src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRF_VdwLJ_GeomP1P1_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gmx_math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomP1P1_VF_avx_256_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_VdwLJ_GeomP1P1_VF_avx_256_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  89     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     int              nvdwtype;
  94     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  98     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  99     __m256d          dummy_mask,cutoff_mask;
 100     __m128           tmpmask0,tmpmask1;
 101     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 102     __m256d          one     = _mm256_set1_pd(1.0);
 103     __m256d          two     = _mm256_set1_pd(2.0);
 104     x                = xx[0];
 105     f                = ff[0];
 106
 107     nri              = nlist->nri;
 108     iinr             = nlist->iinr;
 109     jindex           = nlist->jindex;
 110     jjnr             = nlist->jjnr;
 111     shiftidx         = nlist->shift;
 112     gid              = nlist->gid;
 113     shiftvec         = fr->shift_vec[0];
 114     fshift           = fr->fshift[0];
 115     facel            = _mm256_set1_pd(fr->epsfac);
 116     charge           = mdatoms->chargeA;
 117     krf              = _mm256_set1_pd(fr->ic->k_rf);
 118     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 119     crf              = _mm256_set1_pd(fr->ic->c_rf);
 120     nvdwtype         = fr->ntype;
 121     vdwparam         = fr->nbfp;
 122     vdwtype          = mdatoms->typeA;
 123
 124     /* Avoid stupid compiler warnings */
 125     jnrA = jnrB = jnrC = jnrD = 0;
 126     j_coord_offsetA = 0;
 127     j_coord_offsetB = 0;
 128     j_coord_offsetC = 0;
 129     j_coord_offsetD = 0;
 130
 131     outeriter        = 0;
 132     inneriter        = 0;
 133
 134     for(iidx=0;iidx<4*DIM;iidx++)
 135     {
 136         scratch[iidx] = 0.0;
 137     }
 138
 139     /* Start outer loop over neighborlists */
 140     for(iidx=0; iidx<nri; iidx++)
 141     {
 142         /* Load shift vector for this list */
 143         i_shift_offset   = DIM*shiftidx[iidx];
 144
 145         /* Load limits for loop over neighbors */
 146         j_index_start    = jindex[iidx];
 147         j_index_end      = jindex[iidx+1];
 148
 149         /* Get outer coordinate index */
 150         inr              = iinr[iidx];
 151         i_coord_offset   = DIM*inr;
 152
 153         /* Load i particle coords and add shift vector */
 154         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 155
 156         fix0             = _mm256_setzero_pd();
 157         fiy0             = _mm256_setzero_pd();
 158         fiz0             = _mm256_setzero_pd();
 159
 160         /* Load parameters for i particles */
 161         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 162         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 163
 164         /* Reset potential sums */
 165         velecsum         = _mm256_setzero_pd();
 166         vvdwsum          = _mm256_setzero_pd();
 167
 168         /* Start inner kernel loop */
 169         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 170         {
 171
 172             /* Get j neighbor index, and coordinate index */
 173             jnrA             = jjnr[jidx];
 174             jnrB             = jjnr[jidx+1];
 175             jnrC             = jjnr[jidx+2];
 176             jnrD             = jjnr[jidx+3];
 177             j_coord_offsetA  = DIM*jnrA;
 178             j_coord_offsetB  = DIM*jnrB;
 179             j_coord_offsetC  = DIM*jnrC;
 180             j_coord_offsetD  = DIM*jnrD;
 181
 182             /* load j atom coordinates */
 183             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 184                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 185                                                  &jx0,&jy0,&jz0);
 186
 187             /* Calculate displacement vector */
 188             dx00             = _mm256_sub_pd(ix0,jx0);
 189             dy00             = _mm256_sub_pd(iy0,jy0);
 190             dz00             = _mm256_sub_pd(iz0,jz0);
 191
 192             /* Calculate squared distance and things based on it */
 193             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 194
 195             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 196
 197             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 198
 199             /* Load parameters for j particles */
 200             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 201                                                                  charge+jnrC+0,charge+jnrD+0);
 202             vdwjidx0A        = 2*vdwtype[jnrA+0];
 203             vdwjidx0B        = 2*vdwtype[jnrB+0];
 204             vdwjidx0C        = 2*vdwtype[jnrC+0];
 205             vdwjidx0D        = 2*vdwtype[jnrD+0];
 206
 207             /**************************
 208              * CALCULATE INTERACTIONS *
 209              **************************/
 210
 211             /* Compute parameters for interactions between i and j atoms */
 212             qq00             = _mm256_mul_pd(iq0,jq0);
 213             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 214                                             vdwioffsetptr0+vdwjidx0B,
 215                                             vdwioffsetptr0+vdwjidx0C,
 216                                             vdwioffsetptr0+vdwjidx0D,
 217                                             &c6_00,&c12_00);
 218
 219             /* REACTION-FIELD ELECTROSTATICS */
 220             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 221             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 222
 223             /* LENNARD-JONES DISPERSION/REPULSION */
 224
 225             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 226             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 227             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 228             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 229             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 230
 231             /* Update potential sum for this i atom from the interaction with this j atom. */
 232             velecsum         = _mm256_add_pd(velecsum,velec);
 233             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 234
 235             fscal            = _mm256_add_pd(felec,fvdw);
 236
 237             /* Calculate temporary vectorial force */
 238             tx               = _mm256_mul_pd(fscal,dx00);
 239             ty               = _mm256_mul_pd(fscal,dy00);
 240             tz               = _mm256_mul_pd(fscal,dz00);
 241
 242             /* Update vectorial force */
 243             fix0             = _mm256_add_pd(fix0,tx);
 244             fiy0             = _mm256_add_pd(fiy0,ty);
 245             fiz0             = _mm256_add_pd(fiz0,tz);
 246
 247             fjptrA             = f+j_coord_offsetA;
 248             fjptrB             = f+j_coord_offsetB;
 249             fjptrC             = f+j_coord_offsetC;
 250             fjptrD             = f+j_coord_offsetD;
 251             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 252
 253             /* Inner loop uses 44 flops */
 254         }
 255
 256         if(jidx<j_index_end)
 257         {
 258
 259             /* Get j neighbor index, and coordinate index */
 260             jnrlistA         = jjnr[jidx];
 261             jnrlistB         = jjnr[jidx+1];
 262             jnrlistC         = jjnr[jidx+2];
 263             jnrlistD         = jjnr[jidx+3];
 264             /* Sign of each element will be negative for non-real atoms.
 265              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 266              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 267              */
 268             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 269
 270             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 271             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 272             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 273
 274             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 275             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 276             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 277             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 278             j_coord_offsetA  = DIM*jnrA;
 279             j_coord_offsetB  = DIM*jnrB;
 280             j_coord_offsetC  = DIM*jnrC;
 281             j_coord_offsetD  = DIM*jnrD;
 282
 283             /* load j atom coordinates */
 284             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 285                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 286                                                  &jx0,&jy0,&jz0);
 287
 288             /* Calculate displacement vector */
 289             dx00             = _mm256_sub_pd(ix0,jx0);
 290             dy00             = _mm256_sub_pd(iy0,jy0);
 291             dz00             = _mm256_sub_pd(iz0,jz0);
 292
 293             /* Calculate squared distance and things based on it */
 294             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 295
 296             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 297
 298             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 299
 300             /* Load parameters for j particles */
 301             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 302                                                                  charge+jnrC+0,charge+jnrD+0);
 303             vdwjidx0A        = 2*vdwtype[jnrA+0];
 304             vdwjidx0B        = 2*vdwtype[jnrB+0];
 305             vdwjidx0C        = 2*vdwtype[jnrC+0];
 306             vdwjidx0D        = 2*vdwtype[jnrD+0];
 307
 308             /**************************
 309              * CALCULATE INTERACTIONS *
 310              **************************/
 311
 312             /* Compute parameters for interactions between i and j atoms */
 313             qq00             = _mm256_mul_pd(iq0,jq0);
 314             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 315                                             vdwioffsetptr0+vdwjidx0B,
 316                                             vdwioffsetptr0+vdwjidx0C,
 317                                             vdwioffsetptr0+vdwjidx0D,
 318                                             &c6_00,&c12_00);
 319
 320             /* REACTION-FIELD ELECTROSTATICS */
 321             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 322             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 323
 324             /* LENNARD-JONES DISPERSION/REPULSION */
 325
 326             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 327             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 328             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 329             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 330             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 331
 332             /* Update potential sum for this i atom from the interaction with this j atom. */
 333             velec            = _mm256_andnot_pd(dummy_mask,velec);
 334             velecsum         = _mm256_add_pd(velecsum,velec);
 335             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 336             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 337
 338             fscal            = _mm256_add_pd(felec,fvdw);
 339
 340             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 341
 342             /* Calculate temporary vectorial force */
 343             tx               = _mm256_mul_pd(fscal,dx00);
 344             ty               = _mm256_mul_pd(fscal,dy00);
 345             tz               = _mm256_mul_pd(fscal,dz00);
 346
 347             /* Update vectorial force */
 348             fix0             = _mm256_add_pd(fix0,tx);
 349             fiy0             = _mm256_add_pd(fiy0,ty);
 350             fiz0             = _mm256_add_pd(fiz0,tz);
 351
 352             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 353             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 354             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 355             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 356             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 357
 358             /* Inner loop uses 44 flops */
 359         }
 360
 361         /* End of innermost loop */
 362
 363         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 364                                                  f+i_coord_offset,fshift+i_shift_offset);
 365
 366         ggid                        = gid[iidx];
 367         /* Update potential energies */
 368         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 369         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 370
 371         /* Increment number of inner iterations */
 372         inneriter                  += j_index_end - j_index_start;
 373
 374         /* Outer loop uses 9 flops */
 375     }
 376
 377     /* Increment number of outer iterations */
 378     outeriter        += nri;
 379
 380     /* Update outer/inner flops */
 381
 382     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*44);
 383 }
 384 /*
 385  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_avx_256_double
 386  * Electrostatics interaction: ReactionField
 387  * VdW interaction:            LennardJones
 388  * Geometry:                   Particle-Particle
 389  * Calculate force/pot:        Force
 390  */
 391 void
 392 nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_avx_256_double
 393                     (t_nblist                    * gmx_restrict       nlist,
 394                      rvec                        * gmx_restrict          xx,
 395                      rvec                        * gmx_restrict          ff,
 396                      t_forcerec                  * gmx_restrict          fr,
 397                      t_mdatoms                   * gmx_restrict     mdatoms,
 398                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 399                      t_nrnb                      * gmx_restrict        nrnb)
 400 {
 401     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 402      * just 0 for non-waters.
 403      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 404      * jnr indices corresponding to data put in the four positions in the SIMD register.
 405      */
 406     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 407     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 408     int              jnrA,jnrB,jnrC,jnrD;
 409     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 410     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 411     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 412     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 413     real             rcutoff_scalar;
 414     real             *shiftvec,*fshift,*x,*f;
 415     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 416     real             scratch[4*DIM];
 417     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 418     real *           vdwioffsetptr0;
 419     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 420     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 421     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 422     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 423     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 424     real             *charge;
 425     int              nvdwtype;
 426     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 427     int              *vdwtype;
 428     real             *vdwparam;
 429     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 430     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 431     __m256d          dummy_mask,cutoff_mask;
 432     __m128           tmpmask0,tmpmask1;
 433     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 434     __m256d          one     = _mm256_set1_pd(1.0);
 435     __m256d          two     = _mm256_set1_pd(2.0);
 436     x                = xx[0];
 437     f                = ff[0];
 438
 439     nri              = nlist->nri;
 440     iinr             = nlist->iinr;
 441     jindex           = nlist->jindex;
 442     jjnr             = nlist->jjnr;
 443     shiftidx         = nlist->shift;
 444     gid              = nlist->gid;
 445     shiftvec         = fr->shift_vec[0];
 446     fshift           = fr->fshift[0];
 447     facel            = _mm256_set1_pd(fr->epsfac);
 448     charge           = mdatoms->chargeA;
 449     krf              = _mm256_set1_pd(fr->ic->k_rf);
 450     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 451     crf              = _mm256_set1_pd(fr->ic->c_rf);
 452     nvdwtype         = fr->ntype;
 453     vdwparam         = fr->nbfp;
 454     vdwtype          = mdatoms->typeA;
 455
 456     /* Avoid stupid compiler warnings */
 457     jnrA = jnrB = jnrC = jnrD = 0;
 458     j_coord_offsetA = 0;
 459     j_coord_offsetB = 0;
 460     j_coord_offsetC = 0;
 461     j_coord_offsetD = 0;
 462
 463     outeriter        = 0;
 464     inneriter        = 0;
 465
 466     for(iidx=0;iidx<4*DIM;iidx++)
 467     {
 468         scratch[iidx] = 0.0;
 469     }
 470
 471     /* Start outer loop over neighborlists */
 472     for(iidx=0; iidx<nri; iidx++)
 473     {
 474         /* Load shift vector for this list */
 475         i_shift_offset   = DIM*shiftidx[iidx];
 476
 477         /* Load limits for loop over neighbors */
 478         j_index_start    = jindex[iidx];
 479         j_index_end      = jindex[iidx+1];
 480
 481         /* Get outer coordinate index */
 482         inr              = iinr[iidx];
 483         i_coord_offset   = DIM*inr;
 484
 485         /* Load i particle coords and add shift vector */
 486         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 487
 488         fix0             = _mm256_setzero_pd();
 489         fiy0             = _mm256_setzero_pd();
 490         fiz0             = _mm256_setzero_pd();
 491
 492         /* Load parameters for i particles */
 493         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 494         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 495
 496         /* Start inner kernel loop */
 497         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 498         {
 499
 500             /* Get j neighbor index, and coordinate index */
 501             jnrA             = jjnr[jidx];
 502             jnrB             = jjnr[jidx+1];
 503             jnrC             = jjnr[jidx+2];
 504             jnrD             = jjnr[jidx+3];
 505             j_coord_offsetA  = DIM*jnrA;
 506             j_coord_offsetB  = DIM*jnrB;
 507             j_coord_offsetC  = DIM*jnrC;
 508             j_coord_offsetD  = DIM*jnrD;
 509
 510             /* load j atom coordinates */
 511             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 512                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 513                                                  &jx0,&jy0,&jz0);
 514
 515             /* Calculate displacement vector */
 516             dx00             = _mm256_sub_pd(ix0,jx0);
 517             dy00             = _mm256_sub_pd(iy0,jy0);
 518             dz00             = _mm256_sub_pd(iz0,jz0);
 519
 520             /* Calculate squared distance and things based on it */
 521             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 522
 523             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 524
 525             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 526
 527             /* Load parameters for j particles */
 528             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 529                                                                  charge+jnrC+0,charge+jnrD+0);
 530             vdwjidx0A        = 2*vdwtype[jnrA+0];
 531             vdwjidx0B        = 2*vdwtype[jnrB+0];
 532             vdwjidx0C        = 2*vdwtype[jnrC+0];
 533             vdwjidx0D        = 2*vdwtype[jnrD+0];
 534
 535             /**************************
 536              * CALCULATE INTERACTIONS *
 537              **************************/
 538
 539             /* Compute parameters for interactions between i and j atoms */
 540             qq00             = _mm256_mul_pd(iq0,jq0);
 541             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 542                                             vdwioffsetptr0+vdwjidx0B,
 543                                             vdwioffsetptr0+vdwjidx0C,
 544                                             vdwioffsetptr0+vdwjidx0D,
 545                                             &c6_00,&c12_00);
 546
 547             /* REACTION-FIELD ELECTROSTATICS */
 548             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 549
 550             /* LENNARD-JONES DISPERSION/REPULSION */
 551
 552             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 553             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 554
 555             fscal            = _mm256_add_pd(felec,fvdw);
 556
 557             /* Calculate temporary vectorial force */
 558             tx               = _mm256_mul_pd(fscal,dx00);
 559             ty               = _mm256_mul_pd(fscal,dy00);
 560             tz               = _mm256_mul_pd(fscal,dz00);
 561
 562             /* Update vectorial force */
 563             fix0             = _mm256_add_pd(fix0,tx);
 564             fiy0             = _mm256_add_pd(fiy0,ty);
 565             fiz0             = _mm256_add_pd(fiz0,tz);
 566
 567             fjptrA             = f+j_coord_offsetA;
 568             fjptrB             = f+j_coord_offsetB;
 569             fjptrC             = f+j_coord_offsetC;
 570             fjptrD             = f+j_coord_offsetD;
 571             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 572
 573             /* Inner loop uses 34 flops */
 574         }
 575
 576         if(jidx<j_index_end)
 577         {
 578
 579             /* Get j neighbor index, and coordinate index */
 580             jnrlistA         = jjnr[jidx];
 581             jnrlistB         = jjnr[jidx+1];
 582             jnrlistC         = jjnr[jidx+2];
 583             jnrlistD         = jjnr[jidx+3];
 584             /* Sign of each element will be negative for non-real atoms.
 585              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 586              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 587              */
 588             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 589
 590             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 591             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 592             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 593
 594             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 595             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 596             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 597             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 598             j_coord_offsetA  = DIM*jnrA;
 599             j_coord_offsetB  = DIM*jnrB;
 600             j_coord_offsetC  = DIM*jnrC;
 601             j_coord_offsetD  = DIM*jnrD;
 602
 603             /* load j atom coordinates */
 604             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 605                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 606                                                  &jx0,&jy0,&jz0);
 607
 608             /* Calculate displacement vector */
 609             dx00             = _mm256_sub_pd(ix0,jx0);
 610             dy00             = _mm256_sub_pd(iy0,jy0);
 611             dz00             = _mm256_sub_pd(iz0,jz0);
 612
 613             /* Calculate squared distance and things based on it */
 614             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 615
 616             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 617
 618             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 619
 620             /* Load parameters for j particles */
 621             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 622                                                                  charge+jnrC+0,charge+jnrD+0);
 623             vdwjidx0A        = 2*vdwtype[jnrA+0];
 624             vdwjidx0B        = 2*vdwtype[jnrB+0];
 625             vdwjidx0C        = 2*vdwtype[jnrC+0];
 626             vdwjidx0D        = 2*vdwtype[jnrD+0];
 627
 628             /**************************
 629              * CALCULATE INTERACTIONS *
 630              **************************/
 631
 632             /* Compute parameters for interactions between i and j atoms */
 633             qq00             = _mm256_mul_pd(iq0,jq0);
 634             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 635                                             vdwioffsetptr0+vdwjidx0B,
 636                                             vdwioffsetptr0+vdwjidx0C,
 637                                             vdwioffsetptr0+vdwjidx0D,
 638                                             &c6_00,&c12_00);
 639
 640             /* REACTION-FIELD ELECTROSTATICS */
 641             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 642
 643             /* LENNARD-JONES DISPERSION/REPULSION */
 644
 645             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 646             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 647
 648             fscal            = _mm256_add_pd(felec,fvdw);
 649
 650             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 651
 652             /* Calculate temporary vectorial force */
 653             tx               = _mm256_mul_pd(fscal,dx00);
 654             ty               = _mm256_mul_pd(fscal,dy00);
 655             tz               = _mm256_mul_pd(fscal,dz00);
 656
 657             /* Update vectorial force */
 658             fix0             = _mm256_add_pd(fix0,tx);
 659             fiy0             = _mm256_add_pd(fiy0,ty);
 660             fiz0             = _mm256_add_pd(fiz0,tz);
 661
 662             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 663             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 664             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 665             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 666             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 667
 668             /* Inner loop uses 34 flops */
 669         }
 670
 671         /* End of innermost loop */
 672
 673         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 674                                                  f+i_coord_offset,fshift+i_shift_offset);
 675
 676         /* Increment number of inner iterations */
 677         inneriter                  += j_index_end - j_index_start;
 678
 679         /* Outer loop uses 7 flops */
 680     }
 681
 682     /* Increment number of outer iterations */
 683     outeriter        += nri;
 684
 685     /* Update outer/inner flops */
 686
 687     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*34);
 688 }