src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecNone_VdwLJ_GeomP1P1_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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_VdwLJ_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_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     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     /* Avoid stupid compiler warnings */
 115     jnrA = jnrB = jnrC = jnrD = 0;
 116     j_coord_offsetA = 0;
 117     j_coord_offsetB = 0;
 118     j_coord_offsetC = 0;
 119     j_coord_offsetD = 0;
 120
 121     outeriter        = 0;
 122     inneriter        = 0;
 123
 124     for(iidx=0;iidx<4*DIM;iidx++)
 125     {
 126         scratch[iidx] = 0.0;
 127     }
 128
 129     /* Start outer loop over neighborlists */
 130     for(iidx=0; iidx<nri; iidx++)
 131     {
 132         /* Load shift vector for this list */
 133         i_shift_offset   = DIM*shiftidx[iidx];
 134
 135         /* Load limits for loop over neighbors */
 136         j_index_start    = jindex[iidx];
 137         j_index_end      = jindex[iidx+1];
 138
 139         /* Get outer coordinate index */
 140         inr              = iinr[iidx];
 141         i_coord_offset   = DIM*inr;
 142
 143         /* Load i particle coords and add shift vector */
 144         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 145
 146         fix0             = _mm256_setzero_pd();
 147         fiy0             = _mm256_setzero_pd();
 148         fiz0             = _mm256_setzero_pd();
 149
 150         /* Load parameters for i particles */
 151         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 152
 153         /* Reset potential sums */
 154         vvdwsum          = _mm256_setzero_pd();
 155
 156         /* Start inner kernel loop */
 157         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 158         {
 159
 160             /* Get j neighbor index, and coordinate index */
 161             jnrA             = jjnr[jidx];
 162             jnrB             = jjnr[jidx+1];
 163             jnrC             = jjnr[jidx+2];
 164             jnrD             = jjnr[jidx+3];
 165             j_coord_offsetA  = DIM*jnrA;
 166             j_coord_offsetB  = DIM*jnrB;
 167             j_coord_offsetC  = DIM*jnrC;
 168             j_coord_offsetD  = DIM*jnrD;
 169
 170             /* load j atom coordinates */
 171             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 172                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 173                                                  &jx0,&jy0,&jz0);
 174
 175             /* Calculate displacement vector */
 176             dx00             = _mm256_sub_pd(ix0,jx0);
 177             dy00             = _mm256_sub_pd(iy0,jy0);
 178             dz00             = _mm256_sub_pd(iz0,jz0);
 179
 180             /* Calculate squared distance and things based on it */
 181             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 182
 183             rinvsq00         = avx256_inv_d(rsq00);
 184
 185             /* Load parameters for j particles */
 186             vdwjidx0A        = 2*vdwtype[jnrA+0];
 187             vdwjidx0B        = 2*vdwtype[jnrB+0];
 188             vdwjidx0C        = 2*vdwtype[jnrC+0];
 189             vdwjidx0D        = 2*vdwtype[jnrD+0];
 190
 191             /**************************
 192              * CALCULATE INTERACTIONS *
 193              **************************/
 194
 195             /* Compute parameters for interactions between i and j atoms */
 196             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 197                                             vdwioffsetptr0+vdwjidx0B,
 198                                             vdwioffsetptr0+vdwjidx0C,
 199                                             vdwioffsetptr0+vdwjidx0D,
 200                                             &c6_00,&c12_00);
 201
 202             /* LENNARD-JONES DISPERSION/REPULSION */
 203
 204             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 205             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 206             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 207             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 208             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 209
 210             /* Update potential sum for this i atom from the interaction with this j atom. */
 211             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 212
 213             fscal            = fvdw;
 214
 215             /* Calculate temporary vectorial force */
 216             tx               = _mm256_mul_pd(fscal,dx00);
 217             ty               = _mm256_mul_pd(fscal,dy00);
 218             tz               = _mm256_mul_pd(fscal,dz00);
 219
 220             /* Update vectorial force */
 221             fix0             = _mm256_add_pd(fix0,tx);
 222             fiy0             = _mm256_add_pd(fiy0,ty);
 223             fiz0             = _mm256_add_pd(fiz0,tz);
 224
 225             fjptrA             = f+j_coord_offsetA;
 226             fjptrB             = f+j_coord_offsetB;
 227             fjptrC             = f+j_coord_offsetC;
 228             fjptrD             = f+j_coord_offsetD;
 229             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 230
 231             /* Inner loop uses 32 flops */
 232         }
 233
 234         if(jidx<j_index_end)
 235         {
 236
 237             /* Get j neighbor index, and coordinate index */
 238             jnrlistA         = jjnr[jidx];
 239             jnrlistB         = jjnr[jidx+1];
 240             jnrlistC         = jjnr[jidx+2];
 241             jnrlistD         = jjnr[jidx+3];
 242             /* Sign of each element will be negative for non-real atoms.
 243              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 244              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 245              */
 246             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 247
 248             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 249             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 250             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 251
 252             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 253             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 254             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 255             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 256             j_coord_offsetA  = DIM*jnrA;
 257             j_coord_offsetB  = DIM*jnrB;
 258             j_coord_offsetC  = DIM*jnrC;
 259             j_coord_offsetD  = DIM*jnrD;
 260
 261             /* load j atom coordinates */
 262             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 263                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 264                                                  &jx0,&jy0,&jz0);
 265
 266             /* Calculate displacement vector */
 267             dx00             = _mm256_sub_pd(ix0,jx0);
 268             dy00             = _mm256_sub_pd(iy0,jy0);
 269             dz00             = _mm256_sub_pd(iz0,jz0);
 270
 271             /* Calculate squared distance and things based on it */
 272             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 273
 274             rinvsq00         = avx256_inv_d(rsq00);
 275
 276             /* Load parameters for j particles */
 277             vdwjidx0A        = 2*vdwtype[jnrA+0];
 278             vdwjidx0B        = 2*vdwtype[jnrB+0];
 279             vdwjidx0C        = 2*vdwtype[jnrC+0];
 280             vdwjidx0D        = 2*vdwtype[jnrD+0];
 281
 282             /**************************
 283              * CALCULATE INTERACTIONS *
 284              **************************/
 285
 286             /* Compute parameters for interactions between i and j atoms */
 287             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 288                                             vdwioffsetptr0+vdwjidx0B,
 289                                             vdwioffsetptr0+vdwjidx0C,
 290                                             vdwioffsetptr0+vdwjidx0D,
 291                                             &c6_00,&c12_00);
 292
 293             /* LENNARD-JONES DISPERSION/REPULSION */
 294
 295             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 296             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 297             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 298             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 299             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 300
 301             /* Update potential sum for this i atom from the interaction with this j atom. */
 302             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 303             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 304
 305             fscal            = fvdw;
 306
 307             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 308
 309             /* Calculate temporary vectorial force */
 310             tx               = _mm256_mul_pd(fscal,dx00);
 311             ty               = _mm256_mul_pd(fscal,dy00);
 312             tz               = _mm256_mul_pd(fscal,dz00);
 313
 314             /* Update vectorial force */
 315             fix0             = _mm256_add_pd(fix0,tx);
 316             fiy0             = _mm256_add_pd(fiy0,ty);
 317             fiz0             = _mm256_add_pd(fiz0,tz);
 318
 319             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 320             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 321             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 322             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 323             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 324
 325             /* Inner loop uses 32 flops */
 326         }
 327
 328         /* End of innermost loop */
 329
 330         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 331                                                  f+i_coord_offset,fshift+i_shift_offset);
 332
 333         ggid                        = gid[iidx];
 334         /* Update potential energies */
 335         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 336
 337         /* Increment number of inner iterations */
 338         inneriter                  += j_index_end - j_index_start;
 339
 340         /* Outer loop uses 7 flops */
 341     }
 342
 343     /* Increment number of outer iterations */
 344     outeriter        += nri;
 345
 346     /* Update outer/inner flops */
 347
 348     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*32);
 349 }
 350 /*
 351  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJ_GeomP1P1_F_avx_256_double
 352  * Electrostatics interaction: None
 353  * VdW interaction:            LennardJones
 354  * Geometry:                   Particle-Particle
 355  * Calculate force/pot:        Force
 356  */
 357 void
 358 nb_kernel_ElecNone_VdwLJ_GeomP1P1_F_avx_256_double
 359                     (t_nblist                    * gmx_restrict       nlist,
 360                      rvec                        * gmx_restrict          xx,
 361                      rvec                        * gmx_restrict          ff,
 362                      struct t_forcerec           * gmx_restrict          fr,
 363                      t_mdatoms                   * gmx_restrict     mdatoms,
 364                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 365                      t_nrnb                      * gmx_restrict        nrnb)
 366 {
 367     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 368      * just 0 for non-waters.
 369      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 370      * jnr indices corresponding to data put in the four positions in the SIMD register.
 371      */
 372     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 373     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 374     int              jnrA,jnrB,jnrC,jnrD;
 375     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 376     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 377     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 378     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 379     real             rcutoff_scalar;
 380     real             *shiftvec,*fshift,*x,*f;
 381     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 382     real             scratch[4*DIM];
 383     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 384     real *           vdwioffsetptr0;
 385     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 386     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 387     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 388     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 389     int              nvdwtype;
 390     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 391     int              *vdwtype;
 392     real             *vdwparam;
 393     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 394     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 395     __m256d          dummy_mask,cutoff_mask;
 396     __m128           tmpmask0,tmpmask1;
 397     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 398     __m256d          one     = _mm256_set1_pd(1.0);
 399     __m256d          two     = _mm256_set1_pd(2.0);
 400     x                = xx[0];
 401     f                = ff[0];
 402
 403     nri              = nlist->nri;
 404     iinr             = nlist->iinr;
 405     jindex           = nlist->jindex;
 406     jjnr             = nlist->jjnr;
 407     shiftidx         = nlist->shift;
 408     gid              = nlist->gid;
 409     shiftvec         = fr->shift_vec[0];
 410     fshift           = fr->fshift[0];
 411     nvdwtype         = fr->ntype;
 412     vdwparam         = fr->nbfp;
 413     vdwtype          = mdatoms->typeA;
 414
 415     /* Avoid stupid compiler warnings */
 416     jnrA = jnrB = jnrC = jnrD = 0;
 417     j_coord_offsetA = 0;
 418     j_coord_offsetB = 0;
 419     j_coord_offsetC = 0;
 420     j_coord_offsetD = 0;
 421
 422     outeriter        = 0;
 423     inneriter        = 0;
 424
 425     for(iidx=0;iidx<4*DIM;iidx++)
 426     {
 427         scratch[iidx] = 0.0;
 428     }
 429
 430     /* Start outer loop over neighborlists */
 431     for(iidx=0; iidx<nri; iidx++)
 432     {
 433         /* Load shift vector for this list */
 434         i_shift_offset   = DIM*shiftidx[iidx];
 435
 436         /* Load limits for loop over neighbors */
 437         j_index_start    = jindex[iidx];
 438         j_index_end      = jindex[iidx+1];
 439
 440         /* Get outer coordinate index */
 441         inr              = iinr[iidx];
 442         i_coord_offset   = DIM*inr;
 443
 444         /* Load i particle coords and add shift vector */
 445         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 446
 447         fix0             = _mm256_setzero_pd();
 448         fiy0             = _mm256_setzero_pd();
 449         fiz0             = _mm256_setzero_pd();
 450
 451         /* Load parameters for i particles */
 452         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 453
 454         /* Start inner kernel loop */
 455         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 456         {
 457
 458             /* Get j neighbor index, and coordinate index */
 459             jnrA             = jjnr[jidx];
 460             jnrB             = jjnr[jidx+1];
 461             jnrC             = jjnr[jidx+2];
 462             jnrD             = jjnr[jidx+3];
 463             j_coord_offsetA  = DIM*jnrA;
 464             j_coord_offsetB  = DIM*jnrB;
 465             j_coord_offsetC  = DIM*jnrC;
 466             j_coord_offsetD  = DIM*jnrD;
 467
 468             /* load j atom coordinates */
 469             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 470                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 471                                                  &jx0,&jy0,&jz0);
 472
 473             /* Calculate displacement vector */
 474             dx00             = _mm256_sub_pd(ix0,jx0);
 475             dy00             = _mm256_sub_pd(iy0,jy0);
 476             dz00             = _mm256_sub_pd(iz0,jz0);
 477
 478             /* Calculate squared distance and things based on it */
 479             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 480
 481             rinvsq00         = avx256_inv_d(rsq00);
 482
 483             /* Load parameters for j particles */
 484             vdwjidx0A        = 2*vdwtype[jnrA+0];
 485             vdwjidx0B        = 2*vdwtype[jnrB+0];
 486             vdwjidx0C        = 2*vdwtype[jnrC+0];
 487             vdwjidx0D        = 2*vdwtype[jnrD+0];
 488
 489             /**************************
 490              * CALCULATE INTERACTIONS *
 491              **************************/
 492
 493             /* Compute parameters for interactions between i and j atoms */
 494             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 495                                             vdwioffsetptr0+vdwjidx0B,
 496                                             vdwioffsetptr0+vdwjidx0C,
 497                                             vdwioffsetptr0+vdwjidx0D,
 498                                             &c6_00,&c12_00);
 499
 500             /* LENNARD-JONES DISPERSION/REPULSION */
 501
 502             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 503             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 504
 505             fscal            = fvdw;
 506
 507             /* Calculate temporary vectorial force */
 508             tx               = _mm256_mul_pd(fscal,dx00);
 509             ty               = _mm256_mul_pd(fscal,dy00);
 510             tz               = _mm256_mul_pd(fscal,dz00);
 511
 512             /* Update vectorial force */
 513             fix0             = _mm256_add_pd(fix0,tx);
 514             fiy0             = _mm256_add_pd(fiy0,ty);
 515             fiz0             = _mm256_add_pd(fiz0,tz);
 516
 517             fjptrA             = f+j_coord_offsetA;
 518             fjptrB             = f+j_coord_offsetB;
 519             fjptrC             = f+j_coord_offsetC;
 520             fjptrD             = f+j_coord_offsetD;
 521             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 522
 523             /* Inner loop uses 27 flops */
 524         }
 525
 526         if(jidx<j_index_end)
 527         {
 528
 529             /* Get j neighbor index, and coordinate index */
 530             jnrlistA         = jjnr[jidx];
 531             jnrlistB         = jjnr[jidx+1];
 532             jnrlistC         = jjnr[jidx+2];
 533             jnrlistD         = jjnr[jidx+3];
 534             /* Sign of each element will be negative for non-real atoms.
 535              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 536              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 537              */
 538             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 539
 540             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 541             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 542             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 543
 544             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 545             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 546             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 547             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 548             j_coord_offsetA  = DIM*jnrA;
 549             j_coord_offsetB  = DIM*jnrB;
 550             j_coord_offsetC  = DIM*jnrC;
 551             j_coord_offsetD  = DIM*jnrD;
 552
 553             /* load j atom coordinates */
 554             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 555                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 556                                                  &jx0,&jy0,&jz0);
 557
 558             /* Calculate displacement vector */
 559             dx00             = _mm256_sub_pd(ix0,jx0);
 560             dy00             = _mm256_sub_pd(iy0,jy0);
 561             dz00             = _mm256_sub_pd(iz0,jz0);
 562
 563             /* Calculate squared distance and things based on it */
 564             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 565
 566             rinvsq00         = avx256_inv_d(rsq00);
 567
 568             /* Load parameters for j particles */
 569             vdwjidx0A        = 2*vdwtype[jnrA+0];
 570             vdwjidx0B        = 2*vdwtype[jnrB+0];
 571             vdwjidx0C        = 2*vdwtype[jnrC+0];
 572             vdwjidx0D        = 2*vdwtype[jnrD+0];
 573
 574             /**************************
 575              * CALCULATE INTERACTIONS *
 576              **************************/
 577
 578             /* Compute parameters for interactions between i and j atoms */
 579             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 580                                             vdwioffsetptr0+vdwjidx0B,
 581                                             vdwioffsetptr0+vdwjidx0C,
 582                                             vdwioffsetptr0+vdwjidx0D,
 583                                             &c6_00,&c12_00);
 584
 585             /* LENNARD-JONES DISPERSION/REPULSION */
 586
 587             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 588             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 589
 590             fscal            = fvdw;
 591
 592             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 593
 594             /* Calculate temporary vectorial force */
 595             tx               = _mm256_mul_pd(fscal,dx00);
 596             ty               = _mm256_mul_pd(fscal,dy00);
 597             tz               = _mm256_mul_pd(fscal,dz00);
 598
 599             /* Update vectorial force */
 600             fix0             = _mm256_add_pd(fix0,tx);
 601             fiy0             = _mm256_add_pd(fiy0,ty);
 602             fiz0             = _mm256_add_pd(fiz0,tz);
 603
 604             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 605             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 606             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 607             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 608             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 609
 610             /* Inner loop uses 27 flops */
 611         }
 612
 613         /* End of innermost loop */
 614
 615         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 616                                                  f+i_coord_offset,fshift+i_shift_offset);
 617
 618         /* Increment number of inner iterations */
 619         inneriter                  += j_index_end - j_index_start;
 620
 621         /* Outer loop uses 6 flops */
 622     }
 623
 624     /* Increment number of outer iterations */
 625     outeriter        += nri;
 626
 627     /* Update outer/inner flops */
 628
 629     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*27);
 630 }