src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCoul_VdwLJ_GeomP1P1_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 "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/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_ElecCoul_VdwLJ_GeomP1P1_VF_avx_256_double
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_VdwLJ_GeomP1P1_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     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  87     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  90     real             *charge;
  91     int              nvdwtype;
  92     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  93     int              *vdwtype;
  94     real             *vdwparam;
  95     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  96     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  97     __m256d          dummy_mask,cutoff_mask;
  98     __m128           tmpmask0,tmpmask1;
  99     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 100     __m256d          one     = _mm256_set1_pd(1.0);
 101     __m256d          two     = _mm256_set1_pd(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm256_set1_pd(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115     nvdwtype         = fr->ntype;
 116     vdwparam         = fr->nbfp;
 117     vdwtype          = mdatoms->typeA;
 118
 119     /* Avoid stupid compiler warnings */
 120     jnrA = jnrB = jnrC = jnrD = 0;
 121     j_coord_offsetA = 0;
 122     j_coord_offsetB = 0;
 123     j_coord_offsetC = 0;
 124     j_coord_offsetD = 0;
 125
 126     outeriter        = 0;
 127     inneriter        = 0;
 128
 129     for(iidx=0;iidx<4*DIM;iidx++)
 130     {
 131         scratch[iidx] = 0.0;
 132     }
 133
 134     /* Start outer loop over neighborlists */
 135     for(iidx=0; iidx<nri; iidx++)
 136     {
 137         /* Load shift vector for this list */
 138         i_shift_offset   = DIM*shiftidx[iidx];
 139
 140         /* Load limits for loop over neighbors */
 141         j_index_start    = jindex[iidx];
 142         j_index_end      = jindex[iidx+1];
 143
 144         /* Get outer coordinate index */
 145         inr              = iinr[iidx];
 146         i_coord_offset   = DIM*inr;
 147
 148         /* Load i particle coords and add shift vector */
 149         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 150
 151         fix0             = _mm256_setzero_pd();
 152         fiy0             = _mm256_setzero_pd();
 153         fiz0             = _mm256_setzero_pd();
 154
 155         /* Load parameters for i particles */
 156         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 157         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 158
 159         /* Reset potential sums */
 160         velecsum         = _mm256_setzero_pd();
 161         vvdwsum          = _mm256_setzero_pd();
 162
 163         /* Start inner kernel loop */
 164         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 165         {
 166
 167             /* Get j neighbor index, and coordinate index */
 168             jnrA             = jjnr[jidx];
 169             jnrB             = jjnr[jidx+1];
 170             jnrC             = jjnr[jidx+2];
 171             jnrD             = jjnr[jidx+3];
 172             j_coord_offsetA  = DIM*jnrA;
 173             j_coord_offsetB  = DIM*jnrB;
 174             j_coord_offsetC  = DIM*jnrC;
 175             j_coord_offsetD  = DIM*jnrD;
 176
 177             /* load j atom coordinates */
 178             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 179                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 180                                                  &jx0,&jy0,&jz0);
 181
 182             /* Calculate displacement vector */
 183             dx00             = _mm256_sub_pd(ix0,jx0);
 184             dy00             = _mm256_sub_pd(iy0,jy0);
 185             dz00             = _mm256_sub_pd(iz0,jz0);
 186
 187             /* Calculate squared distance and things based on it */
 188             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 189
 190             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 191
 192             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 193
 194             /* Load parameters for j particles */
 195             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 196                                                                  charge+jnrC+0,charge+jnrD+0);
 197             vdwjidx0A        = 2*vdwtype[jnrA+0];
 198             vdwjidx0B        = 2*vdwtype[jnrB+0];
 199             vdwjidx0C        = 2*vdwtype[jnrC+0];
 200             vdwjidx0D        = 2*vdwtype[jnrD+0];
 201
 202             /**************************
 203              * CALCULATE INTERACTIONS *
 204              **************************/
 205
 206             /* Compute parameters for interactions between i and j atoms */
 207             qq00             = _mm256_mul_pd(iq0,jq0);
 208             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 209                                             vdwioffsetptr0+vdwjidx0B,
 210                                             vdwioffsetptr0+vdwjidx0C,
 211                                             vdwioffsetptr0+vdwjidx0D,
 212                                             &c6_00,&c12_00);
 213
 214             /* COULOMB ELECTROSTATICS */
 215             velec            = _mm256_mul_pd(qq00,rinv00);
 216             felec            = _mm256_mul_pd(velec,rinvsq00);
 217
 218             /* LENNARD-JONES DISPERSION/REPULSION */
 219
 220             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 221             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 222             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 223             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 224             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 225
 226             /* Update potential sum for this i atom from the interaction with this j atom. */
 227             velecsum         = _mm256_add_pd(velecsum,velec);
 228             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 229
 230             fscal            = _mm256_add_pd(felec,fvdw);
 231
 232             /* Calculate temporary vectorial force */
 233             tx               = _mm256_mul_pd(fscal,dx00);
 234             ty               = _mm256_mul_pd(fscal,dy00);
 235             tz               = _mm256_mul_pd(fscal,dz00);
 236
 237             /* Update vectorial force */
 238             fix0             = _mm256_add_pd(fix0,tx);
 239             fiy0             = _mm256_add_pd(fiy0,ty);
 240             fiz0             = _mm256_add_pd(fiz0,tz);
 241
 242             fjptrA             = f+j_coord_offsetA;
 243             fjptrB             = f+j_coord_offsetB;
 244             fjptrC             = f+j_coord_offsetC;
 245             fjptrD             = f+j_coord_offsetD;
 246             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 247
 248             /* Inner loop uses 39 flops */
 249         }
 250
 251         if(jidx<j_index_end)
 252         {
 253
 254             /* Get j neighbor index, and coordinate index */
 255             jnrlistA         = jjnr[jidx];
 256             jnrlistB         = jjnr[jidx+1];
 257             jnrlistC         = jjnr[jidx+2];
 258             jnrlistD         = jjnr[jidx+3];
 259             /* Sign of each element will be negative for non-real atoms.
 260              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 261              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 262              */
 263             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 264
 265             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 266             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 267             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 268
 269             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 270             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 271             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 272             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 273             j_coord_offsetA  = DIM*jnrA;
 274             j_coord_offsetB  = DIM*jnrB;
 275             j_coord_offsetC  = DIM*jnrC;
 276             j_coord_offsetD  = DIM*jnrD;
 277
 278             /* load j atom coordinates */
 279             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 280                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 281                                                  &jx0,&jy0,&jz0);
 282
 283             /* Calculate displacement vector */
 284             dx00             = _mm256_sub_pd(ix0,jx0);
 285             dy00             = _mm256_sub_pd(iy0,jy0);
 286             dz00             = _mm256_sub_pd(iz0,jz0);
 287
 288             /* Calculate squared distance and things based on it */
 289             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 290
 291             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 292
 293             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 294
 295             /* Load parameters for j particles */
 296             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 297                                                                  charge+jnrC+0,charge+jnrD+0);
 298             vdwjidx0A        = 2*vdwtype[jnrA+0];
 299             vdwjidx0B        = 2*vdwtype[jnrB+0];
 300             vdwjidx0C        = 2*vdwtype[jnrC+0];
 301             vdwjidx0D        = 2*vdwtype[jnrD+0];
 302
 303             /**************************
 304              * CALCULATE INTERACTIONS *
 305              **************************/
 306
 307             /* Compute parameters for interactions between i and j atoms */
 308             qq00             = _mm256_mul_pd(iq0,jq0);
 309             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 310                                             vdwioffsetptr0+vdwjidx0B,
 311                                             vdwioffsetptr0+vdwjidx0C,
 312                                             vdwioffsetptr0+vdwjidx0D,
 313                                             &c6_00,&c12_00);
 314
 315             /* COULOMB ELECTROSTATICS */
 316             velec            = _mm256_mul_pd(qq00,rinv00);
 317             felec            = _mm256_mul_pd(velec,rinvsq00);
 318
 319             /* LENNARD-JONES DISPERSION/REPULSION */
 320
 321             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 322             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 323             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 324             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 325             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 326
 327             /* Update potential sum for this i atom from the interaction with this j atom. */
 328             velec            = _mm256_andnot_pd(dummy_mask,velec);
 329             velecsum         = _mm256_add_pd(velecsum,velec);
 330             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 331             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 332
 333             fscal            = _mm256_add_pd(felec,fvdw);
 334
 335             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 336
 337             /* Calculate temporary vectorial force */
 338             tx               = _mm256_mul_pd(fscal,dx00);
 339             ty               = _mm256_mul_pd(fscal,dy00);
 340             tz               = _mm256_mul_pd(fscal,dz00);
 341
 342             /* Update vectorial force */
 343             fix0             = _mm256_add_pd(fix0,tx);
 344             fiy0             = _mm256_add_pd(fiy0,ty);
 345             fiz0             = _mm256_add_pd(fiz0,tz);
 346
 347             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 348             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 349             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 350             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 351             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 352
 353             /* Inner loop uses 39 flops */
 354         }
 355
 356         /* End of innermost loop */
 357
 358         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 359                                                  f+i_coord_offset,fshift+i_shift_offset);
 360
 361         ggid                        = gid[iidx];
 362         /* Update potential energies */
 363         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 364         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 365
 366         /* Increment number of inner iterations */
 367         inneriter                  += j_index_end - j_index_start;
 368
 369         /* Outer loop uses 9 flops */
 370     }
 371
 372     /* Increment number of outer iterations */
 373     outeriter        += nri;
 374
 375     /* Update outer/inner flops */
 376
 377     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*39);
 378 }
 379 /*
 380  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomP1P1_F_avx_256_double
 381  * Electrostatics interaction: Coulomb
 382  * VdW interaction:            LennardJones
 383  * Geometry:                   Particle-Particle
 384  * Calculate force/pot:        Force
 385  */
 386 void
 387 nb_kernel_ElecCoul_VdwLJ_GeomP1P1_F_avx_256_double
 388                     (t_nblist                    * gmx_restrict       nlist,
 389                      rvec                        * gmx_restrict          xx,
 390                      rvec                        * gmx_restrict          ff,
 391                      t_forcerec                  * gmx_restrict          fr,
 392                      t_mdatoms                   * gmx_restrict     mdatoms,
 393                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 394                      t_nrnb                      * gmx_restrict        nrnb)
 395 {
 396     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 397      * just 0 for non-waters.
 398      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 399      * jnr indices corresponding to data put in the four positions in the SIMD register.
 400      */
 401     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 402     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 403     int              jnrA,jnrB,jnrC,jnrD;
 404     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 405     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 406     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 407     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 408     real             rcutoff_scalar;
 409     real             *shiftvec,*fshift,*x,*f;
 410     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 411     real             scratch[4*DIM];
 412     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 413     real *           vdwioffsetptr0;
 414     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 415     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 416     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 417     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 418     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 419     real             *charge;
 420     int              nvdwtype;
 421     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 422     int              *vdwtype;
 423     real             *vdwparam;
 424     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 425     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 426     __m256d          dummy_mask,cutoff_mask;
 427     __m128           tmpmask0,tmpmask1;
 428     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 429     __m256d          one     = _mm256_set1_pd(1.0);
 430     __m256d          two     = _mm256_set1_pd(2.0);
 431     x                = xx[0];
 432     f                = ff[0];
 433
 434     nri              = nlist->nri;
 435     iinr             = nlist->iinr;
 436     jindex           = nlist->jindex;
 437     jjnr             = nlist->jjnr;
 438     shiftidx         = nlist->shift;
 439     gid              = nlist->gid;
 440     shiftvec         = fr->shift_vec[0];
 441     fshift           = fr->fshift[0];
 442     facel            = _mm256_set1_pd(fr->epsfac);
 443     charge           = mdatoms->chargeA;
 444     nvdwtype         = fr->ntype;
 445     vdwparam         = fr->nbfp;
 446     vdwtype          = mdatoms->typeA;
 447
 448     /* Avoid stupid compiler warnings */
 449     jnrA = jnrB = jnrC = jnrD = 0;
 450     j_coord_offsetA = 0;
 451     j_coord_offsetB = 0;
 452     j_coord_offsetC = 0;
 453     j_coord_offsetD = 0;
 454
 455     outeriter        = 0;
 456     inneriter        = 0;
 457
 458     for(iidx=0;iidx<4*DIM;iidx++)
 459     {
 460         scratch[iidx] = 0.0;
 461     }
 462
 463     /* Start outer loop over neighborlists */
 464     for(iidx=0; iidx<nri; iidx++)
 465     {
 466         /* Load shift vector for this list */
 467         i_shift_offset   = DIM*shiftidx[iidx];
 468
 469         /* Load limits for loop over neighbors */
 470         j_index_start    = jindex[iidx];
 471         j_index_end      = jindex[iidx+1];
 472
 473         /* Get outer coordinate index */
 474         inr              = iinr[iidx];
 475         i_coord_offset   = DIM*inr;
 476
 477         /* Load i particle coords and add shift vector */
 478         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 479
 480         fix0             = _mm256_setzero_pd();
 481         fiy0             = _mm256_setzero_pd();
 482         fiz0             = _mm256_setzero_pd();
 483
 484         /* Load parameters for i particles */
 485         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 486         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 487
 488         /* Start inner kernel loop */
 489         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 490         {
 491
 492             /* Get j neighbor index, and coordinate index */
 493             jnrA             = jjnr[jidx];
 494             jnrB             = jjnr[jidx+1];
 495             jnrC             = jjnr[jidx+2];
 496             jnrD             = jjnr[jidx+3];
 497             j_coord_offsetA  = DIM*jnrA;
 498             j_coord_offsetB  = DIM*jnrB;
 499             j_coord_offsetC  = DIM*jnrC;
 500             j_coord_offsetD  = DIM*jnrD;
 501
 502             /* load j atom coordinates */
 503             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 504                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 505                                                  &jx0,&jy0,&jz0);
 506
 507             /* Calculate displacement vector */
 508             dx00             = _mm256_sub_pd(ix0,jx0);
 509             dy00             = _mm256_sub_pd(iy0,jy0);
 510             dz00             = _mm256_sub_pd(iz0,jz0);
 511
 512             /* Calculate squared distance and things based on it */
 513             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 514
 515             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 516
 517             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 518
 519             /* Load parameters for j particles */
 520             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 521                                                                  charge+jnrC+0,charge+jnrD+0);
 522             vdwjidx0A        = 2*vdwtype[jnrA+0];
 523             vdwjidx0B        = 2*vdwtype[jnrB+0];
 524             vdwjidx0C        = 2*vdwtype[jnrC+0];
 525             vdwjidx0D        = 2*vdwtype[jnrD+0];
 526
 527             /**************************
 528              * CALCULATE INTERACTIONS *
 529              **************************/
 530
 531             /* Compute parameters for interactions between i and j atoms */
 532             qq00             = _mm256_mul_pd(iq0,jq0);
 533             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 534                                             vdwioffsetptr0+vdwjidx0B,
 535                                             vdwioffsetptr0+vdwjidx0C,
 536                                             vdwioffsetptr0+vdwjidx0D,
 537                                             &c6_00,&c12_00);
 538
 539             /* COULOMB ELECTROSTATICS */
 540             velec            = _mm256_mul_pd(qq00,rinv00);
 541             felec            = _mm256_mul_pd(velec,rinvsq00);
 542
 543             /* LENNARD-JONES DISPERSION/REPULSION */
 544
 545             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 546             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 547
 548             fscal            = _mm256_add_pd(felec,fvdw);
 549
 550             /* Calculate temporary vectorial force */
 551             tx               = _mm256_mul_pd(fscal,dx00);
 552             ty               = _mm256_mul_pd(fscal,dy00);
 553             tz               = _mm256_mul_pd(fscal,dz00);
 554
 555             /* Update vectorial force */
 556             fix0             = _mm256_add_pd(fix0,tx);
 557             fiy0             = _mm256_add_pd(fiy0,ty);
 558             fiz0             = _mm256_add_pd(fiz0,tz);
 559
 560             fjptrA             = f+j_coord_offsetA;
 561             fjptrB             = f+j_coord_offsetB;
 562             fjptrC             = f+j_coord_offsetC;
 563             fjptrD             = f+j_coord_offsetD;
 564             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 565
 566             /* Inner loop uses 33 flops */
 567         }
 568
 569         if(jidx<j_index_end)
 570         {
 571
 572             /* Get j neighbor index, and coordinate index */
 573             jnrlistA         = jjnr[jidx];
 574             jnrlistB         = jjnr[jidx+1];
 575             jnrlistC         = jjnr[jidx+2];
 576             jnrlistD         = jjnr[jidx+3];
 577             /* Sign of each element will be negative for non-real atoms.
 578              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 579              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 580              */
 581             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 582
 583             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 584             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 585             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 586
 587             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 588             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 589             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 590             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 591             j_coord_offsetA  = DIM*jnrA;
 592             j_coord_offsetB  = DIM*jnrB;
 593             j_coord_offsetC  = DIM*jnrC;
 594             j_coord_offsetD  = DIM*jnrD;
 595
 596             /* load j atom coordinates */
 597             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 598                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 599                                                  &jx0,&jy0,&jz0);
 600
 601             /* Calculate displacement vector */
 602             dx00             = _mm256_sub_pd(ix0,jx0);
 603             dy00             = _mm256_sub_pd(iy0,jy0);
 604             dz00             = _mm256_sub_pd(iz0,jz0);
 605
 606             /* Calculate squared distance and things based on it */
 607             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 608
 609             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 610
 611             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 612
 613             /* Load parameters for j particles */
 614             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 615                                                                  charge+jnrC+0,charge+jnrD+0);
 616             vdwjidx0A        = 2*vdwtype[jnrA+0];
 617             vdwjidx0B        = 2*vdwtype[jnrB+0];
 618             vdwjidx0C        = 2*vdwtype[jnrC+0];
 619             vdwjidx0D        = 2*vdwtype[jnrD+0];
 620
 621             /**************************
 622              * CALCULATE INTERACTIONS *
 623              **************************/
 624
 625             /* Compute parameters for interactions between i and j atoms */
 626             qq00             = _mm256_mul_pd(iq0,jq0);
 627             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 628                                             vdwioffsetptr0+vdwjidx0B,
 629                                             vdwioffsetptr0+vdwjidx0C,
 630                                             vdwioffsetptr0+vdwjidx0D,
 631                                             &c6_00,&c12_00);
 632
 633             /* COULOMB ELECTROSTATICS */
 634             velec            = _mm256_mul_pd(qq00,rinv00);
 635             felec            = _mm256_mul_pd(velec,rinvsq00);
 636
 637             /* LENNARD-JONES DISPERSION/REPULSION */
 638
 639             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 640             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 641
 642             fscal            = _mm256_add_pd(felec,fvdw);
 643
 644             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 645
 646             /* Calculate temporary vectorial force */
 647             tx               = _mm256_mul_pd(fscal,dx00);
 648             ty               = _mm256_mul_pd(fscal,dy00);
 649             tz               = _mm256_mul_pd(fscal,dz00);
 650
 651             /* Update vectorial force */
 652             fix0             = _mm256_add_pd(fix0,tx);
 653             fiy0             = _mm256_add_pd(fiy0,ty);
 654             fiz0             = _mm256_add_pd(fiz0,tz);
 655
 656             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 657             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 658             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 659             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 660             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 661
 662             /* Inner loop uses 33 flops */
 663         }
 664
 665         /* End of innermost loop */
 666
 667         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 668                                                  f+i_coord_offset,fshift+i_shift_offset);
 669
 670         /* Increment number of inner iterations */
 671         inneriter                  += j_index_end - j_index_start;
 672
 673         /* Outer loop uses 7 flops */
 674     }
 675
 676     /* Increment number of outer iterations */
 677     outeriter        += nri;
 678
 679     /* Update outer/inner flops */
 680
 681     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*33);
 682 }