src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCoul_VdwLJ_GeomP1P1_avx_256_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_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_ElecCoul_VdwLJ_GeomP1P1_VF_avx_256_double
  51  * Electrostatics interaction: Coulomb
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCoul_VdwLJ_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     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  89     real             *charge;
  90     int              nvdwtype;
  91     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  92     int              *vdwtype;
  93     real             *vdwparam;
  94     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  95     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  96     __m256d          dummy_mask,cutoff_mask;
  97     __m128           tmpmask0,tmpmask1;
  98     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
  99     __m256d          one     = _mm256_set1_pd(1.0);
 100     __m256d          two     = _mm256_set1_pd(2.0);
 101     x                = xx[0];
 102     f                = ff[0];
 103
 104     nri              = nlist->nri;
 105     iinr             = nlist->iinr;
 106     jindex           = nlist->jindex;
 107     jjnr             = nlist->jjnr;
 108     shiftidx         = nlist->shift;
 109     gid              = nlist->gid;
 110     shiftvec         = fr->shift_vec[0];
 111     fshift           = fr->fshift[0];
 112     facel            = _mm256_set1_pd(fr->ic->epsfac);
 113     charge           = mdatoms->chargeA;
 114     nvdwtype         = fr->ntype;
 115     vdwparam         = fr->nbfp;
 116     vdwtype          = mdatoms->typeA;
 117
 118     /* Avoid stupid compiler warnings */
 119     jnrA = jnrB = jnrC = jnrD = 0;
 120     j_coord_offsetA = 0;
 121     j_coord_offsetB = 0;
 122     j_coord_offsetC = 0;
 123     j_coord_offsetD = 0;
 124
 125     outeriter        = 0;
 126     inneriter        = 0;
 127
 128     for(iidx=0;iidx<4*DIM;iidx++)
 129     {
 130         scratch[iidx] = 0.0;
 131     }
 132
 133     /* Start outer loop over neighborlists */
 134     for(iidx=0; iidx<nri; iidx++)
 135     {
 136         /* Load shift vector for this list */
 137         i_shift_offset   = DIM*shiftidx[iidx];
 138
 139         /* Load limits for loop over neighbors */
 140         j_index_start    = jindex[iidx];
 141         j_index_end      = jindex[iidx+1];
 142
 143         /* Get outer coordinate index */
 144         inr              = iinr[iidx];
 145         i_coord_offset   = DIM*inr;
 146
 147         /* Load i particle coords and add shift vector */
 148         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 149
 150         fix0             = _mm256_setzero_pd();
 151         fiy0             = _mm256_setzero_pd();
 152         fiz0             = _mm256_setzero_pd();
 153
 154         /* Load parameters for i particles */
 155         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 156         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 157
 158         /* Reset potential sums */
 159         velecsum         = _mm256_setzero_pd();
 160         vvdwsum          = _mm256_setzero_pd();
 161
 162         /* Start inner kernel loop */
 163         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 164         {
 165
 166             /* Get j neighbor index, and coordinate index */
 167             jnrA             = jjnr[jidx];
 168             jnrB             = jjnr[jidx+1];
 169             jnrC             = jjnr[jidx+2];
 170             jnrD             = jjnr[jidx+3];
 171             j_coord_offsetA  = DIM*jnrA;
 172             j_coord_offsetB  = DIM*jnrB;
 173             j_coord_offsetC  = DIM*jnrC;
 174             j_coord_offsetD  = DIM*jnrD;
 175
 176             /* load j atom coordinates */
 177             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 178                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 179                                                  &jx0,&jy0,&jz0);
 180
 181             /* Calculate displacement vector */
 182             dx00             = _mm256_sub_pd(ix0,jx0);
 183             dy00             = _mm256_sub_pd(iy0,jy0);
 184             dz00             = _mm256_sub_pd(iz0,jz0);
 185
 186             /* Calculate squared distance and things based on it */
 187             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 188
 189             rinv00           = avx256_invsqrt_d(rsq00);
 190
 191             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 192
 193             /* Load parameters for j particles */
 194             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 195                                                                  charge+jnrC+0,charge+jnrD+0);
 196             vdwjidx0A        = 2*vdwtype[jnrA+0];
 197             vdwjidx0B        = 2*vdwtype[jnrB+0];
 198             vdwjidx0C        = 2*vdwtype[jnrC+0];
 199             vdwjidx0D        = 2*vdwtype[jnrD+0];
 200
 201             /**************************
 202              * CALCULATE INTERACTIONS *
 203              **************************/
 204
 205             /* Compute parameters for interactions between i and j atoms */
 206             qq00             = _mm256_mul_pd(iq0,jq0);
 207             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 208                                             vdwioffsetptr0+vdwjidx0B,
 209                                             vdwioffsetptr0+vdwjidx0C,
 210                                             vdwioffsetptr0+vdwjidx0D,
 211                                             &c6_00,&c12_00);
 212
 213             /* COULOMB ELECTROSTATICS */
 214             velec            = _mm256_mul_pd(qq00,rinv00);
 215             felec            = _mm256_mul_pd(velec,rinvsq00);
 216
 217             /* LENNARD-JONES DISPERSION/REPULSION */
 218
 219             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 220             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 221             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 222             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 223             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 224
 225             /* Update potential sum for this i atom from the interaction with this j atom. */
 226             velecsum         = _mm256_add_pd(velecsum,velec);
 227             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 228
 229             fscal            = _mm256_add_pd(felec,fvdw);
 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             /* Inner loop uses 39 flops */
 248         }
 249
 250         if(jidx<j_index_end)
 251         {
 252
 253             /* Get j neighbor index, and coordinate index */
 254             jnrlistA         = jjnr[jidx];
 255             jnrlistB         = jjnr[jidx+1];
 256             jnrlistC         = jjnr[jidx+2];
 257             jnrlistD         = jjnr[jidx+3];
 258             /* Sign of each element will be negative for non-real atoms.
 259              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 260              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 261              */
 262             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 263
 264             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 265             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 266             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 267
 268             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 269             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 270             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 271             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 272             j_coord_offsetA  = DIM*jnrA;
 273             j_coord_offsetB  = DIM*jnrB;
 274             j_coord_offsetC  = DIM*jnrC;
 275             j_coord_offsetD  = DIM*jnrD;
 276
 277             /* load j atom coordinates */
 278             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 279                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 280                                                  &jx0,&jy0,&jz0);
 281
 282             /* Calculate displacement vector */
 283             dx00             = _mm256_sub_pd(ix0,jx0);
 284             dy00             = _mm256_sub_pd(iy0,jy0);
 285             dz00             = _mm256_sub_pd(iz0,jz0);
 286
 287             /* Calculate squared distance and things based on it */
 288             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 289
 290             rinv00           = avx256_invsqrt_d(rsq00);
 291
 292             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 293
 294             /* Load parameters for j particles */
 295             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 296                                                                  charge+jnrC+0,charge+jnrD+0);
 297             vdwjidx0A        = 2*vdwtype[jnrA+0];
 298             vdwjidx0B        = 2*vdwtype[jnrB+0];
 299             vdwjidx0C        = 2*vdwtype[jnrC+0];
 300             vdwjidx0D        = 2*vdwtype[jnrD+0];
 301
 302             /**************************
 303              * CALCULATE INTERACTIONS *
 304              **************************/
 305
 306             /* Compute parameters for interactions between i and j atoms */
 307             qq00             = _mm256_mul_pd(iq0,jq0);
 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             /* COULOMB ELECTROSTATICS */
 315             velec            = _mm256_mul_pd(qq00,rinv00);
 316             felec            = _mm256_mul_pd(velec,rinvsq00);
 317
 318             /* LENNARD-JONES DISPERSION/REPULSION */
 319
 320             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 321             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 322             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 323             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 324             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 325
 326             /* Update potential sum for this i atom from the interaction with this j atom. */
 327             velec            = _mm256_andnot_pd(dummy_mask,velec);
 328             velecsum         = _mm256_add_pd(velecsum,velec);
 329             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 330             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 331
 332             fscal            = _mm256_add_pd(felec,fvdw);
 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             /* Inner loop uses 39 flops */
 353         }
 354
 355         /* End of innermost loop */
 356
 357         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 358                                                  f+i_coord_offset,fshift+i_shift_offset);
 359
 360         ggid                        = gid[iidx];
 361         /* Update potential energies */
 362         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 363         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 364
 365         /* Increment number of inner iterations */
 366         inneriter                  += j_index_end - j_index_start;
 367
 368         /* Outer loop uses 9 flops */
 369     }
 370
 371     /* Increment number of outer iterations */
 372     outeriter        += nri;
 373
 374     /* Update outer/inner flops */
 375
 376     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*39);
 377 }
 378 /*
 379  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomP1P1_F_avx_256_double
 380  * Electrostatics interaction: Coulomb
 381  * VdW interaction:            LennardJones
 382  * Geometry:                   Particle-Particle
 383  * Calculate force/pot:        Force
 384  */
 385 void
 386 nb_kernel_ElecCoul_VdwLJ_GeomP1P1_F_avx_256_double
 387                     (t_nblist                    * gmx_restrict       nlist,
 388                      rvec                        * gmx_restrict          xx,
 389                      rvec                        * gmx_restrict          ff,
 390                      struct t_forcerec           * gmx_restrict          fr,
 391                      t_mdatoms                   * gmx_restrict     mdatoms,
 392                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 393                      t_nrnb                      * gmx_restrict        nrnb)
 394 {
 395     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 396      * just 0 for non-waters.
 397      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 398      * jnr indices corresponding to data put in the four positions in the SIMD register.
 399      */
 400     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 401     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 402     int              jnrA,jnrB,jnrC,jnrD;
 403     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 404     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 405     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 406     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 407     real             rcutoff_scalar;
 408     real             *shiftvec,*fshift,*x,*f;
 409     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 410     real             scratch[4*DIM];
 411     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 412     real *           vdwioffsetptr0;
 413     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 414     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 415     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 416     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 417     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 418     real             *charge;
 419     int              nvdwtype;
 420     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 421     int              *vdwtype;
 422     real             *vdwparam;
 423     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 424     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 425     __m256d          dummy_mask,cutoff_mask;
 426     __m128           tmpmask0,tmpmask1;
 427     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 428     __m256d          one     = _mm256_set1_pd(1.0);
 429     __m256d          two     = _mm256_set1_pd(2.0);
 430     x                = xx[0];
 431     f                = ff[0];
 432
 433     nri              = nlist->nri;
 434     iinr             = nlist->iinr;
 435     jindex           = nlist->jindex;
 436     jjnr             = nlist->jjnr;
 437     shiftidx         = nlist->shift;
 438     gid              = nlist->gid;
 439     shiftvec         = fr->shift_vec[0];
 440     fshift           = fr->fshift[0];
 441     facel            = _mm256_set1_pd(fr->ic->epsfac);
 442     charge           = mdatoms->chargeA;
 443     nvdwtype         = fr->ntype;
 444     vdwparam         = fr->nbfp;
 445     vdwtype          = mdatoms->typeA;
 446
 447     /* Avoid stupid compiler warnings */
 448     jnrA = jnrB = jnrC = jnrD = 0;
 449     j_coord_offsetA = 0;
 450     j_coord_offsetB = 0;
 451     j_coord_offsetC = 0;
 452     j_coord_offsetD = 0;
 453
 454     outeriter        = 0;
 455     inneriter        = 0;
 456
 457     for(iidx=0;iidx<4*DIM;iidx++)
 458     {
 459         scratch[iidx] = 0.0;
 460     }
 461
 462     /* Start outer loop over neighborlists */
 463     for(iidx=0; iidx<nri; iidx++)
 464     {
 465         /* Load shift vector for this list */
 466         i_shift_offset   = DIM*shiftidx[iidx];
 467
 468         /* Load limits for loop over neighbors */
 469         j_index_start    = jindex[iidx];
 470         j_index_end      = jindex[iidx+1];
 471
 472         /* Get outer coordinate index */
 473         inr              = iinr[iidx];
 474         i_coord_offset   = DIM*inr;
 475
 476         /* Load i particle coords and add shift vector */
 477         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 478
 479         fix0             = _mm256_setzero_pd();
 480         fiy0             = _mm256_setzero_pd();
 481         fiz0             = _mm256_setzero_pd();
 482
 483         /* Load parameters for i particles */
 484         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 485         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 486
 487         /* Start inner kernel loop */
 488         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 489         {
 490
 491             /* Get j neighbor index, and coordinate index */
 492             jnrA             = jjnr[jidx];
 493             jnrB             = jjnr[jidx+1];
 494             jnrC             = jjnr[jidx+2];
 495             jnrD             = jjnr[jidx+3];
 496             j_coord_offsetA  = DIM*jnrA;
 497             j_coord_offsetB  = DIM*jnrB;
 498             j_coord_offsetC  = DIM*jnrC;
 499             j_coord_offsetD  = DIM*jnrD;
 500
 501             /* load j atom coordinates */
 502             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 503                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 504                                                  &jx0,&jy0,&jz0);
 505
 506             /* Calculate displacement vector */
 507             dx00             = _mm256_sub_pd(ix0,jx0);
 508             dy00             = _mm256_sub_pd(iy0,jy0);
 509             dz00             = _mm256_sub_pd(iz0,jz0);
 510
 511             /* Calculate squared distance and things based on it */
 512             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 513
 514             rinv00           = avx256_invsqrt_d(rsq00);
 515
 516             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 517
 518             /* Load parameters for j particles */
 519             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 520                                                                  charge+jnrC+0,charge+jnrD+0);
 521             vdwjidx0A        = 2*vdwtype[jnrA+0];
 522             vdwjidx0B        = 2*vdwtype[jnrB+0];
 523             vdwjidx0C        = 2*vdwtype[jnrC+0];
 524             vdwjidx0D        = 2*vdwtype[jnrD+0];
 525
 526             /**************************
 527              * CALCULATE INTERACTIONS *
 528              **************************/
 529
 530             /* Compute parameters for interactions between i and j atoms */
 531             qq00             = _mm256_mul_pd(iq0,jq0);
 532             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 533                                             vdwioffsetptr0+vdwjidx0B,
 534                                             vdwioffsetptr0+vdwjidx0C,
 535                                             vdwioffsetptr0+vdwjidx0D,
 536                                             &c6_00,&c12_00);
 537
 538             /* COULOMB ELECTROSTATICS */
 539             velec            = _mm256_mul_pd(qq00,rinv00);
 540             felec            = _mm256_mul_pd(velec,rinvsq00);
 541
 542             /* LENNARD-JONES DISPERSION/REPULSION */
 543
 544             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 545             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 546
 547             fscal            = _mm256_add_pd(felec,fvdw);
 548
 549             /* Calculate temporary vectorial force */
 550             tx               = _mm256_mul_pd(fscal,dx00);
 551             ty               = _mm256_mul_pd(fscal,dy00);
 552             tz               = _mm256_mul_pd(fscal,dz00);
 553
 554             /* Update vectorial force */
 555             fix0             = _mm256_add_pd(fix0,tx);
 556             fiy0             = _mm256_add_pd(fiy0,ty);
 557             fiz0             = _mm256_add_pd(fiz0,tz);
 558
 559             fjptrA             = f+j_coord_offsetA;
 560             fjptrB             = f+j_coord_offsetB;
 561             fjptrC             = f+j_coord_offsetC;
 562             fjptrD             = f+j_coord_offsetD;
 563             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 564
 565             /* Inner loop uses 33 flops */
 566         }
 567
 568         if(jidx<j_index_end)
 569         {
 570
 571             /* Get j neighbor index, and coordinate index */
 572             jnrlistA         = jjnr[jidx];
 573             jnrlistB         = jjnr[jidx+1];
 574             jnrlistC         = jjnr[jidx+2];
 575             jnrlistD         = jjnr[jidx+3];
 576             /* Sign of each element will be negative for non-real atoms.
 577              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 578              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 579              */
 580             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 581
 582             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 583             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 584             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 585
 586             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 587             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 588             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 589             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 590             j_coord_offsetA  = DIM*jnrA;
 591             j_coord_offsetB  = DIM*jnrB;
 592             j_coord_offsetC  = DIM*jnrC;
 593             j_coord_offsetD  = DIM*jnrD;
 594
 595             /* load j atom coordinates */
 596             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 597                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 598                                                  &jx0,&jy0,&jz0);
 599
 600             /* Calculate displacement vector */
 601             dx00             = _mm256_sub_pd(ix0,jx0);
 602             dy00             = _mm256_sub_pd(iy0,jy0);
 603             dz00             = _mm256_sub_pd(iz0,jz0);
 604
 605             /* Calculate squared distance and things based on it */
 606             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 607
 608             rinv00           = avx256_invsqrt_d(rsq00);
 609
 610             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 611
 612             /* Load parameters for j particles */
 613             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 614                                                                  charge+jnrC+0,charge+jnrD+0);
 615             vdwjidx0A        = 2*vdwtype[jnrA+0];
 616             vdwjidx0B        = 2*vdwtype[jnrB+0];
 617             vdwjidx0C        = 2*vdwtype[jnrC+0];
 618             vdwjidx0D        = 2*vdwtype[jnrD+0];
 619
 620             /**************************
 621              * CALCULATE INTERACTIONS *
 622              **************************/
 623
 624             /* Compute parameters for interactions between i and j atoms */
 625             qq00             = _mm256_mul_pd(iq0,jq0);
 626             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 627                                             vdwioffsetptr0+vdwjidx0B,
 628                                             vdwioffsetptr0+vdwjidx0C,
 629                                             vdwioffsetptr0+vdwjidx0D,
 630                                             &c6_00,&c12_00);
 631
 632             /* COULOMB ELECTROSTATICS */
 633             velec            = _mm256_mul_pd(qq00,rinv00);
 634             felec            = _mm256_mul_pd(velec,rinvsq00);
 635
 636             /* LENNARD-JONES DISPERSION/REPULSION */
 637
 638             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 639             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 640
 641             fscal            = _mm256_add_pd(felec,fvdw);
 642
 643             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 644
 645             /* Calculate temporary vectorial force */
 646             tx               = _mm256_mul_pd(fscal,dx00);
 647             ty               = _mm256_mul_pd(fscal,dy00);
 648             tz               = _mm256_mul_pd(fscal,dz00);
 649
 650             /* Update vectorial force */
 651             fix0             = _mm256_add_pd(fix0,tx);
 652             fiy0             = _mm256_add_pd(fiy0,ty);
 653             fiz0             = _mm256_add_pd(fiz0,tz);
 654
 655             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 656             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 657             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 658             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 659             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 660
 661             /* Inner loop uses 33 flops */
 662         }
 663
 664         /* End of innermost loop */
 665
 666         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 667                                                  f+i_coord_offset,fshift+i_shift_offset);
 668
 669         /* Increment number of inner iterations */
 670         inneriter                  += j_index_end - j_index_start;
 671
 672         /* Outer loop uses 7 flops */
 673     }
 674
 675     /* Increment number of outer iterations */
 676     outeriter        += nri;
 677
 678     /* Update outer/inner flops */
 679
 680     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*33);
 681 }