src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRF_VdwLJ_GeomW4P1_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_ElecRF_VdwLJ_GeomW4P1_VF_avx_256_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRF_VdwLJ_GeomW4P1_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     real *           vdwioffsetptr3;
  91     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  93     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  96     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  97     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  98     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  99     real             *charge;
 100     int              nvdwtype;
 101     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 102     int              *vdwtype;
 103     real             *vdwparam;
 104     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 105     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 106     __m256d          dummy_mask,cutoff_mask;
 107     __m128           tmpmask0,tmpmask1;
 108     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 109     __m256d          one     = _mm256_set1_pd(1.0);
 110     __m256d          two     = _mm256_set1_pd(2.0);
 111     x                = xx[0];
 112     f                = ff[0];
 113
 114     nri              = nlist->nri;
 115     iinr             = nlist->iinr;
 116     jindex           = nlist->jindex;
 117     jjnr             = nlist->jjnr;
 118     shiftidx         = nlist->shift;
 119     gid              = nlist->gid;
 120     shiftvec         = fr->shift_vec[0];
 121     fshift           = fr->fshift[0];
 122     facel            = _mm256_set1_pd(fr->epsfac);
 123     charge           = mdatoms->chargeA;
 124     krf              = _mm256_set1_pd(fr->ic->k_rf);
 125     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 126     crf              = _mm256_set1_pd(fr->ic->c_rf);
 127     nvdwtype         = fr->ntype;
 128     vdwparam         = fr->nbfp;
 129     vdwtype          = mdatoms->typeA;
 130
 131     /* Setup water-specific parameters */
 132     inr              = nlist->iinr[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     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 136     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 137
 138     /* Avoid stupid compiler warnings */
 139     jnrA = jnrB = jnrC = jnrD = 0;
 140     j_coord_offsetA = 0;
 141     j_coord_offsetB = 0;
 142     j_coord_offsetC = 0;
 143     j_coord_offsetD = 0;
 144
 145     outeriter        = 0;
 146     inneriter        = 0;
 147
 148     for(iidx=0;iidx<4*DIM;iidx++)
 149     {
 150         scratch[iidx] = 0.0;
 151     }
 152
 153     /* Start outer loop over neighborlists */
 154     for(iidx=0; iidx<nri; iidx++)
 155     {
 156         /* Load shift vector for this list */
 157         i_shift_offset   = DIM*shiftidx[iidx];
 158
 159         /* Load limits for loop over neighbors */
 160         j_index_start    = jindex[iidx];
 161         j_index_end      = jindex[iidx+1];
 162
 163         /* Get outer coordinate index */
 164         inr              = iinr[iidx];
 165         i_coord_offset   = DIM*inr;
 166
 167         /* Load i particle coords and add shift vector */
 168         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 169                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 170
 171         fix0             = _mm256_setzero_pd();
 172         fiy0             = _mm256_setzero_pd();
 173         fiz0             = _mm256_setzero_pd();
 174         fix1             = _mm256_setzero_pd();
 175         fiy1             = _mm256_setzero_pd();
 176         fiz1             = _mm256_setzero_pd();
 177         fix2             = _mm256_setzero_pd();
 178         fiy2             = _mm256_setzero_pd();
 179         fiz2             = _mm256_setzero_pd();
 180         fix3             = _mm256_setzero_pd();
 181         fiy3             = _mm256_setzero_pd();
 182         fiz3             = _mm256_setzero_pd();
 183
 184         /* Reset potential sums */
 185         velecsum         = _mm256_setzero_pd();
 186         vvdwsum          = _mm256_setzero_pd();
 187
 188         /* Start inner kernel loop */
 189         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 190         {
 191
 192             /* Get j neighbor index, and coordinate index */
 193             jnrA             = jjnr[jidx];
 194             jnrB             = jjnr[jidx+1];
 195             jnrC             = jjnr[jidx+2];
 196             jnrD             = jjnr[jidx+3];
 197             j_coord_offsetA  = DIM*jnrA;
 198             j_coord_offsetB  = DIM*jnrB;
 199             j_coord_offsetC  = DIM*jnrC;
 200             j_coord_offsetD  = DIM*jnrD;
 201
 202             /* load j atom coordinates */
 203             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 204                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 205                                                  &jx0,&jy0,&jz0);
 206
 207             /* Calculate displacement vector */
 208             dx00             = _mm256_sub_pd(ix0,jx0);
 209             dy00             = _mm256_sub_pd(iy0,jy0);
 210             dz00             = _mm256_sub_pd(iz0,jz0);
 211             dx10             = _mm256_sub_pd(ix1,jx0);
 212             dy10             = _mm256_sub_pd(iy1,jy0);
 213             dz10             = _mm256_sub_pd(iz1,jz0);
 214             dx20             = _mm256_sub_pd(ix2,jx0);
 215             dy20             = _mm256_sub_pd(iy2,jy0);
 216             dz20             = _mm256_sub_pd(iz2,jz0);
 217             dx30             = _mm256_sub_pd(ix3,jx0);
 218             dy30             = _mm256_sub_pd(iy3,jy0);
 219             dz30             = _mm256_sub_pd(iz3,jz0);
 220
 221             /* Calculate squared distance and things based on it */
 222             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 223             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 224             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 225             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 226
 227             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 228             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 229             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 230
 231             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 232             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 233             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 234             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 235
 236             /* Load parameters for j particles */
 237             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 238                                                                  charge+jnrC+0,charge+jnrD+0);
 239             vdwjidx0A        = 2*vdwtype[jnrA+0];
 240             vdwjidx0B        = 2*vdwtype[jnrB+0];
 241             vdwjidx0C        = 2*vdwtype[jnrC+0];
 242             vdwjidx0D        = 2*vdwtype[jnrD+0];
 243
 244             fjx0             = _mm256_setzero_pd();
 245             fjy0             = _mm256_setzero_pd();
 246             fjz0             = _mm256_setzero_pd();
 247
 248             /**************************
 249              * CALCULATE INTERACTIONS *
 250              **************************/
 251
 252             /* Compute parameters for interactions between i and j atoms */
 253             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 254                                             vdwioffsetptr0+vdwjidx0B,
 255                                             vdwioffsetptr0+vdwjidx0C,
 256                                             vdwioffsetptr0+vdwjidx0D,
 257                                             &c6_00,&c12_00);
 258
 259             /* LENNARD-JONES DISPERSION/REPULSION */
 260
 261             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 262             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 263             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 264             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 265             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 266
 267             /* Update potential sum for this i atom from the interaction with this j atom. */
 268             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 269
 270             fscal            = fvdw;
 271
 272             /* Calculate temporary vectorial force */
 273             tx               = _mm256_mul_pd(fscal,dx00);
 274             ty               = _mm256_mul_pd(fscal,dy00);
 275             tz               = _mm256_mul_pd(fscal,dz00);
 276
 277             /* Update vectorial force */
 278             fix0             = _mm256_add_pd(fix0,tx);
 279             fiy0             = _mm256_add_pd(fiy0,ty);
 280             fiz0             = _mm256_add_pd(fiz0,tz);
 281
 282             fjx0             = _mm256_add_pd(fjx0,tx);
 283             fjy0             = _mm256_add_pd(fjy0,ty);
 284             fjz0             = _mm256_add_pd(fjz0,tz);
 285
 286             /**************************
 287              * CALCULATE INTERACTIONS *
 288              **************************/
 289
 290             /* Compute parameters for interactions between i and j atoms */
 291             qq10             = _mm256_mul_pd(iq1,jq0);
 292
 293             /* REACTION-FIELD ELECTROSTATICS */
 294             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 295             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 296
 297             /* Update potential sum for this i atom from the interaction with this j atom. */
 298             velecsum         = _mm256_add_pd(velecsum,velec);
 299
 300             fscal            = felec;
 301
 302             /* Calculate temporary vectorial force */
 303             tx               = _mm256_mul_pd(fscal,dx10);
 304             ty               = _mm256_mul_pd(fscal,dy10);
 305             tz               = _mm256_mul_pd(fscal,dz10);
 306
 307             /* Update vectorial force */
 308             fix1             = _mm256_add_pd(fix1,tx);
 309             fiy1             = _mm256_add_pd(fiy1,ty);
 310             fiz1             = _mm256_add_pd(fiz1,tz);
 311
 312             fjx0             = _mm256_add_pd(fjx0,tx);
 313             fjy0             = _mm256_add_pd(fjy0,ty);
 314             fjz0             = _mm256_add_pd(fjz0,tz);
 315
 316             /**************************
 317              * CALCULATE INTERACTIONS *
 318              **************************/
 319
 320             /* Compute parameters for interactions between i and j atoms */
 321             qq20             = _mm256_mul_pd(iq2,jq0);
 322
 323             /* REACTION-FIELD ELECTROSTATICS */
 324             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 325             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 326
 327             /* Update potential sum for this i atom from the interaction with this j atom. */
 328             velecsum         = _mm256_add_pd(velecsum,velec);
 329
 330             fscal            = felec;
 331
 332             /* Calculate temporary vectorial force */
 333             tx               = _mm256_mul_pd(fscal,dx20);
 334             ty               = _mm256_mul_pd(fscal,dy20);
 335             tz               = _mm256_mul_pd(fscal,dz20);
 336
 337             /* Update vectorial force */
 338             fix2             = _mm256_add_pd(fix2,tx);
 339             fiy2             = _mm256_add_pd(fiy2,ty);
 340             fiz2             = _mm256_add_pd(fiz2,tz);
 341
 342             fjx0             = _mm256_add_pd(fjx0,tx);
 343             fjy0             = _mm256_add_pd(fjy0,ty);
 344             fjz0             = _mm256_add_pd(fjz0,tz);
 345
 346             /**************************
 347              * CALCULATE INTERACTIONS *
 348              **************************/
 349
 350             /* Compute parameters for interactions between i and j atoms */
 351             qq30             = _mm256_mul_pd(iq3,jq0);
 352
 353             /* REACTION-FIELD ELECTROSTATICS */
 354             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
 355             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 356
 357             /* Update potential sum for this i atom from the interaction with this j atom. */
 358             velecsum         = _mm256_add_pd(velecsum,velec);
 359
 360             fscal            = felec;
 361
 362             /* Calculate temporary vectorial force */
 363             tx               = _mm256_mul_pd(fscal,dx30);
 364             ty               = _mm256_mul_pd(fscal,dy30);
 365             tz               = _mm256_mul_pd(fscal,dz30);
 366
 367             /* Update vectorial force */
 368             fix3             = _mm256_add_pd(fix3,tx);
 369             fiy3             = _mm256_add_pd(fiy3,ty);
 370             fiz3             = _mm256_add_pd(fiz3,tz);
 371
 372             fjx0             = _mm256_add_pd(fjx0,tx);
 373             fjy0             = _mm256_add_pd(fjy0,ty);
 374             fjz0             = _mm256_add_pd(fjz0,tz);
 375
 376             fjptrA             = f+j_coord_offsetA;
 377             fjptrB             = f+j_coord_offsetB;
 378             fjptrC             = f+j_coord_offsetC;
 379             fjptrD             = f+j_coord_offsetD;
 380
 381             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 382
 383             /* Inner loop uses 131 flops */
 384         }
 385
 386         if(jidx<j_index_end)
 387         {
 388
 389             /* Get j neighbor index, and coordinate index */
 390             jnrlistA         = jjnr[jidx];
 391             jnrlistB         = jjnr[jidx+1];
 392             jnrlistC         = jjnr[jidx+2];
 393             jnrlistD         = jjnr[jidx+3];
 394             /* Sign of each element will be negative for non-real atoms.
 395              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 396              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 397              */
 398             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 399
 400             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 401             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 402             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 403
 404             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 405             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 406             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 407             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 408             j_coord_offsetA  = DIM*jnrA;
 409             j_coord_offsetB  = DIM*jnrB;
 410             j_coord_offsetC  = DIM*jnrC;
 411             j_coord_offsetD  = DIM*jnrD;
 412
 413             /* load j atom coordinates */
 414             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 415                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 416                                                  &jx0,&jy0,&jz0);
 417
 418             /* Calculate displacement vector */
 419             dx00             = _mm256_sub_pd(ix0,jx0);
 420             dy00             = _mm256_sub_pd(iy0,jy0);
 421             dz00             = _mm256_sub_pd(iz0,jz0);
 422             dx10             = _mm256_sub_pd(ix1,jx0);
 423             dy10             = _mm256_sub_pd(iy1,jy0);
 424             dz10             = _mm256_sub_pd(iz1,jz0);
 425             dx20             = _mm256_sub_pd(ix2,jx0);
 426             dy20             = _mm256_sub_pd(iy2,jy0);
 427             dz20             = _mm256_sub_pd(iz2,jz0);
 428             dx30             = _mm256_sub_pd(ix3,jx0);
 429             dy30             = _mm256_sub_pd(iy3,jy0);
 430             dz30             = _mm256_sub_pd(iz3,jz0);
 431
 432             /* Calculate squared distance and things based on it */
 433             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 434             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 435             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 436             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 437
 438             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 439             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 440             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 441
 442             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 443             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 444             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 445             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 446
 447             /* Load parameters for j particles */
 448             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 449                                                                  charge+jnrC+0,charge+jnrD+0);
 450             vdwjidx0A        = 2*vdwtype[jnrA+0];
 451             vdwjidx0B        = 2*vdwtype[jnrB+0];
 452             vdwjidx0C        = 2*vdwtype[jnrC+0];
 453             vdwjidx0D        = 2*vdwtype[jnrD+0];
 454
 455             fjx0             = _mm256_setzero_pd();
 456             fjy0             = _mm256_setzero_pd();
 457             fjz0             = _mm256_setzero_pd();
 458
 459             /**************************
 460              * CALCULATE INTERACTIONS *
 461              **************************/
 462
 463             /* Compute parameters for interactions between i and j atoms */
 464             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 465                                             vdwioffsetptr0+vdwjidx0B,
 466                                             vdwioffsetptr0+vdwjidx0C,
 467                                             vdwioffsetptr0+vdwjidx0D,
 468                                             &c6_00,&c12_00);
 469
 470             /* LENNARD-JONES DISPERSION/REPULSION */
 471
 472             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 473             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 474             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 475             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 476             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 477
 478             /* Update potential sum for this i atom from the interaction with this j atom. */
 479             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 480             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 481
 482             fscal            = fvdw;
 483
 484             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 485
 486             /* Calculate temporary vectorial force */
 487             tx               = _mm256_mul_pd(fscal,dx00);
 488             ty               = _mm256_mul_pd(fscal,dy00);
 489             tz               = _mm256_mul_pd(fscal,dz00);
 490
 491             /* Update vectorial force */
 492             fix0             = _mm256_add_pd(fix0,tx);
 493             fiy0             = _mm256_add_pd(fiy0,ty);
 494             fiz0             = _mm256_add_pd(fiz0,tz);
 495
 496             fjx0             = _mm256_add_pd(fjx0,tx);
 497             fjy0             = _mm256_add_pd(fjy0,ty);
 498             fjz0             = _mm256_add_pd(fjz0,tz);
 499
 500             /**************************
 501              * CALCULATE INTERACTIONS *
 502              **************************/
 503
 504             /* Compute parameters for interactions between i and j atoms */
 505             qq10             = _mm256_mul_pd(iq1,jq0);
 506
 507             /* REACTION-FIELD ELECTROSTATICS */
 508             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 509             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 510
 511             /* Update potential sum for this i atom from the interaction with this j atom. */
 512             velec            = _mm256_andnot_pd(dummy_mask,velec);
 513             velecsum         = _mm256_add_pd(velecsum,velec);
 514
 515             fscal            = felec;
 516
 517             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 518
 519             /* Calculate temporary vectorial force */
 520             tx               = _mm256_mul_pd(fscal,dx10);
 521             ty               = _mm256_mul_pd(fscal,dy10);
 522             tz               = _mm256_mul_pd(fscal,dz10);
 523
 524             /* Update vectorial force */
 525             fix1             = _mm256_add_pd(fix1,tx);
 526             fiy1             = _mm256_add_pd(fiy1,ty);
 527             fiz1             = _mm256_add_pd(fiz1,tz);
 528
 529             fjx0             = _mm256_add_pd(fjx0,tx);
 530             fjy0             = _mm256_add_pd(fjy0,ty);
 531             fjz0             = _mm256_add_pd(fjz0,tz);
 532
 533             /**************************
 534              * CALCULATE INTERACTIONS *
 535              **************************/
 536
 537             /* Compute parameters for interactions between i and j atoms */
 538             qq20             = _mm256_mul_pd(iq2,jq0);
 539
 540             /* REACTION-FIELD ELECTROSTATICS */
 541             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 542             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 543
 544             /* Update potential sum for this i atom from the interaction with this j atom. */
 545             velec            = _mm256_andnot_pd(dummy_mask,velec);
 546             velecsum         = _mm256_add_pd(velecsum,velec);
 547
 548             fscal            = felec;
 549
 550             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 551
 552             /* Calculate temporary vectorial force */
 553             tx               = _mm256_mul_pd(fscal,dx20);
 554             ty               = _mm256_mul_pd(fscal,dy20);
 555             tz               = _mm256_mul_pd(fscal,dz20);
 556
 557             /* Update vectorial force */
 558             fix2             = _mm256_add_pd(fix2,tx);
 559             fiy2             = _mm256_add_pd(fiy2,ty);
 560             fiz2             = _mm256_add_pd(fiz2,tz);
 561
 562             fjx0             = _mm256_add_pd(fjx0,tx);
 563             fjy0             = _mm256_add_pd(fjy0,ty);
 564             fjz0             = _mm256_add_pd(fjz0,tz);
 565
 566             /**************************
 567              * CALCULATE INTERACTIONS *
 568              **************************/
 569
 570             /* Compute parameters for interactions between i and j atoms */
 571             qq30             = _mm256_mul_pd(iq3,jq0);
 572
 573             /* REACTION-FIELD ELECTROSTATICS */
 574             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
 575             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 576
 577             /* Update potential sum for this i atom from the interaction with this j atom. */
 578             velec            = _mm256_andnot_pd(dummy_mask,velec);
 579             velecsum         = _mm256_add_pd(velecsum,velec);
 580
 581             fscal            = felec;
 582
 583             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 584
 585             /* Calculate temporary vectorial force */
 586             tx               = _mm256_mul_pd(fscal,dx30);
 587             ty               = _mm256_mul_pd(fscal,dy30);
 588             tz               = _mm256_mul_pd(fscal,dz30);
 589
 590             /* Update vectorial force */
 591             fix3             = _mm256_add_pd(fix3,tx);
 592             fiy3             = _mm256_add_pd(fiy3,ty);
 593             fiz3             = _mm256_add_pd(fiz3,tz);
 594
 595             fjx0             = _mm256_add_pd(fjx0,tx);
 596             fjy0             = _mm256_add_pd(fjy0,ty);
 597             fjz0             = _mm256_add_pd(fjz0,tz);
 598
 599             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 600             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 601             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 602             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 603
 604             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 605
 606             /* Inner loop uses 131 flops */
 607         }
 608
 609         /* End of innermost loop */
 610
 611         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 612                                                  f+i_coord_offset,fshift+i_shift_offset);
 613
 614         ggid                        = gid[iidx];
 615         /* Update potential energies */
 616         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 617         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 618
 619         /* Increment number of inner iterations */
 620         inneriter                  += j_index_end - j_index_start;
 621
 622         /* Outer loop uses 26 flops */
 623     }
 624
 625     /* Increment number of outer iterations */
 626     outeriter        += nri;
 627
 628     /* Update outer/inner flops */
 629
 630     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*131);
 631 }
 632 /*
 633  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW4P1_F_avx_256_double
 634  * Electrostatics interaction: ReactionField
 635  * VdW interaction:            LennardJones
 636  * Geometry:                   Water4-Particle
 637  * Calculate force/pot:        Force
 638  */
 639 void
 640 nb_kernel_ElecRF_VdwLJ_GeomW4P1_F_avx_256_double
 641                     (t_nblist                    * gmx_restrict       nlist,
 642                      rvec                        * gmx_restrict          xx,
 643                      rvec                        * gmx_restrict          ff,
 644                      t_forcerec                  * gmx_restrict          fr,
 645                      t_mdatoms                   * gmx_restrict     mdatoms,
 646                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 647                      t_nrnb                      * gmx_restrict        nrnb)
 648 {
 649     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 650      * just 0 for non-waters.
 651      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 652      * jnr indices corresponding to data put in the four positions in the SIMD register.
 653      */
 654     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 655     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 656     int              jnrA,jnrB,jnrC,jnrD;
 657     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 658     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 659     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 660     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 661     real             rcutoff_scalar;
 662     real             *shiftvec,*fshift,*x,*f;
 663     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 664     real             scratch[4*DIM];
 665     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 666     real *           vdwioffsetptr0;
 667     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 668     real *           vdwioffsetptr1;
 669     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 670     real *           vdwioffsetptr2;
 671     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 672     real *           vdwioffsetptr3;
 673     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 674     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 675     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 676     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 677     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 678     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 679     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 680     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 681     real             *charge;
 682     int              nvdwtype;
 683     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 684     int              *vdwtype;
 685     real             *vdwparam;
 686     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 687     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 688     __m256d          dummy_mask,cutoff_mask;
 689     __m128           tmpmask0,tmpmask1;
 690     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 691     __m256d          one     = _mm256_set1_pd(1.0);
 692     __m256d          two     = _mm256_set1_pd(2.0);
 693     x                = xx[0];
 694     f                = ff[0];
 695
 696     nri              = nlist->nri;
 697     iinr             = nlist->iinr;
 698     jindex           = nlist->jindex;
 699     jjnr             = nlist->jjnr;
 700     shiftidx         = nlist->shift;
 701     gid              = nlist->gid;
 702     shiftvec         = fr->shift_vec[0];
 703     fshift           = fr->fshift[0];
 704     facel            = _mm256_set1_pd(fr->epsfac);
 705     charge           = mdatoms->chargeA;
 706     krf              = _mm256_set1_pd(fr->ic->k_rf);
 707     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 708     crf              = _mm256_set1_pd(fr->ic->c_rf);
 709     nvdwtype         = fr->ntype;
 710     vdwparam         = fr->nbfp;
 711     vdwtype          = mdatoms->typeA;
 712
 713     /* Setup water-specific parameters */
 714     inr              = nlist->iinr[0];
 715     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 716     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 717     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 718     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 719
 720     /* Avoid stupid compiler warnings */
 721     jnrA = jnrB = jnrC = jnrD = 0;
 722     j_coord_offsetA = 0;
 723     j_coord_offsetB = 0;
 724     j_coord_offsetC = 0;
 725     j_coord_offsetD = 0;
 726
 727     outeriter        = 0;
 728     inneriter        = 0;
 729
 730     for(iidx=0;iidx<4*DIM;iidx++)
 731     {
 732         scratch[iidx] = 0.0;
 733     }
 734
 735     /* Start outer loop over neighborlists */
 736     for(iidx=0; iidx<nri; iidx++)
 737     {
 738         /* Load shift vector for this list */
 739         i_shift_offset   = DIM*shiftidx[iidx];
 740
 741         /* Load limits for loop over neighbors */
 742         j_index_start    = jindex[iidx];
 743         j_index_end      = jindex[iidx+1];
 744
 745         /* Get outer coordinate index */
 746         inr              = iinr[iidx];
 747         i_coord_offset   = DIM*inr;
 748
 749         /* Load i particle coords and add shift vector */
 750         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 751                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 752
 753         fix0             = _mm256_setzero_pd();
 754         fiy0             = _mm256_setzero_pd();
 755         fiz0             = _mm256_setzero_pd();
 756         fix1             = _mm256_setzero_pd();
 757         fiy1             = _mm256_setzero_pd();
 758         fiz1             = _mm256_setzero_pd();
 759         fix2             = _mm256_setzero_pd();
 760         fiy2             = _mm256_setzero_pd();
 761         fiz2             = _mm256_setzero_pd();
 762         fix3             = _mm256_setzero_pd();
 763         fiy3             = _mm256_setzero_pd();
 764         fiz3             = _mm256_setzero_pd();
 765
 766         /* Start inner kernel loop */
 767         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 768         {
 769
 770             /* Get j neighbor index, and coordinate index */
 771             jnrA             = jjnr[jidx];
 772             jnrB             = jjnr[jidx+1];
 773             jnrC             = jjnr[jidx+2];
 774             jnrD             = jjnr[jidx+3];
 775             j_coord_offsetA  = DIM*jnrA;
 776             j_coord_offsetB  = DIM*jnrB;
 777             j_coord_offsetC  = DIM*jnrC;
 778             j_coord_offsetD  = DIM*jnrD;
 779
 780             /* load j atom coordinates */
 781             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 782                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 783                                                  &jx0,&jy0,&jz0);
 784
 785             /* Calculate displacement vector */
 786             dx00             = _mm256_sub_pd(ix0,jx0);
 787             dy00             = _mm256_sub_pd(iy0,jy0);
 788             dz00             = _mm256_sub_pd(iz0,jz0);
 789             dx10             = _mm256_sub_pd(ix1,jx0);
 790             dy10             = _mm256_sub_pd(iy1,jy0);
 791             dz10             = _mm256_sub_pd(iz1,jz0);
 792             dx20             = _mm256_sub_pd(ix2,jx0);
 793             dy20             = _mm256_sub_pd(iy2,jy0);
 794             dz20             = _mm256_sub_pd(iz2,jz0);
 795             dx30             = _mm256_sub_pd(ix3,jx0);
 796             dy30             = _mm256_sub_pd(iy3,jy0);
 797             dz30             = _mm256_sub_pd(iz3,jz0);
 798
 799             /* Calculate squared distance and things based on it */
 800             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 801             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 802             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 803             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 804
 805             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 806             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 807             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 808
 809             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 810             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 811             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 812             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 813
 814             /* Load parameters for j particles */
 815             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 816                                                                  charge+jnrC+0,charge+jnrD+0);
 817             vdwjidx0A        = 2*vdwtype[jnrA+0];
 818             vdwjidx0B        = 2*vdwtype[jnrB+0];
 819             vdwjidx0C        = 2*vdwtype[jnrC+0];
 820             vdwjidx0D        = 2*vdwtype[jnrD+0];
 821
 822             fjx0             = _mm256_setzero_pd();
 823             fjy0             = _mm256_setzero_pd();
 824             fjz0             = _mm256_setzero_pd();
 825
 826             /**************************
 827              * CALCULATE INTERACTIONS *
 828              **************************/
 829
 830             /* Compute parameters for interactions between i and j atoms */
 831             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 832                                             vdwioffsetptr0+vdwjidx0B,
 833                                             vdwioffsetptr0+vdwjidx0C,
 834                                             vdwioffsetptr0+vdwjidx0D,
 835                                             &c6_00,&c12_00);
 836
 837             /* LENNARD-JONES DISPERSION/REPULSION */
 838
 839             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 840             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 841
 842             fscal            = fvdw;
 843
 844             /* Calculate temporary vectorial force */
 845             tx               = _mm256_mul_pd(fscal,dx00);
 846             ty               = _mm256_mul_pd(fscal,dy00);
 847             tz               = _mm256_mul_pd(fscal,dz00);
 848
 849             /* Update vectorial force */
 850             fix0             = _mm256_add_pd(fix0,tx);
 851             fiy0             = _mm256_add_pd(fiy0,ty);
 852             fiz0             = _mm256_add_pd(fiz0,tz);
 853
 854             fjx0             = _mm256_add_pd(fjx0,tx);
 855             fjy0             = _mm256_add_pd(fjy0,ty);
 856             fjz0             = _mm256_add_pd(fjz0,tz);
 857
 858             /**************************
 859              * CALCULATE INTERACTIONS *
 860              **************************/
 861
 862             /* Compute parameters for interactions between i and j atoms */
 863             qq10             = _mm256_mul_pd(iq1,jq0);
 864
 865             /* REACTION-FIELD ELECTROSTATICS */
 866             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 867
 868             fscal            = felec;
 869
 870             /* Calculate temporary vectorial force */
 871             tx               = _mm256_mul_pd(fscal,dx10);
 872             ty               = _mm256_mul_pd(fscal,dy10);
 873             tz               = _mm256_mul_pd(fscal,dz10);
 874
 875             /* Update vectorial force */
 876             fix1             = _mm256_add_pd(fix1,tx);
 877             fiy1             = _mm256_add_pd(fiy1,ty);
 878             fiz1             = _mm256_add_pd(fiz1,tz);
 879
 880             fjx0             = _mm256_add_pd(fjx0,tx);
 881             fjy0             = _mm256_add_pd(fjy0,ty);
 882             fjz0             = _mm256_add_pd(fjz0,tz);
 883
 884             /**************************
 885              * CALCULATE INTERACTIONS *
 886              **************************/
 887
 888             /* Compute parameters for interactions between i and j atoms */
 889             qq20             = _mm256_mul_pd(iq2,jq0);
 890
 891             /* REACTION-FIELD ELECTROSTATICS */
 892             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 893
 894             fscal            = felec;
 895
 896             /* Calculate temporary vectorial force */
 897             tx               = _mm256_mul_pd(fscal,dx20);
 898             ty               = _mm256_mul_pd(fscal,dy20);
 899             tz               = _mm256_mul_pd(fscal,dz20);
 900
 901             /* Update vectorial force */
 902             fix2             = _mm256_add_pd(fix2,tx);
 903             fiy2             = _mm256_add_pd(fiy2,ty);
 904             fiz2             = _mm256_add_pd(fiz2,tz);
 905
 906             fjx0             = _mm256_add_pd(fjx0,tx);
 907             fjy0             = _mm256_add_pd(fjy0,ty);
 908             fjz0             = _mm256_add_pd(fjz0,tz);
 909
 910             /**************************
 911              * CALCULATE INTERACTIONS *
 912              **************************/
 913
 914             /* Compute parameters for interactions between i and j atoms */
 915             qq30             = _mm256_mul_pd(iq3,jq0);
 916
 917             /* REACTION-FIELD ELECTROSTATICS */
 918             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 919
 920             fscal            = felec;
 921
 922             /* Calculate temporary vectorial force */
 923             tx               = _mm256_mul_pd(fscal,dx30);
 924             ty               = _mm256_mul_pd(fscal,dy30);
 925             tz               = _mm256_mul_pd(fscal,dz30);
 926
 927             /* Update vectorial force */
 928             fix3             = _mm256_add_pd(fix3,tx);
 929             fiy3             = _mm256_add_pd(fiy3,ty);
 930             fiz3             = _mm256_add_pd(fiz3,tz);
 931
 932             fjx0             = _mm256_add_pd(fjx0,tx);
 933             fjy0             = _mm256_add_pd(fjy0,ty);
 934             fjz0             = _mm256_add_pd(fjz0,tz);
 935
 936             fjptrA             = f+j_coord_offsetA;
 937             fjptrB             = f+j_coord_offsetB;
 938             fjptrC             = f+j_coord_offsetC;
 939             fjptrD             = f+j_coord_offsetD;
 940
 941             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 942
 943             /* Inner loop uses 111 flops */
 944         }
 945
 946         if(jidx<j_index_end)
 947         {
 948
 949             /* Get j neighbor index, and coordinate index */
 950             jnrlistA         = jjnr[jidx];
 951             jnrlistB         = jjnr[jidx+1];
 952             jnrlistC         = jjnr[jidx+2];
 953             jnrlistD         = jjnr[jidx+3];
 954             /* Sign of each element will be negative for non-real atoms.
 955              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 956              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 957              */
 958             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 959
 960             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 961             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 962             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 963
 964             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 965             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 966             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 967             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 968             j_coord_offsetA  = DIM*jnrA;
 969             j_coord_offsetB  = DIM*jnrB;
 970             j_coord_offsetC  = DIM*jnrC;
 971             j_coord_offsetD  = DIM*jnrD;
 972
 973             /* load j atom coordinates */
 974             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 975                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 976                                                  &jx0,&jy0,&jz0);
 977
 978             /* Calculate displacement vector */
 979             dx00             = _mm256_sub_pd(ix0,jx0);
 980             dy00             = _mm256_sub_pd(iy0,jy0);
 981             dz00             = _mm256_sub_pd(iz0,jz0);
 982             dx10             = _mm256_sub_pd(ix1,jx0);
 983             dy10             = _mm256_sub_pd(iy1,jy0);
 984             dz10             = _mm256_sub_pd(iz1,jz0);
 985             dx20             = _mm256_sub_pd(ix2,jx0);
 986             dy20             = _mm256_sub_pd(iy2,jy0);
 987             dz20             = _mm256_sub_pd(iz2,jz0);
 988             dx30             = _mm256_sub_pd(ix3,jx0);
 989             dy30             = _mm256_sub_pd(iy3,jy0);
 990             dz30             = _mm256_sub_pd(iz3,jz0);
 991
 992             /* Calculate squared distance and things based on it */
 993             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 994             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 995             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 996             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 997
 998             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 999             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1000             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1001
1002             rinvsq00         = gmx_mm256_inv_pd(rsq00);
1003             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1004             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1005             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1006
1007             /* Load parameters for j particles */
1008             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1009                                                                  charge+jnrC+0,charge+jnrD+0);
1010             vdwjidx0A        = 2*vdwtype[jnrA+0];
1011             vdwjidx0B        = 2*vdwtype[jnrB+0];
1012             vdwjidx0C        = 2*vdwtype[jnrC+0];
1013             vdwjidx0D        = 2*vdwtype[jnrD+0];
1014
1015             fjx0             = _mm256_setzero_pd();
1016             fjy0             = _mm256_setzero_pd();
1017             fjz0             = _mm256_setzero_pd();
1018
1019             /**************************
1020              * CALCULATE INTERACTIONS *
1021              **************************/
1022
1023             /* Compute parameters for interactions between i and j atoms */
1024             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1025                                             vdwioffsetptr0+vdwjidx0B,
1026                                             vdwioffsetptr0+vdwjidx0C,
1027                                             vdwioffsetptr0+vdwjidx0D,
1028                                             &c6_00,&c12_00);
1029
1030             /* LENNARD-JONES DISPERSION/REPULSION */
1031
1032             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1033             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1034
1035             fscal            = fvdw;
1036
1037             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1038
1039             /* Calculate temporary vectorial force */
1040             tx               = _mm256_mul_pd(fscal,dx00);
1041             ty               = _mm256_mul_pd(fscal,dy00);
1042             tz               = _mm256_mul_pd(fscal,dz00);
1043
1044             /* Update vectorial force */
1045             fix0             = _mm256_add_pd(fix0,tx);
1046             fiy0             = _mm256_add_pd(fiy0,ty);
1047             fiz0             = _mm256_add_pd(fiz0,tz);
1048
1049             fjx0             = _mm256_add_pd(fjx0,tx);
1050             fjy0             = _mm256_add_pd(fjy0,ty);
1051             fjz0             = _mm256_add_pd(fjz0,tz);
1052
1053             /**************************
1054              * CALCULATE INTERACTIONS *
1055              **************************/
1056
1057             /* Compute parameters for interactions between i and j atoms */
1058             qq10             = _mm256_mul_pd(iq1,jq0);
1059
1060             /* REACTION-FIELD ELECTROSTATICS */
1061             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1062
1063             fscal            = felec;
1064
1065             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1066
1067             /* Calculate temporary vectorial force */
1068             tx               = _mm256_mul_pd(fscal,dx10);
1069             ty               = _mm256_mul_pd(fscal,dy10);
1070             tz               = _mm256_mul_pd(fscal,dz10);
1071
1072             /* Update vectorial force */
1073             fix1             = _mm256_add_pd(fix1,tx);
1074             fiy1             = _mm256_add_pd(fiy1,ty);
1075             fiz1             = _mm256_add_pd(fiz1,tz);
1076
1077             fjx0             = _mm256_add_pd(fjx0,tx);
1078             fjy0             = _mm256_add_pd(fjy0,ty);
1079             fjz0             = _mm256_add_pd(fjz0,tz);
1080
1081             /**************************
1082              * CALCULATE INTERACTIONS *
1083              **************************/
1084
1085             /* Compute parameters for interactions between i and j atoms */
1086             qq20             = _mm256_mul_pd(iq2,jq0);
1087
1088             /* REACTION-FIELD ELECTROSTATICS */
1089             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1090
1091             fscal            = felec;
1092
1093             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1094
1095             /* Calculate temporary vectorial force */
1096             tx               = _mm256_mul_pd(fscal,dx20);
1097             ty               = _mm256_mul_pd(fscal,dy20);
1098             tz               = _mm256_mul_pd(fscal,dz20);
1099
1100             /* Update vectorial force */
1101             fix2             = _mm256_add_pd(fix2,tx);
1102             fiy2             = _mm256_add_pd(fiy2,ty);
1103             fiz2             = _mm256_add_pd(fiz2,tz);
1104
1105             fjx0             = _mm256_add_pd(fjx0,tx);
1106             fjy0             = _mm256_add_pd(fjy0,ty);
1107             fjz0             = _mm256_add_pd(fjz0,tz);
1108
1109             /**************************
1110              * CALCULATE INTERACTIONS *
1111              **************************/
1112
1113             /* Compute parameters for interactions between i and j atoms */
1114             qq30             = _mm256_mul_pd(iq3,jq0);
1115
1116             /* REACTION-FIELD ELECTROSTATICS */
1117             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1118
1119             fscal            = felec;
1120
1121             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1122
1123             /* Calculate temporary vectorial force */
1124             tx               = _mm256_mul_pd(fscal,dx30);
1125             ty               = _mm256_mul_pd(fscal,dy30);
1126             tz               = _mm256_mul_pd(fscal,dz30);
1127
1128             /* Update vectorial force */
1129             fix3             = _mm256_add_pd(fix3,tx);
1130             fiy3             = _mm256_add_pd(fiy3,ty);
1131             fiz3             = _mm256_add_pd(fiz3,tz);
1132
1133             fjx0             = _mm256_add_pd(fjx0,tx);
1134             fjy0             = _mm256_add_pd(fjy0,ty);
1135             fjz0             = _mm256_add_pd(fjz0,tz);
1136
1137             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1138             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1139             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1140             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1141
1142             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1143
1144             /* Inner loop uses 111 flops */
1145         }
1146
1147         /* End of innermost loop */
1148
1149         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1150                                                  f+i_coord_offset,fshift+i_shift_offset);
1151
1152         /* Increment number of inner iterations */
1153         inneriter                  += j_index_end - j_index_start;
1154
1155         /* Outer loop uses 24 flops */
1156     }
1157
1158     /* Increment number of outer iterations */
1159     outeriter        += nri;
1160
1161     /* Update outer/inner flops */
1162
1163     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*111);
1164 }