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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomP1P1_VF_avx_256_double
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_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     nvdwtype         = fr->ntype;
 118     vdwparam         = fr->nbfp;
 119     vdwtype          = mdatoms->typeA;
 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         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 159         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 160
 161         /* Reset potential sums */
 162         velecsum         = _mm256_setzero_pd();
 163         vvdwsum          = _mm256_setzero_pd();
 164
 165         /* Start inner kernel loop */
 166         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 167         {
 168
 169             /* Get j neighbor index, and coordinate index */
 170             jnrA             = jjnr[jidx];
 171             jnrB             = jjnr[jidx+1];
 172             jnrC             = jjnr[jidx+2];
 173             jnrD             = jjnr[jidx+3];
 174             j_coord_offsetA  = DIM*jnrA;
 175             j_coord_offsetB  = DIM*jnrB;
 176             j_coord_offsetC  = DIM*jnrC;
 177             j_coord_offsetD  = DIM*jnrD;
 178
 179             /* load j atom coordinates */
 180             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 181                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 182                                                  &jx0,&jy0,&jz0);
 183
 184             /* Calculate displacement vector */
 185             dx00             = _mm256_sub_pd(ix0,jx0);
 186             dy00             = _mm256_sub_pd(iy0,jy0);
 187             dz00             = _mm256_sub_pd(iz0,jz0);
 188
 189             /* Calculate squared distance and things based on it */
 190             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 191
 192             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 193
 194             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 195
 196             /* Load parameters for j particles */
 197             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 198                                                                  charge+jnrC+0,charge+jnrD+0);
 199             vdwjidx0A        = 2*vdwtype[jnrA+0];
 200             vdwjidx0B        = 2*vdwtype[jnrB+0];
 201             vdwjidx0C        = 2*vdwtype[jnrC+0];
 202             vdwjidx0D        = 2*vdwtype[jnrD+0];
 203
 204             /**************************
 205              * CALCULATE INTERACTIONS *
 206              **************************/
 207
 208             /* Compute parameters for interactions between i and j atoms */
 209             qq00             = _mm256_mul_pd(iq0,jq0);
 210             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 211                                             vdwioffsetptr0+vdwjidx0B,
 212                                             vdwioffsetptr0+vdwjidx0C,
 213                                             vdwioffsetptr0+vdwjidx0D,
 214                                             &c6_00,&c12_00);
 215
 216             /* COULOMB ELECTROSTATICS */
 217             velec            = _mm256_mul_pd(qq00,rinv00);
 218             felec            = _mm256_mul_pd(velec,rinvsq00);
 219
 220             /* LENNARD-JONES DISPERSION/REPULSION */
 221
 222             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 223             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 224             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 225             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 226             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 227
 228             /* Update potential sum for this i atom from the interaction with this j atom. */
 229             velecsum         = _mm256_add_pd(velecsum,velec);
 230             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 231
 232             fscal            = _mm256_add_pd(felec,fvdw);
 233
 234             /* Calculate temporary vectorial force */
 235             tx               = _mm256_mul_pd(fscal,dx00);
 236             ty               = _mm256_mul_pd(fscal,dy00);
 237             tz               = _mm256_mul_pd(fscal,dz00);
 238
 239             /* Update vectorial force */
 240             fix0             = _mm256_add_pd(fix0,tx);
 241             fiy0             = _mm256_add_pd(fiy0,ty);
 242             fiz0             = _mm256_add_pd(fiz0,tz);
 243
 244             fjptrA             = f+j_coord_offsetA;
 245             fjptrB             = f+j_coord_offsetB;
 246             fjptrC             = f+j_coord_offsetC;
 247             fjptrD             = f+j_coord_offsetD;
 248             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 249
 250             /* Inner loop uses 39 flops */
 251         }
 252
 253         if(jidx<j_index_end)
 254         {
 255
 256             /* Get j neighbor index, and coordinate index */
 257             jnrlistA         = jjnr[jidx];
 258             jnrlistB         = jjnr[jidx+1];
 259             jnrlistC         = jjnr[jidx+2];
 260             jnrlistD         = jjnr[jidx+3];
 261             /* Sign of each element will be negative for non-real atoms.
 262              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 263              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 264              */
 265             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 266
 267             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 268             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 269             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 270
 271             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 272             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 273             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 274             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 275             j_coord_offsetA  = DIM*jnrA;
 276             j_coord_offsetB  = DIM*jnrB;
 277             j_coord_offsetC  = DIM*jnrC;
 278             j_coord_offsetD  = DIM*jnrD;
 279
 280             /* load j atom coordinates */
 281             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 282                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 283                                                  &jx0,&jy0,&jz0);
 284
 285             /* Calculate displacement vector */
 286             dx00             = _mm256_sub_pd(ix0,jx0);
 287             dy00             = _mm256_sub_pd(iy0,jy0);
 288             dz00             = _mm256_sub_pd(iz0,jz0);
 289
 290             /* Calculate squared distance and things based on it */
 291             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 292
 293             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 294
 295             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 296
 297             /* Load parameters for j particles */
 298             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 299                                                                  charge+jnrC+0,charge+jnrD+0);
 300             vdwjidx0A        = 2*vdwtype[jnrA+0];
 301             vdwjidx0B        = 2*vdwtype[jnrB+0];
 302             vdwjidx0C        = 2*vdwtype[jnrC+0];
 303             vdwjidx0D        = 2*vdwtype[jnrD+0];
 304
 305             /**************************
 306              * CALCULATE INTERACTIONS *
 307              **************************/
 308
 309             /* Compute parameters for interactions between i and j atoms */
 310             qq00             = _mm256_mul_pd(iq0,jq0);
 311             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 312                                             vdwioffsetptr0+vdwjidx0B,
 313                                             vdwioffsetptr0+vdwjidx0C,
 314                                             vdwioffsetptr0+vdwjidx0D,
 315                                             &c6_00,&c12_00);
 316
 317             /* COULOMB ELECTROSTATICS */
 318             velec            = _mm256_mul_pd(qq00,rinv00);
 319             felec            = _mm256_mul_pd(velec,rinvsq00);
 320
 321             /* LENNARD-JONES DISPERSION/REPULSION */
 322
 323             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 324             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 325             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 326             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 327             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 328
 329             /* Update potential sum for this i atom from the interaction with this j atom. */
 330             velec            = _mm256_andnot_pd(dummy_mask,velec);
 331             velecsum         = _mm256_add_pd(velecsum,velec);
 332             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 333             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 334
 335             fscal            = _mm256_add_pd(felec,fvdw);
 336
 337             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 338
 339             /* Calculate temporary vectorial force */
 340             tx               = _mm256_mul_pd(fscal,dx00);
 341             ty               = _mm256_mul_pd(fscal,dy00);
 342             tz               = _mm256_mul_pd(fscal,dz00);
 343
 344             /* Update vectorial force */
 345             fix0             = _mm256_add_pd(fix0,tx);
 346             fiy0             = _mm256_add_pd(fiy0,ty);
 347             fiz0             = _mm256_add_pd(fiz0,tz);
 348
 349             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 350             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 351             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 352             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 353             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 354
 355             /* Inner loop uses 39 flops */
 356         }
 357
 358         /* End of innermost loop */
 359
 360         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 361                                                  f+i_coord_offset,fshift+i_shift_offset);
 362
 363         ggid                        = gid[iidx];
 364         /* Update potential energies */
 365         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 366         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 367
 368         /* Increment number of inner iterations */
 369         inneriter                  += j_index_end - j_index_start;
 370
 371         /* Outer loop uses 9 flops */
 372     }
 373
 374     /* Increment number of outer iterations */
 375     outeriter        += nri;
 376
 377     /* Update outer/inner flops */
 378
 379     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*39);
 380 }
 381 /*
 382  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomP1P1_F_avx_256_double
 383  * Electrostatics interaction: Coulomb
 384  * VdW interaction:            LennardJones
 385  * Geometry:                   Particle-Particle
 386  * Calculate force/pot:        Force
 387  */
 388 void
 389 nb_kernel_ElecCoul_VdwLJ_GeomP1P1_F_avx_256_double
 390                     (t_nblist                    * gmx_restrict       nlist,
 391                      rvec                        * gmx_restrict          xx,
 392                      rvec                        * gmx_restrict          ff,
 393                      t_forcerec                  * gmx_restrict          fr,
 394                      t_mdatoms                   * gmx_restrict     mdatoms,
 395                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 396                      t_nrnb                      * gmx_restrict        nrnb)
 397 {
 398     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 399      * just 0 for non-waters.
 400      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 401      * jnr indices corresponding to data put in the four positions in the SIMD register.
 402      */
 403     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 404     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 405     int              jnrA,jnrB,jnrC,jnrD;
 406     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 407     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 408     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 409     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 410     real             rcutoff_scalar;
 411     real             *shiftvec,*fshift,*x,*f;
 412     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 413     real             scratch[4*DIM];
 414     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 415     real *           vdwioffsetptr0;
 416     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 417     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 418     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 419     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 420     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 421     real             *charge;
 422     int              nvdwtype;
 423     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 424     int              *vdwtype;
 425     real             *vdwparam;
 426     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 427     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 428     __m256d          dummy_mask,cutoff_mask;
 429     __m128           tmpmask0,tmpmask1;
 430     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 431     __m256d          one     = _mm256_set1_pd(1.0);
 432     __m256d          two     = _mm256_set1_pd(2.0);
 433     x                = xx[0];
 434     f                = ff[0];
 435
 436     nri              = nlist->nri;
 437     iinr             = nlist->iinr;
 438     jindex           = nlist->jindex;
 439     jjnr             = nlist->jjnr;
 440     shiftidx         = nlist->shift;
 441     gid              = nlist->gid;
 442     shiftvec         = fr->shift_vec[0];
 443     fshift           = fr->fshift[0];
 444     facel            = _mm256_set1_pd(fr->epsfac);
 445     charge           = mdatoms->chargeA;
 446     nvdwtype         = fr->ntype;
 447     vdwparam         = fr->nbfp;
 448     vdwtype          = mdatoms->typeA;
 449
 450     /* Avoid stupid compiler warnings */
 451     jnrA = jnrB = jnrC = jnrD = 0;
 452     j_coord_offsetA = 0;
 453     j_coord_offsetB = 0;
 454     j_coord_offsetC = 0;
 455     j_coord_offsetD = 0;
 456
 457     outeriter        = 0;
 458     inneriter        = 0;
 459
 460     for(iidx=0;iidx<4*DIM;iidx++)
 461     {
 462         scratch[iidx] = 0.0;
 463     }
 464
 465     /* Start outer loop over neighborlists */
 466     for(iidx=0; iidx<nri; iidx++)
 467     {
 468         /* Load shift vector for this list */
 469         i_shift_offset   = DIM*shiftidx[iidx];
 470
 471         /* Load limits for loop over neighbors */
 472         j_index_start    = jindex[iidx];
 473         j_index_end      = jindex[iidx+1];
 474
 475         /* Get outer coordinate index */
 476         inr              = iinr[iidx];
 477         i_coord_offset   = DIM*inr;
 478
 479         /* Load i particle coords and add shift vector */
 480         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 481
 482         fix0             = _mm256_setzero_pd();
 483         fiy0             = _mm256_setzero_pd();
 484         fiz0             = _mm256_setzero_pd();
 485
 486         /* Load parameters for i particles */
 487         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 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             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 518
 519             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 520
 521             /* Load parameters for j particles */
 522             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 523                                                                  charge+jnrC+0,charge+jnrD+0);
 524             vdwjidx0A        = 2*vdwtype[jnrA+0];
 525             vdwjidx0B        = 2*vdwtype[jnrB+0];
 526             vdwjidx0C        = 2*vdwtype[jnrC+0];
 527             vdwjidx0D        = 2*vdwtype[jnrD+0];
 528
 529             /**************************
 530              * CALCULATE INTERACTIONS *
 531              **************************/
 532
 533             /* Compute parameters for interactions between i and j atoms */
 534             qq00             = _mm256_mul_pd(iq0,jq0);
 535             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 536                                             vdwioffsetptr0+vdwjidx0B,
 537                                             vdwioffsetptr0+vdwjidx0C,
 538                                             vdwioffsetptr0+vdwjidx0D,
 539                                             &c6_00,&c12_00);
 540
 541             /* COULOMB ELECTROSTATICS */
 542             velec            = _mm256_mul_pd(qq00,rinv00);
 543             felec            = _mm256_mul_pd(velec,rinvsq00);
 544
 545             /* LENNARD-JONES DISPERSION/REPULSION */
 546
 547             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 548             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 549
 550             fscal            = _mm256_add_pd(felec,fvdw);
 551
 552             /* Calculate temporary vectorial force */
 553             tx               = _mm256_mul_pd(fscal,dx00);
 554             ty               = _mm256_mul_pd(fscal,dy00);
 555             tz               = _mm256_mul_pd(fscal,dz00);
 556
 557             /* Update vectorial force */
 558             fix0             = _mm256_add_pd(fix0,tx);
 559             fiy0             = _mm256_add_pd(fiy0,ty);
 560             fiz0             = _mm256_add_pd(fiz0,tz);
 561
 562             fjptrA             = f+j_coord_offsetA;
 563             fjptrB             = f+j_coord_offsetB;
 564             fjptrC             = f+j_coord_offsetC;
 565             fjptrD             = f+j_coord_offsetD;
 566             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 567
 568             /* Inner loop uses 33 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             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 612
 613             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 614
 615             /* Load parameters for j particles */
 616             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 617                                                                  charge+jnrC+0,charge+jnrD+0);
 618             vdwjidx0A        = 2*vdwtype[jnrA+0];
 619             vdwjidx0B        = 2*vdwtype[jnrB+0];
 620             vdwjidx0C        = 2*vdwtype[jnrC+0];
 621             vdwjidx0D        = 2*vdwtype[jnrD+0];
 622
 623             /**************************
 624              * CALCULATE INTERACTIONS *
 625              **************************/
 626
 627             /* Compute parameters for interactions between i and j atoms */
 628             qq00             = _mm256_mul_pd(iq0,jq0);
 629             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 630                                             vdwioffsetptr0+vdwjidx0B,
 631                                             vdwioffsetptr0+vdwjidx0C,
 632                                             vdwioffsetptr0+vdwjidx0D,
 633                                             &c6_00,&c12_00);
 634
 635             /* COULOMB ELECTROSTATICS */
 636             velec            = _mm256_mul_pd(qq00,rinv00);
 637             felec            = _mm256_mul_pd(velec,rinvsq00);
 638
 639             /* LENNARD-JONES DISPERSION/REPULSION */
 640
 641             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 642             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 643
 644             fscal            = _mm256_add_pd(felec,fvdw);
 645
 646             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 647
 648             /* Calculate temporary vectorial force */
 649             tx               = _mm256_mul_pd(fscal,dx00);
 650             ty               = _mm256_mul_pd(fscal,dy00);
 651             tz               = _mm256_mul_pd(fscal,dz00);
 652
 653             /* Update vectorial force */
 654             fix0             = _mm256_add_pd(fix0,tx);
 655             fiy0             = _mm256_add_pd(fiy0,ty);
 656             fiz0             = _mm256_add_pd(fiz0,tz);
 657
 658             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 659             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 660             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 661             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 662             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 663
 664             /* Inner loop uses 33 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 7 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_ELEC_VDW_F,outeriter*7 + inneriter*33);
 684 }