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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, 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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_double.h"
  48 #include "kernelutil_x86_avx_256_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSh_GeomP1P1_VF_avx_256_double
  52  * Electrostatics interaction: None
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecNone_VdwLJSh_GeomP1P1_VF_avx_256_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  84     real *           vdwioffsetptr0;
  85     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  86     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  87     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     int              nvdwtype;
  90     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  91     int              *vdwtype;
  92     real             *vdwparam;
  93     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  94     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  95     __m256d          dummy_mask,cutoff_mask;
  96     __m128           tmpmask0,tmpmask1;
  97     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
  98     __m256d          one     = _mm256_set1_pd(1.0);
  99     __m256d          two     = _mm256_set1_pd(2.0);
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     nvdwtype         = fr->ntype;
 112     vdwparam         = fr->nbfp;
 113     vdwtype          = mdatoms->typeA;
 114
 115     rcutoff_scalar   = fr->rvdw;
 116     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 117     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 118
 119     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 120     rvdw             = _mm256_set1_pd(fr->rvdw);
 121
 122     /* Avoid stupid compiler warnings */
 123     jnrA = jnrB = jnrC = jnrD = 0;
 124     j_coord_offsetA = 0;
 125     j_coord_offsetB = 0;
 126     j_coord_offsetC = 0;
 127     j_coord_offsetD = 0;
 128
 129     outeriter        = 0;
 130     inneriter        = 0;
 131
 132     for(iidx=0;iidx<4*DIM;iidx++)
 133     {
 134         scratch[iidx] = 0.0;
 135     }
 136
 137     /* Start outer loop over neighborlists */
 138     for(iidx=0; iidx<nri; iidx++)
 139     {
 140         /* Load shift vector for this list */
 141         i_shift_offset   = DIM*shiftidx[iidx];
 142
 143         /* Load limits for loop over neighbors */
 144         j_index_start    = jindex[iidx];
 145         j_index_end      = jindex[iidx+1];
 146
 147         /* Get outer coordinate index */
 148         inr              = iinr[iidx];
 149         i_coord_offset   = DIM*inr;
 150
 151         /* Load i particle coords and add shift vector */
 152         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 153
 154         fix0             = _mm256_setzero_pd();
 155         fiy0             = _mm256_setzero_pd();
 156         fiz0             = _mm256_setzero_pd();
 157
 158         /* Load parameters for i particles */
 159         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 160
 161         /* Reset potential sums */
 162         vvdwsum          = _mm256_setzero_pd();
 163
 164         /* Start inner kernel loop */
 165         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 166         {
 167
 168             /* Get j neighbor index, and coordinate index */
 169             jnrA             = jjnr[jidx];
 170             jnrB             = jjnr[jidx+1];
 171             jnrC             = jjnr[jidx+2];
 172             jnrD             = jjnr[jidx+3];
 173             j_coord_offsetA  = DIM*jnrA;
 174             j_coord_offsetB  = DIM*jnrB;
 175             j_coord_offsetC  = DIM*jnrC;
 176             j_coord_offsetD  = DIM*jnrD;
 177
 178             /* load j atom coordinates */
 179             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 180                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 181                                                  &jx0,&jy0,&jz0);
 182
 183             /* Calculate displacement vector */
 184             dx00             = _mm256_sub_pd(ix0,jx0);
 185             dy00             = _mm256_sub_pd(iy0,jy0);
 186             dz00             = _mm256_sub_pd(iz0,jz0);
 187
 188             /* Calculate squared distance and things based on it */
 189             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 190
 191             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 192
 193             /* Load parameters for j particles */
 194             vdwjidx0A        = 2*vdwtype[jnrA+0];
 195             vdwjidx0B        = 2*vdwtype[jnrB+0];
 196             vdwjidx0C        = 2*vdwtype[jnrC+0];
 197             vdwjidx0D        = 2*vdwtype[jnrD+0];
 198
 199             /**************************
 200              * CALCULATE INTERACTIONS *
 201              **************************/
 202
 203             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 204             {
 205
 206             /* Compute parameters for interactions between i and j atoms */
 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             /* LENNARD-JONES DISPERSION/REPULSION */
 214
 215             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 216             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 217             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 218             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 219                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 220             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 221
 222             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 223
 224             /* Update potential sum for this i atom from the interaction with this j atom. */
 225             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 226             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 227
 228             fscal            = fvdw;
 229
 230             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 231
 232             /* Calculate temporary vectorial force */
 233             tx               = _mm256_mul_pd(fscal,dx00);
 234             ty               = _mm256_mul_pd(fscal,dy00);
 235             tz               = _mm256_mul_pd(fscal,dz00);
 236
 237             /* Update vectorial force */
 238             fix0             = _mm256_add_pd(fix0,tx);
 239             fiy0             = _mm256_add_pd(fiy0,ty);
 240             fiz0             = _mm256_add_pd(fiz0,tz);
 241
 242             fjptrA             = f+j_coord_offsetA;
 243             fjptrB             = f+j_coord_offsetB;
 244             fjptrC             = f+j_coord_offsetC;
 245             fjptrD             = f+j_coord_offsetD;
 246             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 247
 248             }
 249
 250             /* Inner loop uses 41 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             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 294
 295             /* Load parameters for j particles */
 296             vdwjidx0A        = 2*vdwtype[jnrA+0];
 297             vdwjidx0B        = 2*vdwtype[jnrB+0];
 298             vdwjidx0C        = 2*vdwtype[jnrC+0];
 299             vdwjidx0D        = 2*vdwtype[jnrD+0];
 300
 301             /**************************
 302              * CALCULATE INTERACTIONS *
 303              **************************/
 304
 305             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 306             {
 307
 308             /* Compute parameters for interactions between i and j atoms */
 309             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 310                                             vdwioffsetptr0+vdwjidx0B,
 311                                             vdwioffsetptr0+vdwjidx0C,
 312                                             vdwioffsetptr0+vdwjidx0D,
 313                                             &c6_00,&c12_00);
 314
 315             /* LENNARD-JONES DISPERSION/REPULSION */
 316
 317             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 318             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 319             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 320             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 321                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 322             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 323
 324             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 325
 326             /* Update potential sum for this i atom from the interaction with this j atom. */
 327             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 328             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 329             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 330
 331             fscal            = fvdw;
 332
 333             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 334
 335             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 336
 337             /* Calculate temporary vectorial force */
 338             tx               = _mm256_mul_pd(fscal,dx00);
 339             ty               = _mm256_mul_pd(fscal,dy00);
 340             tz               = _mm256_mul_pd(fscal,dz00);
 341
 342             /* Update vectorial force */
 343             fix0             = _mm256_add_pd(fix0,tx);
 344             fiy0             = _mm256_add_pd(fiy0,ty);
 345             fiz0             = _mm256_add_pd(fiz0,tz);
 346
 347             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 348             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 349             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 350             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 351             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 352
 353             }
 354
 355             /* Inner loop uses 41 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(vvdwsum,kernel_data->energygrp_vdw+ggid);
 366
 367         /* Increment number of inner iterations */
 368         inneriter                  += j_index_end - j_index_start;
 369
 370         /* Outer loop uses 7 flops */
 371     }
 372
 373     /* Increment number of outer iterations */
 374     outeriter        += nri;
 375
 376     /* Update outer/inner flops */
 377
 378     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*41);
 379 }
 380 /*
 381  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSh_GeomP1P1_F_avx_256_double
 382  * Electrostatics interaction: None
 383  * VdW interaction:            LennardJones
 384  * Geometry:                   Particle-Particle
 385  * Calculate force/pot:        Force
 386  */
 387 void
 388 nb_kernel_ElecNone_VdwLJSh_GeomP1P1_F_avx_256_double
 389                     (t_nblist                    * gmx_restrict       nlist,
 390                      rvec                        * gmx_restrict          xx,
 391                      rvec                        * gmx_restrict          ff,
 392                      t_forcerec                  * gmx_restrict          fr,
 393                      t_mdatoms                   * gmx_restrict     mdatoms,
 394                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 395                      t_nrnb                      * gmx_restrict        nrnb)
 396 {
 397     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 398      * just 0 for non-waters.
 399      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 400      * jnr indices corresponding to data put in the four positions in the SIMD register.
 401      */
 402     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 403     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 404     int              jnrA,jnrB,jnrC,jnrD;
 405     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 406     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 407     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 408     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 409     real             rcutoff_scalar;
 410     real             *shiftvec,*fshift,*x,*f;
 411     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 412     real             scratch[4*DIM];
 413     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 414     real *           vdwioffsetptr0;
 415     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 416     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 417     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 418     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 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     nvdwtype         = fr->ntype;
 442     vdwparam         = fr->nbfp;
 443     vdwtype          = mdatoms->typeA;
 444
 445     rcutoff_scalar   = fr->rvdw;
 446     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 447     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 448
 449     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 450     rvdw             = _mm256_set1_pd(fr->rvdw);
 451
 452     /* Avoid stupid compiler warnings */
 453     jnrA = jnrB = jnrC = jnrD = 0;
 454     j_coord_offsetA = 0;
 455     j_coord_offsetB = 0;
 456     j_coord_offsetC = 0;
 457     j_coord_offsetD = 0;
 458
 459     outeriter        = 0;
 460     inneriter        = 0;
 461
 462     for(iidx=0;iidx<4*DIM;iidx++)
 463     {
 464         scratch[iidx] = 0.0;
 465     }
 466
 467     /* Start outer loop over neighborlists */
 468     for(iidx=0; iidx<nri; iidx++)
 469     {
 470         /* Load shift vector for this list */
 471         i_shift_offset   = DIM*shiftidx[iidx];
 472
 473         /* Load limits for loop over neighbors */
 474         j_index_start    = jindex[iidx];
 475         j_index_end      = jindex[iidx+1];
 476
 477         /* Get outer coordinate index */
 478         inr              = iinr[iidx];
 479         i_coord_offset   = DIM*inr;
 480
 481         /* Load i particle coords and add shift vector */
 482         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 483
 484         fix0             = _mm256_setzero_pd();
 485         fiy0             = _mm256_setzero_pd();
 486         fiz0             = _mm256_setzero_pd();
 487
 488         /* Load parameters for i particles */
 489         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 490
 491         /* Start inner kernel loop */
 492         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 493         {
 494
 495             /* Get j neighbor index, and coordinate index */
 496             jnrA             = jjnr[jidx];
 497             jnrB             = jjnr[jidx+1];
 498             jnrC             = jjnr[jidx+2];
 499             jnrD             = jjnr[jidx+3];
 500             j_coord_offsetA  = DIM*jnrA;
 501             j_coord_offsetB  = DIM*jnrB;
 502             j_coord_offsetC  = DIM*jnrC;
 503             j_coord_offsetD  = DIM*jnrD;
 504
 505             /* load j atom coordinates */
 506             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 507                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 508                                                  &jx0,&jy0,&jz0);
 509
 510             /* Calculate displacement vector */
 511             dx00             = _mm256_sub_pd(ix0,jx0);
 512             dy00             = _mm256_sub_pd(iy0,jy0);
 513             dz00             = _mm256_sub_pd(iz0,jz0);
 514
 515             /* Calculate squared distance and things based on it */
 516             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 517
 518             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 519
 520             /* Load parameters for j particles */
 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             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 531             {
 532
 533             /* Compute parameters for interactions between i and j atoms */
 534             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 535                                             vdwioffsetptr0+vdwjidx0B,
 536                                             vdwioffsetptr0+vdwjidx0C,
 537                                             vdwioffsetptr0+vdwjidx0D,
 538                                             &c6_00,&c12_00);
 539
 540             /* LENNARD-JONES DISPERSION/REPULSION */
 541
 542             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 543             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 544
 545             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 546
 547             fscal            = fvdw;
 548
 549             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 550
 551             /* Calculate temporary vectorial force */
 552             tx               = _mm256_mul_pd(fscal,dx00);
 553             ty               = _mm256_mul_pd(fscal,dy00);
 554             tz               = _mm256_mul_pd(fscal,dz00);
 555
 556             /* Update vectorial force */
 557             fix0             = _mm256_add_pd(fix0,tx);
 558             fiy0             = _mm256_add_pd(fiy0,ty);
 559             fiz0             = _mm256_add_pd(fiz0,tz);
 560
 561             fjptrA             = f+j_coord_offsetA;
 562             fjptrB             = f+j_coord_offsetB;
 563             fjptrC             = f+j_coord_offsetC;
 564             fjptrD             = f+j_coord_offsetD;
 565             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 566
 567             }
 568
 569             /* Inner loop uses 30 flops */
 570         }
 571
 572         if(jidx<j_index_end)
 573         {
 574
 575             /* Get j neighbor index, and coordinate index */
 576             jnrlistA         = jjnr[jidx];
 577             jnrlistB         = jjnr[jidx+1];
 578             jnrlistC         = jjnr[jidx+2];
 579             jnrlistD         = jjnr[jidx+3];
 580             /* Sign of each element will be negative for non-real atoms.
 581              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 582              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 583              */
 584             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 585
 586             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 587             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 588             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 589
 590             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 591             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 592             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 593             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 594             j_coord_offsetA  = DIM*jnrA;
 595             j_coord_offsetB  = DIM*jnrB;
 596             j_coord_offsetC  = DIM*jnrC;
 597             j_coord_offsetD  = DIM*jnrD;
 598
 599             /* load j atom coordinates */
 600             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 601                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 602                                                  &jx0,&jy0,&jz0);
 603
 604             /* Calculate displacement vector */
 605             dx00             = _mm256_sub_pd(ix0,jx0);
 606             dy00             = _mm256_sub_pd(iy0,jy0);
 607             dz00             = _mm256_sub_pd(iz0,jz0);
 608
 609             /* Calculate squared distance and things based on it */
 610             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 611
 612             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 613
 614             /* Load parameters for j particles */
 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             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 625             {
 626
 627             /* Compute parameters for interactions between i and j atoms */
 628             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 629                                             vdwioffsetptr0+vdwjidx0B,
 630                                             vdwioffsetptr0+vdwjidx0C,
 631                                             vdwioffsetptr0+vdwjidx0D,
 632                                             &c6_00,&c12_00);
 633
 634             /* LENNARD-JONES DISPERSION/REPULSION */
 635
 636             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 637             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 638
 639             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 640
 641             fscal            = fvdw;
 642
 643             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 644
 645             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 646
 647             /* Calculate temporary vectorial force */
 648             tx               = _mm256_mul_pd(fscal,dx00);
 649             ty               = _mm256_mul_pd(fscal,dy00);
 650             tz               = _mm256_mul_pd(fscal,dz00);
 651
 652             /* Update vectorial force */
 653             fix0             = _mm256_add_pd(fix0,tx);
 654             fiy0             = _mm256_add_pd(fiy0,ty);
 655             fiz0             = _mm256_add_pd(fiz0,tz);
 656
 657             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 658             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 659             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 660             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 661             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 662
 663             }
 664
 665             /* Inner loop uses 30 flops */
 666         }
 667
 668         /* End of innermost loop */
 669
 670         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 671                                                  f+i_coord_offset,fshift+i_shift_offset);
 672
 673         /* Increment number of inner iterations */
 674         inneriter                  += j_index_end - j_index_start;
 675
 676         /* Outer loop uses 6 flops */
 677     }
 678
 679     /* Increment number of outer iterations */
 680     outeriter        += nri;
 681
 682     /* Update outer/inner flops */
 683
 684     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*30);
 685 }