src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecNone_VdwLJSh_GeomP1P1_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSh_GeomP1P1_VF_avx_256_double
  51  * Electrostatics interaction: None
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecNone_VdwLJSh_GeomP1P1_VF_avx_256_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     real *           vdwioffsetptr0;
  84     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  86     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     int              nvdwtype;
  89     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  90     int              *vdwtype;
  91     real             *vdwparam;
  92     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  93     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  94     __m256d          dummy_mask,cutoff_mask;
  95     __m128           tmpmask0,tmpmask1;
  96     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
  97     __m256d          one     = _mm256_set1_pd(1.0);
  98     __m256d          two     = _mm256_set1_pd(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     nvdwtype         = fr->ntype;
 111     vdwparam         = fr->nbfp;
 112     vdwtype          = mdatoms->typeA;
 113
 114     rcutoff_scalar   = fr->ic->rvdw;
 115     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 116     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 117
 118     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 119     rvdw             = _mm256_set1_pd(fr->ic->rvdw);
 120
 121     /* Avoid stupid compiler warnings */
 122     jnrA = jnrB = jnrC = jnrD = 0;
 123     j_coord_offsetA = 0;
 124     j_coord_offsetB = 0;
 125     j_coord_offsetC = 0;
 126     j_coord_offsetD = 0;
 127
 128     outeriter        = 0;
 129     inneriter        = 0;
 130
 131     for(iidx=0;iidx<4*DIM;iidx++)
 132     {
 133         scratch[iidx] = 0.0;
 134     }
 135
 136     /* Start outer loop over neighborlists */
 137     for(iidx=0; iidx<nri; iidx++)
 138     {
 139         /* Load shift vector for this list */
 140         i_shift_offset   = DIM*shiftidx[iidx];
 141
 142         /* Load limits for loop over neighbors */
 143         j_index_start    = jindex[iidx];
 144         j_index_end      = jindex[iidx+1];
 145
 146         /* Get outer coordinate index */
 147         inr              = iinr[iidx];
 148         i_coord_offset   = DIM*inr;
 149
 150         /* Load i particle coords and add shift vector */
 151         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 152
 153         fix0             = _mm256_setzero_pd();
 154         fiy0             = _mm256_setzero_pd();
 155         fiz0             = _mm256_setzero_pd();
 156
 157         /* Load parameters for i particles */
 158         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 159
 160         /* Reset potential sums */
 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             rinvsq00         = avx256_inv_d(rsq00);
 191
 192             /* Load parameters for j particles */
 193             vdwjidx0A        = 2*vdwtype[jnrA+0];
 194             vdwjidx0B        = 2*vdwtype[jnrB+0];
 195             vdwjidx0C        = 2*vdwtype[jnrC+0];
 196             vdwjidx0D        = 2*vdwtype[jnrD+0];
 197
 198             /**************************
 199              * CALCULATE INTERACTIONS *
 200              **************************/
 201
 202             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 203             {
 204
 205             /* Compute parameters for interactions between i and j atoms */
 206             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 207                                             vdwioffsetptr0+vdwjidx0B,
 208                                             vdwioffsetptr0+vdwjidx0C,
 209                                             vdwioffsetptr0+vdwjidx0D,
 210                                             &c6_00,&c12_00);
 211
 212             /* LENNARD-JONES DISPERSION/REPULSION */
 213
 214             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 215             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 216             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 217             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 218                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 219             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 220
 221             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 222
 223             /* Update potential sum for this i atom from the interaction with this j atom. */
 224             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 225             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 226
 227             fscal            = fvdw;
 228
 229             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 230
 231             /* Calculate temporary vectorial force */
 232             tx               = _mm256_mul_pd(fscal,dx00);
 233             ty               = _mm256_mul_pd(fscal,dy00);
 234             tz               = _mm256_mul_pd(fscal,dz00);
 235
 236             /* Update vectorial force */
 237             fix0             = _mm256_add_pd(fix0,tx);
 238             fiy0             = _mm256_add_pd(fiy0,ty);
 239             fiz0             = _mm256_add_pd(fiz0,tz);
 240
 241             fjptrA             = f+j_coord_offsetA;
 242             fjptrB             = f+j_coord_offsetB;
 243             fjptrC             = f+j_coord_offsetC;
 244             fjptrD             = f+j_coord_offsetD;
 245             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 246
 247             }
 248
 249             /* Inner loop uses 41 flops */
 250         }
 251
 252         if(jidx<j_index_end)
 253         {
 254
 255             /* Get j neighbor index, and coordinate index */
 256             jnrlistA         = jjnr[jidx];
 257             jnrlistB         = jjnr[jidx+1];
 258             jnrlistC         = jjnr[jidx+2];
 259             jnrlistD         = jjnr[jidx+3];
 260             /* Sign of each element will be negative for non-real atoms.
 261              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 262              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 263              */
 264             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 265
 266             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 267             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 268             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 269
 270             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 271             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 272             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 273             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 274             j_coord_offsetA  = DIM*jnrA;
 275             j_coord_offsetB  = DIM*jnrB;
 276             j_coord_offsetC  = DIM*jnrC;
 277             j_coord_offsetD  = DIM*jnrD;
 278
 279             /* load j atom coordinates */
 280             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 281                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 282                                                  &jx0,&jy0,&jz0);
 283
 284             /* Calculate displacement vector */
 285             dx00             = _mm256_sub_pd(ix0,jx0);
 286             dy00             = _mm256_sub_pd(iy0,jy0);
 287             dz00             = _mm256_sub_pd(iz0,jz0);
 288
 289             /* Calculate squared distance and things based on it */
 290             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 291
 292             rinvsq00         = avx256_inv_d(rsq00);
 293
 294             /* Load parameters for j particles */
 295             vdwjidx0A        = 2*vdwtype[jnrA+0];
 296             vdwjidx0B        = 2*vdwtype[jnrB+0];
 297             vdwjidx0C        = 2*vdwtype[jnrC+0];
 298             vdwjidx0D        = 2*vdwtype[jnrD+0];
 299
 300             /**************************
 301              * CALCULATE INTERACTIONS *
 302              **************************/
 303
 304             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 305             {
 306
 307             /* Compute parameters for interactions between i and j atoms */
 308             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 309                                             vdwioffsetptr0+vdwjidx0B,
 310                                             vdwioffsetptr0+vdwjidx0C,
 311                                             vdwioffsetptr0+vdwjidx0D,
 312                                             &c6_00,&c12_00);
 313
 314             /* LENNARD-JONES DISPERSION/REPULSION */
 315
 316             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 317             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 318             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 319             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 320                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 321             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 322
 323             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 324
 325             /* Update potential sum for this i atom from the interaction with this j atom. */
 326             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 327             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 328             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 329
 330             fscal            = fvdw;
 331
 332             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 333
 334             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 335
 336             /* Calculate temporary vectorial force */
 337             tx               = _mm256_mul_pd(fscal,dx00);
 338             ty               = _mm256_mul_pd(fscal,dy00);
 339             tz               = _mm256_mul_pd(fscal,dz00);
 340
 341             /* Update vectorial force */
 342             fix0             = _mm256_add_pd(fix0,tx);
 343             fiy0             = _mm256_add_pd(fiy0,ty);
 344             fiz0             = _mm256_add_pd(fiz0,tz);
 345
 346             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 347             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 348             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 349             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 350             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 351
 352             }
 353
 354             /* Inner loop uses 41 flops */
 355         }
 356
 357         /* End of innermost loop */
 358
 359         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 360                                                  f+i_coord_offset,fshift+i_shift_offset);
 361
 362         ggid                        = gid[iidx];
 363         /* Update potential energies */
 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 7 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_VDW_VF,outeriter*7 + inneriter*41);
 378 }
 379 /*
 380  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSh_GeomP1P1_F_avx_256_double
 381  * Electrostatics interaction: None
 382  * VdW interaction:            LennardJones
 383  * Geometry:                   Particle-Particle
 384  * Calculate force/pot:        Force
 385  */
 386 void
 387 nb_kernel_ElecNone_VdwLJSh_GeomP1P1_F_avx_256_double
 388                     (t_nblist                    * gmx_restrict       nlist,
 389                      rvec                        * gmx_restrict          xx,
 390                      rvec                        * gmx_restrict          ff,
 391                      struct 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     int              nvdwtype;
 419     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 420     int              *vdwtype;
 421     real             *vdwparam;
 422     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 423     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 424     __m256d          dummy_mask,cutoff_mask;
 425     __m128           tmpmask0,tmpmask1;
 426     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 427     __m256d          one     = _mm256_set1_pd(1.0);
 428     __m256d          two     = _mm256_set1_pd(2.0);
 429     x                = xx[0];
 430     f                = ff[0];
 431
 432     nri              = nlist->nri;
 433     iinr             = nlist->iinr;
 434     jindex           = nlist->jindex;
 435     jjnr             = nlist->jjnr;
 436     shiftidx         = nlist->shift;
 437     gid              = nlist->gid;
 438     shiftvec         = fr->shift_vec[0];
 439     fshift           = fr->fshift[0];
 440     nvdwtype         = fr->ntype;
 441     vdwparam         = fr->nbfp;
 442     vdwtype          = mdatoms->typeA;
 443
 444     rcutoff_scalar   = fr->ic->rvdw;
 445     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 446     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 447
 448     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 449     rvdw             = _mm256_set1_pd(fr->ic->rvdw);
 450
 451     /* Avoid stupid compiler warnings */
 452     jnrA = jnrB = jnrC = jnrD = 0;
 453     j_coord_offsetA = 0;
 454     j_coord_offsetB = 0;
 455     j_coord_offsetC = 0;
 456     j_coord_offsetD = 0;
 457
 458     outeriter        = 0;
 459     inneriter        = 0;
 460
 461     for(iidx=0;iidx<4*DIM;iidx++)
 462     {
 463         scratch[iidx] = 0.0;
 464     }
 465
 466     /* Start outer loop over neighborlists */
 467     for(iidx=0; iidx<nri; iidx++)
 468     {
 469         /* Load shift vector for this list */
 470         i_shift_offset   = DIM*shiftidx[iidx];
 471
 472         /* Load limits for loop over neighbors */
 473         j_index_start    = jindex[iidx];
 474         j_index_end      = jindex[iidx+1];
 475
 476         /* Get outer coordinate index */
 477         inr              = iinr[iidx];
 478         i_coord_offset   = DIM*inr;
 479
 480         /* Load i particle coords and add shift vector */
 481         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 482
 483         fix0             = _mm256_setzero_pd();
 484         fiy0             = _mm256_setzero_pd();
 485         fiz0             = _mm256_setzero_pd();
 486
 487         /* Load parameters for i particles */
 488         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 489
 490         /* Start inner kernel loop */
 491         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 492         {
 493
 494             /* Get j neighbor index, and coordinate index */
 495             jnrA             = jjnr[jidx];
 496             jnrB             = jjnr[jidx+1];
 497             jnrC             = jjnr[jidx+2];
 498             jnrD             = jjnr[jidx+3];
 499             j_coord_offsetA  = DIM*jnrA;
 500             j_coord_offsetB  = DIM*jnrB;
 501             j_coord_offsetC  = DIM*jnrC;
 502             j_coord_offsetD  = DIM*jnrD;
 503
 504             /* load j atom coordinates */
 505             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 506                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 507                                                  &jx0,&jy0,&jz0);
 508
 509             /* Calculate displacement vector */
 510             dx00             = _mm256_sub_pd(ix0,jx0);
 511             dy00             = _mm256_sub_pd(iy0,jy0);
 512             dz00             = _mm256_sub_pd(iz0,jz0);
 513
 514             /* Calculate squared distance and things based on it */
 515             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 516
 517             rinvsq00         = avx256_inv_d(rsq00);
 518
 519             /* Load parameters for j particles */
 520             vdwjidx0A        = 2*vdwtype[jnrA+0];
 521             vdwjidx0B        = 2*vdwtype[jnrB+0];
 522             vdwjidx0C        = 2*vdwtype[jnrC+0];
 523             vdwjidx0D        = 2*vdwtype[jnrD+0];
 524
 525             /**************************
 526              * CALCULATE INTERACTIONS *
 527              **************************/
 528
 529             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 530             {
 531
 532             /* Compute parameters for interactions between i and j atoms */
 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             /* LENNARD-JONES DISPERSION/REPULSION */
 540
 541             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 542             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 543
 544             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 545
 546             fscal            = fvdw;
 547
 548             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 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             }
 567
 568             /* Inner loop uses 30 flops */
 569         }
 570
 571         if(jidx<j_index_end)
 572         {
 573
 574             /* Get j neighbor index, and coordinate index */
 575             jnrlistA         = jjnr[jidx];
 576             jnrlistB         = jjnr[jidx+1];
 577             jnrlistC         = jjnr[jidx+2];
 578             jnrlistD         = jjnr[jidx+3];
 579             /* Sign of each element will be negative for non-real atoms.
 580              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 581              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 582              */
 583             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 584
 585             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 586             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 587             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 588
 589             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 590             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 591             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 592             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 593             j_coord_offsetA  = DIM*jnrA;
 594             j_coord_offsetB  = DIM*jnrB;
 595             j_coord_offsetC  = DIM*jnrC;
 596             j_coord_offsetD  = DIM*jnrD;
 597
 598             /* load j atom coordinates */
 599             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 600                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 601                                                  &jx0,&jy0,&jz0);
 602
 603             /* Calculate displacement vector */
 604             dx00             = _mm256_sub_pd(ix0,jx0);
 605             dy00             = _mm256_sub_pd(iy0,jy0);
 606             dz00             = _mm256_sub_pd(iz0,jz0);
 607
 608             /* Calculate squared distance and things based on it */
 609             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 610
 611             rinvsq00         = avx256_inv_d(rsq00);
 612
 613             /* Load parameters for j particles */
 614             vdwjidx0A        = 2*vdwtype[jnrA+0];
 615             vdwjidx0B        = 2*vdwtype[jnrB+0];
 616             vdwjidx0C        = 2*vdwtype[jnrC+0];
 617             vdwjidx0D        = 2*vdwtype[jnrD+0];
 618
 619             /**************************
 620              * CALCULATE INTERACTIONS *
 621              **************************/
 622
 623             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 624             {
 625
 626             /* Compute parameters for interactions between i and j atoms */
 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             /* LENNARD-JONES DISPERSION/REPULSION */
 634
 635             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 636             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 637
 638             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 639
 640             fscal            = fvdw;
 641
 642             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 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             }
 663
 664             /* Inner loop uses 30 flops */
 665         }
 666
 667         /* End of innermost loop */
 668
 669         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 670                                                  f+i_coord_offset,fshift+i_shift_offset);
 671
 672         /* Increment number of inner iterations */
 673         inneriter                  += j_index_end - j_index_start;
 674
 675         /* Outer loop uses 6 flops */
 676     }
 677
 678     /* Increment number of outer iterations */
 679     outeriter        += nri;
 680
 681     /* Update outer/inner flops */
 682
 683     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*30);
 684 }