
 642     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 643     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 644     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 645     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 646     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 647     real             rcutoff_scalar;
 648     real             *shiftvec,*fshift,*x,*f;
 649     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 650     real             scratch[4*DIM];
 651     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 652     real *           vdwioffsetptr0;
 653     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 654     real *           vdwioffsetptr1;
 655     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 656     real *           vdwioffsetptr2;
 657     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 658     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 659     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 660     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 661     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 662     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 663     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 664     real             *charge;
 665     int              nvdwtype;
 666     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 667     int              *vdwtype;
 668     real             *vdwparam;
 669     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 670     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 671     __m256           dummy_mask,cutoff_mask;
 672     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 673     __m256           one     = _mm256_set1_ps(1.0);
 674     __m256           two     = _mm256_set1_ps(2.0);
 675     x                = xx[0];
 676     f                = ff[0];
 677
 678     nri              = nlist->nri;
 679     iinr             = nlist->iinr;
 680     jindex           = nlist->jindex;
 681     jjnr             = nlist->jjnr;
 682     shiftidx         = nlist->shift;
 683     gid              = nlist->gid;
 684     shiftvec         = fr->shift_vec[0];
 685     fshift           = fr->fshift[0];
 686     facel            = _mm256_set1_ps(fr->ic->epsfac);
 687     charge           = mdatoms->chargeA;
 688     nvdwtype         = fr->ntype;
 689     vdwparam         = fr->nbfp;
 690     vdwtype          = mdatoms->typeA;
 691
 692     /* Setup water-specific parameters */
 693     inr              = nlist->iinr[0];
 694     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 695     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 696     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 697     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 698
 699     /* Avoid stupid compiler warnings */
 700     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 701     j_coord_offsetA = 0;
 702     j_coord_offsetB = 0;
 703     j_coord_offsetC = 0;
 704     j_coord_offsetD = 0;
 705     j_coord_offsetE = 0;
 706     j_coord_offsetF = 0;
 707     j_coord_offsetG = 0;
 708     j_coord_offsetH = 0;
 709
 710     outeriter        = 0;
 711     inneriter        = 0;
 712
 713     for(iidx=0;iidx<4*DIM;iidx++)
 714     {
 715         scratch[iidx] = 0.0;
 716     }
 717
 718     /* Start outer loop over neighborlists */
 719     for(iidx=0; iidx<nri; iidx++)
 720     {
 721         /* Load shift vector for this list */
 722         i_shift_offset   = DIM*shiftidx[iidx];
 723
 724         /* Load limits for loop over neighbors */
 725         j_index_start    = jindex[iidx];
 726         j_index_end      = jindex[iidx+1];
 727
 728         /* Get outer coordinate index */
 729         inr              = iinr[iidx];
 730         i_coord_offset   = DIM*inr;
 731
 732         /* Load i particle coords and add shift vector */
 733         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 734                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 735
 736         fix0             = _mm256_setzero_ps();
 737         fiy0             = _mm256_setzero_ps();
 738         fiz0             = _mm256_setzero_ps();
 739         fix1             = _mm256_setzero_ps();
 740         fiy1             = _mm256_setzero_ps();
 741         fiz1             = _mm256_setzero_ps();
 742         fix2             = _mm256_setzero_ps();
 743         fiy2             = _mm256_setzero_ps();
 744         fiz2             = _mm256_setzero_ps();
 745
 746         /* Start inner kernel loop */
 747         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 748         {
 749
 750             /* Get j neighbor index, and coordinate index */
 751             jnrA             = jjnr[jidx];
 752             jnrB             = jjnr[jidx+1];
 753             jnrC             = jjnr[jidx+2];
 754             jnrD             = jjnr[jidx+3];
 755             jnrE             = jjnr[jidx+4];
 756             jnrF             = jjnr[jidx+5];
 757             jnrG             = jjnr[jidx+6];
 758             jnrH             = jjnr[jidx+7];
 759             j_coord_offsetA  = DIM*jnrA;
 760             j_coord_offsetB  = DIM*jnrB;
 761             j_coord_offsetC  = DIM*jnrC;
 762             j_coord_offsetD  = DIM*jnrD;
 763             j_coord_offsetE  = DIM*jnrE;
 764             j_coord_offsetF  = DIM*jnrF;
 765             j_coord_offsetG  = DIM*jnrG;
 766             j_coord_offsetH  = DIM*jnrH;
 767
 768             /* load j atom coordinates */
 769             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 770                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 771                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 772                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 773                                                  &jx0,&jy0,&jz0);
 774
 775             /* Calculate displacement vector */
 776             dx00             = _mm256_sub_ps(ix0,jx0);
 777             dy00             = _mm256_sub_ps(iy0,jy0);
 778             dz00             = _mm256_sub_ps(iz0,jz0);
 779             dx10             = _mm256_sub_ps(ix1,jx0);
 780             dy10             = _mm256_sub_ps(iy1,jy0);
 781             dz10             = _mm256_sub_ps(iz1,jz0);
 782             dx20             = _mm256_sub_ps(ix2,jx0);
 783             dy20             = _mm256_sub_ps(iy2,jy0);
 784             dz20             = _mm256_sub_ps(iz2,jz0);
 785
 786             /* Calculate squared distance and things based on it */
 787             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 788             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 789             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 790
 791             rinv00           = avx256_invsqrt_f(rsq00);
 792             rinv10           = avx256_invsqrt_f(rsq10);
 793             rinv20           = avx256_invsqrt_f(rsq20);
 794
 795             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 796             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 797             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 798
 799             /* Load parameters for j particles */
 800             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 801                                                                  charge+jnrC+0,charge+jnrD+0,
 802                                                                  charge+jnrE+0,charge+jnrF+0,
 803                                                                  charge+jnrG+0,charge+jnrH+0);
 804             vdwjidx0A        = 2*vdwtype[jnrA+0];
 805             vdwjidx0B        = 2*vdwtype[jnrB+0];
 806             vdwjidx0C        = 2*vdwtype[jnrC+0];
 807             vdwjidx0D        = 2*vdwtype[jnrD+0];
 808             vdwjidx0E        = 2*vdwtype[jnrE+0];
 809             vdwjidx0F        = 2*vdwtype[jnrF+0];
 810             vdwjidx0G        = 2*vdwtype[jnrG+0];
 811             vdwjidx0H        = 2*vdwtype[jnrH+0];
 812
 813             fjx0             = _mm256_setzero_ps();
 814             fjy0             = _mm256_setzero_ps();
 815             fjz0             = _mm256_setzero_ps();
 816
 817             /**************************
 818              * CALCULATE INTERACTIONS *
 819              **************************/
 820
 821             /* Compute parameters for interactions between i and j atoms */
 822             qq00             = _mm256_mul_ps(iq0,jq0);
 823             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 824                                             vdwioffsetptr0+vdwjidx0B,
 825                                             vdwioffsetptr0+vdwjidx0C,
 826                                             vdwioffsetptr0+vdwjidx0D,
 827                                             vdwioffsetptr0+vdwjidx0E,
 828                                             vdwioffsetptr0+vdwjidx0F,
 829                                             vdwioffsetptr0+vdwjidx0G,
 830                                             vdwioffsetptr0+vdwjidx0H,
 831                                             &c6_00,&c12_00);
 832
 833             /* COULOMB ELECTROSTATICS */
 834             velec            = _mm256_mul_ps(qq00,rinv00);
 835             felec            = _mm256_mul_ps(velec,rinvsq00);
 836
 837             /* LENNARD-JONES DISPERSION/REPULSION */
 838
 839             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 840             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 841
 842             fscal            = _mm256_add_ps(felec,fvdw);
 843
 844             /* Calculate temporary vectorial force */
 845             tx               = _mm256_mul_ps(fscal,dx00);
 846             ty               = _mm256_mul_ps(fscal,dy00);
 847             tz               = _mm256_mul_ps(fscal,dz00);
 848
 849             /* Update vectorial force */
 850             fix0             = _mm256_add_ps(fix0,tx);
 851             fiy0             = _mm256_add_ps(fiy0,ty);
 852             fiz0             = _mm256_add_ps(fiz0,tz);
 853
 854             fjx0             = _mm256_add_ps(fjx0,tx);
 855             fjy0             = _mm256_add_ps(fjy0,ty);
 856             fjz0             = _mm256_add_ps(fjz0,tz);
 857
 858             /**************************
 859              * CALCULATE INTERACTIONS *
 860              **************************/
 861
 862             /* Compute parameters for interactions between i and j atoms */
 863             qq10             = _mm256_mul_ps(iq1,jq0);
 864
 865             /* COULOMB ELECTROSTATICS */
 866             velec            = _mm256_mul_ps(qq10,rinv10);
 867             felec            = _mm256_mul_ps(velec,rinvsq10);
 868
 869             fscal            = felec;
 870
 871             /* Calculate temporary vectorial force */
 872             tx               = _mm256_mul_ps(fscal,dx10);
 873             ty               = _mm256_mul_ps(fscal,dy10);
 874             tz               = _mm256_mul_ps(fscal,dz10);
 875
 876             /* Update vectorial force */
 877             fix1             = _mm256_add_ps(fix1,tx);
 878             fiy1             = _mm256_add_ps(fiy1,ty);
 879             fiz1             = _mm256_add_ps(fiz1,tz);
 880
 881             fjx0             = _mm256_add_ps(fjx0,tx);
 882             fjy0             = _mm256_add_ps(fjy0,ty);
 883             fjz0             = _mm256_add_ps(fjz0,tz);
 884
 885             /**************************
 886              * CALCULATE INTERACTIONS *
 887              **************************/
 888
 889             /* Compute parameters for interactions between i and j atoms */
 890             qq20             = _mm256_mul_ps(iq2,jq0);
 891
 892             /* COULOMB ELECTROSTATICS */
 893             velec            = _mm256_mul_ps(qq20,rinv20);
 894             felec            = _mm256_mul_ps(velec,rinvsq20);
 895
 896             fscal            = felec;
 897
 898             /* Calculate temporary vectorial force */
 899             tx               = _mm256_mul_ps(fscal,dx20);
 900             ty               = _mm256_mul_ps(fscal,dy20);
 901             tz               = _mm256_mul_ps(fscal,dz20);
 902
 903             /* Update vectorial force */
 904             fix2             = _mm256_add_ps(fix2,tx);
 905             fiy2             = _mm256_add_ps(fiy2,ty);
 906             fiz2             = _mm256_add_ps(fiz2,tz);
 907
 908             fjx0             = _mm256_add_ps(fjx0,tx);
 909             fjy0             = _mm256_add_ps(fjy0,ty);
 910             fjz0             = _mm256_add_ps(fjz0,tz);
 911
 912             fjptrA             = f+j_coord_offsetA;
 913             fjptrB             = f+j_coord_offsetB;
 914             fjptrC             = f+j_coord_offsetC;
 915             fjptrD             = f+j_coord_offsetD;
 916             fjptrE             = f+j_coord_offsetE;
 917             fjptrF             = f+j_coord_offsetF;
 918             fjptrG             = f+j_coord_offsetG;
 919             fjptrH             = f+j_coord_offsetH;
 920
 921             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 922
 923             /* Inner loop uses 88 flops */
 924         }
 925
 926         if(jidx<j_index_end)
 927         {
 928
 929             /* Get j neighbor index, and coordinate index */
 930             jnrlistA         = jjnr[jidx];
 931             jnrlistB         = jjnr[jidx+1];
 932             jnrlistC         = jjnr[jidx+2];
 933             jnrlistD         = jjnr[jidx+3];
 934             jnrlistE         = jjnr[jidx+4];
 935             jnrlistF         = jjnr[jidx+5];
 936             jnrlistG         = jjnr[jidx+6];
 937             jnrlistH         = jjnr[jidx+7];
 938             /* Sign of each element will be negative for non-real atoms.
 939              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 940              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 941              */
 942             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 943                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 944
 945             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 946             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 947             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 948             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 949             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 950             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 951             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 952             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 953             j_coord_offsetA  = DIM*jnrA;
 954             j_coord_offsetB  = DIM*jnrB;
 955             j_coord_offsetC  = DIM*jnrC;
 956             j_coord_offsetD  = DIM*jnrD;
 957             j_coord_offsetE  = DIM*jnrE;
 958             j_coord_offsetF  = DIM*jnrF;
 959             j_coord_offsetG  = DIM*jnrG;
 960             j_coord_offsetH  = DIM*jnrH;
 961
 962             /* load j atom coordinates */
 963             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 964                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 965                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 966                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 967                                                  &jx0,&jy0,&jz0);
 968
 969             /* Calculate displacement vector */
 970             dx00             = _mm256_sub_ps(ix0,jx0);
 971             dy00             = _mm256_sub_ps(iy0,jy0);
 972             dz00             = _mm256_sub_ps(iz0,jz0);
 973             dx10             = _mm256_sub_ps(ix1,jx0);
 974             dy10             = _mm256_sub_ps(iy1,jy0);
 975             dz10             = _mm256_sub_ps(iz1,jz0);
 976             dx20             = _mm256_sub_ps(ix2,jx0);
 977             dy20             = _mm256_sub_ps(iy2,jy0);
 978             dz20             = _mm256_sub_ps(iz2,jz0);
 979
 980             /* Calculate squared distance and things based on it */
 981             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 982             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 983             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 984
 985             rinv00           = avx256_invsqrt_f(rsq00);
 986             rinv10           = avx256_invsqrt_f(rsq10);
 987             rinv20           = avx256_invsqrt_f(rsq20);
 988
 989             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 990             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 991             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 992
 993             /* Load parameters for j particles */
 994             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 995                                                                  charge+jnrC+0,charge+jnrD+0,
 996                                                                  charge+jnrE+0,charge+jnrF+0,
 997                                                                  charge+jnrG+0,charge+jnrH+0);
 998             vdwjidx0A        = 2*vdwtype[jnrA+0];
 999             vdwjidx0B        = 2*vdwtype[jnrB+0];
1000             vdwjidx0C        = 2*vdwtype[jnrC+0];
1001             vdwjidx0D        = 2*vdwtype[jnrD+0];
1002             vdwjidx0E        = 2*vdwtype[jnrE+0];
1003             vdwjidx0F        = 2*vdwtype[jnrF+0];
1004             vdwjidx0G        = 2*vdwtype[jnrG+0];
1005             vdwjidx0H        = 2*vdwtype[jnrH+0];
1006
1007             fjx0             = _mm256_setzero_ps();
1008             fjy0             = _mm256_setzero_ps();
1009             fjz0             = _mm256_setzero_ps();
1010
1011             /**************************
1012              * CALCULATE INTERACTIONS *
1013              **************************/
1014
1015             /* Compute parameters for interactions between i and j atoms */
1016             qq00             = _mm256_mul_ps(iq0,jq0);
1017             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1018                                             vdwioffsetptr0+vdwjidx0B,
1019                                             vdwioffsetptr0+vdwjidx0C,
1020                                             vdwioffsetptr0+vdwjidx0D,
1021                                             vdwioffsetptr0+vdwjidx0E,
1022                                             vdwioffsetptr0+vdwjidx0F,
1023                                             vdwioffsetptr0+vdwjidx0G,
1024                                             vdwioffsetptr0+vdwjidx0H,
1025                                             &c6_00,&c12_00);
1026
1027             /* COULOMB ELECTROSTATICS */
1028             velec            = _mm256_mul_ps(qq00,rinv00);
1029             felec            = _mm256_mul_ps(velec,rinvsq00);
1030
1031             /* LENNARD-JONES DISPERSION/REPULSION */
1032
1033             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1034             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1035
1036             fscal            = _mm256_add_ps(felec,fvdw);
1037
1038             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1039
1040             /* Calculate temporary vectorial force */
1041             tx               = _mm256_mul_ps(fscal,dx00);
1042             ty               = _mm256_mul_ps(fscal,dy00);
1043             tz               = _mm256_mul_ps(fscal,dz00);
1044
1045             /* Update vectorial force */
1046             fix0             = _mm256_add_ps(fix0,tx);
1047             fiy0             = _mm256_add_ps(fiy0,ty);
1048             fiz0             = _mm256_add_ps(fiz0,tz);
1049
1050             fjx0             = _mm256_add_ps(fjx0,tx);
1051             fjy0             = _mm256_add_ps(fjy0,ty);
1052             fjz0             = _mm256_add_ps(fjz0,tz);
1053
1054             /**************************
1055              * CALCULATE INTERACTIONS *
1056              **************************/
1057
1058             /* Compute parameters for interactions between i and j atoms */
1059             qq10             = _mm256_mul_ps(iq1,jq0);
1060
1061             /* COULOMB ELECTROSTATICS */
1062             velec            = _mm256_mul_ps(qq10,rinv10);
1063             felec            = _mm256_mul_ps(velec,rinvsq10);
1064
1065             fscal            = felec;
1066
1067             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1068
1069             /* Calculate temporary vectorial force */
1070             tx               = _mm256_mul_ps(fscal,dx10);
1071             ty               = _mm256_mul_ps(fscal,dy10);
1072             tz               = _mm256_mul_ps(fscal,dz10);
1073
1074             /* Update vectorial force */
1075             fix1             = _mm256_add_ps(fix1,tx);
1076             fiy1             = _mm256_add_ps(fiy1,ty);
1077             fiz1             = _mm256_add_ps(fiz1,tz);
1078
1079             fjx0             = _mm256_add_ps(fjx0,tx);
1080             fjy0             = _mm256_add_ps(fjy0,ty);
1081             fjz0             = _mm256_add_ps(fjz0,tz);
1082
1083             /**************************
1084              * CALCULATE INTERACTIONS *
1085              **************************/
1086
1087             /* Compute parameters for interactions between i and j atoms */
1088             qq20             = _mm256_mul_ps(iq2,jq0);
1089
1090             /* COULOMB ELECTROSTATICS */
1091             velec            = _mm256_mul_ps(qq20,rinv20);
1092             felec            = _mm256_mul_ps(velec,rinvsq20);
1093
1094             fscal            = felec;
1095
1096             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1097
1098             /* Calculate temporary vectorial force */
1099             tx               = _mm256_mul_ps(fscal,dx20);
1100             ty               = _mm256_mul_ps(fscal,dy20);
1101             tz               = _mm256_mul_ps(fscal,dz20);
1102
1103             /* Update vectorial force */
1104             fix2             = _mm256_add_ps(fix2,tx);
1105             fiy2             = _mm256_add_ps(fiy2,ty);
1106             fiz2             = _mm256_add_ps(fiz2,tz);
1107
1108             fjx0             = _mm256_add_ps(fjx0,tx);
1109             fjy0             = _mm256_add_ps(fjy0,ty);
1110             fjz0             = _mm256_add_ps(fjz0,tz);
1111
1112             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1113             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1114             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1115             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1116             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1117             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1118             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1119             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1120
1121             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1122
1123             /* Inner loop uses 88 flops */
1124         }
1125
1126         /* End of innermost loop */
1127
1128         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1129                                                  f+i_coord_offset,fshift+i_shift_offset);
1130
1131         /* Increment number of inner iterations */
1132         inneriter                  += j_index_end - j_index_start;
1133
1134         /* Outer loop uses 18 flops */
1135     }
1136
1137     /* Increment number of outer iterations */
1138     outeriter        += nri;
1139
1140     /* Update outer/inner flops */
1141
1142     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*88);
1143 }