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, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "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_ElecNone_VdwLJSh_GeomP1P1_VF_avx_256_double
  54  * Electrostatics interaction: None
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecNone_VdwLJSh_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     int              nvdwtype;
  92     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  93     int              *vdwtype;
  94     real             *vdwparam;
  95     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  96     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  97     __m256d          dummy_mask,cutoff_mask;
  98     __m128           tmpmask0,tmpmask1;
  99     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 100     __m256d          one     = _mm256_set1_pd(1.0);
 101     __m256d          two     = _mm256_set1_pd(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     nvdwtype         = fr->ntype;
 114     vdwparam         = fr->nbfp;
 115     vdwtype          = mdatoms->typeA;
 116
 117     rcutoff_scalar   = fr->rvdw;
 118     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 119     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 120
 121     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 122     rvdw             = _mm256_set1_pd(fr->rvdw);
 123
 124     /* Avoid stupid compiler warnings */
 125     jnrA = jnrB = jnrC = jnrD = 0;
 126     j_coord_offsetA = 0;
 127     j_coord_offsetB = 0;
 128     j_coord_offsetC = 0;
 129     j_coord_offsetD = 0;
 130
 131     outeriter        = 0;
 132     inneriter        = 0;
 133
 134     for(iidx=0;iidx<4*DIM;iidx++)
 135     {
 136         scratch[iidx] = 0.0;
 137     }
 138
 139     /* Start outer loop over neighborlists */
 140     for(iidx=0; iidx<nri; iidx++)
 141     {
 142         /* Load shift vector for this list */
 143         i_shift_offset   = DIM*shiftidx[iidx];
 144
 145         /* Load limits for loop over neighbors */
 146         j_index_start    = jindex[iidx];
 147         j_index_end      = jindex[iidx+1];
 148
 149         /* Get outer coordinate index */
 150         inr              = iinr[iidx];
 151         i_coord_offset   = DIM*inr;
 152
 153         /* Load i particle coords and add shift vector */
 154         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 155
 156         fix0             = _mm256_setzero_pd();
 157         fiy0             = _mm256_setzero_pd();
 158         fiz0             = _mm256_setzero_pd();
 159
 160         /* Load parameters for i particles */
 161         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 162
 163         /* Reset potential sums */
 164         vvdwsum          = _mm256_setzero_pd();
 165
 166         /* Start inner kernel loop */
 167         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 168         {
 169
 170             /* Get j neighbor index, and coordinate index */
 171             jnrA             = jjnr[jidx];
 172             jnrB             = jjnr[jidx+1];
 173             jnrC             = jjnr[jidx+2];
 174             jnrD             = jjnr[jidx+3];
 175             j_coord_offsetA  = DIM*jnrA;
 176             j_coord_offsetB  = DIM*jnrB;
 177             j_coord_offsetC  = DIM*jnrC;
 178             j_coord_offsetD  = DIM*jnrD;
 179
 180             /* load j atom coordinates */
 181             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 182                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 183                                                  &jx0,&jy0,&jz0);
 184
 185             /* Calculate displacement vector */
 186             dx00             = _mm256_sub_pd(ix0,jx0);
 187             dy00             = _mm256_sub_pd(iy0,jy0);
 188             dz00             = _mm256_sub_pd(iz0,jz0);
 189
 190             /* Calculate squared distance and things based on it */
 191             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 192
 193             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 194
 195             /* Load parameters for j particles */
 196             vdwjidx0A        = 2*vdwtype[jnrA+0];
 197             vdwjidx0B        = 2*vdwtype[jnrB+0];
 198             vdwjidx0C        = 2*vdwtype[jnrC+0];
 199             vdwjidx0D        = 2*vdwtype[jnrD+0];
 200
 201             /**************************
 202              * CALCULATE INTERACTIONS *
 203              **************************/
 204
 205             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 206             {
 207
 208             /* Compute parameters for interactions between i and j atoms */
 209             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 210                                             vdwioffsetptr0+vdwjidx0B,
 211                                             vdwioffsetptr0+vdwjidx0C,
 212                                             vdwioffsetptr0+vdwjidx0D,
 213                                             &c6_00,&c12_00);
 214
 215             /* LENNARD-JONES DISPERSION/REPULSION */
 216
 217             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 218             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 219             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 220             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) ,
 221                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 222             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 223
 224             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 225
 226             /* Update potential sum for this i atom from the interaction with this j atom. */
 227             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 228             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 229
 230             fscal            = fvdw;
 231
 232             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 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             }
 251
 252             /* Inner loop uses 41 flops */
 253         }
 254
 255         if(jidx<j_index_end)
 256         {
 257
 258             /* Get j neighbor index, and coordinate index */
 259             jnrlistA         = jjnr[jidx];
 260             jnrlistB         = jjnr[jidx+1];
 261             jnrlistC         = jjnr[jidx+2];
 262             jnrlistD         = jjnr[jidx+3];
 263             /* Sign of each element will be negative for non-real atoms.
 264              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 265              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 266              */
 267             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 268
 269             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 270             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 271             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 272
 273             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 274             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 275             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 276             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 277             j_coord_offsetA  = DIM*jnrA;
 278             j_coord_offsetB  = DIM*jnrB;
 279             j_coord_offsetC  = DIM*jnrC;
 280             j_coord_offsetD  = DIM*jnrD;
 281
 282             /* load j atom coordinates */
 283             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 284                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 285                                                  &jx0,&jy0,&jz0);
 286
 287             /* Calculate displacement vector */
 288             dx00             = _mm256_sub_pd(ix0,jx0);
 289             dy00             = _mm256_sub_pd(iy0,jy0);
 290             dz00             = _mm256_sub_pd(iz0,jz0);
 291
 292             /* Calculate squared distance and things based on it */
 293             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 294
 295             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 296
 297             /* Load parameters for j particles */
 298             vdwjidx0A        = 2*vdwtype[jnrA+0];
 299             vdwjidx0B        = 2*vdwtype[jnrB+0];
 300             vdwjidx0C        = 2*vdwtype[jnrC+0];
 301             vdwjidx0D        = 2*vdwtype[jnrD+0];
 302
 303             /**************************
 304              * CALCULATE INTERACTIONS *
 305              **************************/
 306
 307             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 308             {
 309
 310             /* Compute parameters for interactions between i and j atoms */
 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             /* LENNARD-JONES DISPERSION/REPULSION */
 318
 319             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 320             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 321             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 322             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) ,
 323                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 324             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 325
 326             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 327
 328             /* Update potential sum for this i atom from the interaction with this j atom. */
 329             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 330             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 331             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 332
 333             fscal            = fvdw;
 334
 335             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 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             }
 356
 357             /* Inner loop uses 41 flops */
 358         }
 359
 360         /* End of innermost loop */
 361
 362         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 363                                                  f+i_coord_offset,fshift+i_shift_offset);
 364
 365         ggid                        = gid[iidx];
 366         /* Update potential energies */
 367         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 368
 369         /* Increment number of inner iterations */
 370         inneriter                  += j_index_end - j_index_start;
 371
 372         /* Outer loop uses 7 flops */
 373     }
 374
 375     /* Increment number of outer iterations */
 376     outeriter        += nri;
 377
 378     /* Update outer/inner flops */
 379
 380     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*41);
 381 }
 382 /*
 383  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSh_GeomP1P1_F_avx_256_double
 384  * Electrostatics interaction: None
 385  * VdW interaction:            LennardJones
 386  * Geometry:                   Particle-Particle
 387  * Calculate force/pot:        Force
 388  */
 389 void
 390 nb_kernel_ElecNone_VdwLJSh_GeomP1P1_F_avx_256_double
 391                     (t_nblist                    * gmx_restrict       nlist,
 392                      rvec                        * gmx_restrict          xx,
 393                      rvec                        * gmx_restrict          ff,
 394                      t_forcerec                  * gmx_restrict          fr,
 395                      t_mdatoms                   * gmx_restrict     mdatoms,
 396                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 397                      t_nrnb                      * gmx_restrict        nrnb)
 398 {
 399     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 400      * just 0 for non-waters.
 401      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 402      * jnr indices corresponding to data put in the four positions in the SIMD register.
 403      */
 404     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 405     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 406     int              jnrA,jnrB,jnrC,jnrD;
 407     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 408     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 409     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 410     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 411     real             rcutoff_scalar;
 412     real             *shiftvec,*fshift,*x,*f;
 413     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 414     real             scratch[4*DIM];
 415     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 416     real *           vdwioffsetptr0;
 417     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 418     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 419     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 420     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 421     int              nvdwtype;
 422     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 423     int              *vdwtype;
 424     real             *vdwparam;
 425     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 426     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 427     __m256d          dummy_mask,cutoff_mask;
 428     __m128           tmpmask0,tmpmask1;
 429     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 430     __m256d          one     = _mm256_set1_pd(1.0);
 431     __m256d          two     = _mm256_set1_pd(2.0);
 432     x                = xx[0];
 433     f                = ff[0];
 434
 435     nri              = nlist->nri;
 436     iinr             = nlist->iinr;
 437     jindex           = nlist->jindex;
 438     jjnr             = nlist->jjnr;
 439     shiftidx         = nlist->shift;
 440     gid              = nlist->gid;
 441     shiftvec         = fr->shift_vec[0];
 442     fshift           = fr->fshift[0];
 443     nvdwtype         = fr->ntype;
 444     vdwparam         = fr->nbfp;
 445     vdwtype          = mdatoms->typeA;
 446
 447     rcutoff_scalar   = fr->rvdw;
 448     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 449     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 450
 451     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 452     rvdw             = _mm256_set1_pd(fr->rvdw);
 453
 454     /* Avoid stupid compiler warnings */
 455     jnrA = jnrB = jnrC = jnrD = 0;
 456     j_coord_offsetA = 0;
 457     j_coord_offsetB = 0;
 458     j_coord_offsetC = 0;
 459     j_coord_offsetD = 0;
 460
 461     outeriter        = 0;
 462     inneriter        = 0;
 463
 464     for(iidx=0;iidx<4*DIM;iidx++)
 465     {
 466         scratch[iidx] = 0.0;
 467     }
 468
 469     /* Start outer loop over neighborlists */
 470     for(iidx=0; iidx<nri; iidx++)
 471     {
 472         /* Load shift vector for this list */
 473         i_shift_offset   = DIM*shiftidx[iidx];
 474
 475         /* Load limits for loop over neighbors */
 476         j_index_start    = jindex[iidx];
 477         j_index_end      = jindex[iidx+1];
 478
 479         /* Get outer coordinate index */
 480         inr              = iinr[iidx];
 481         i_coord_offset   = DIM*inr;
 482
 483         /* Load i particle coords and add shift vector */
 484         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 485
 486         fix0             = _mm256_setzero_pd();
 487         fiy0             = _mm256_setzero_pd();
 488         fiz0             = _mm256_setzero_pd();
 489
 490         /* Load parameters for i particles */
 491         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 492
 493         /* Start inner kernel loop */
 494         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 495         {
 496
 497             /* Get j neighbor index, and coordinate index */
 498             jnrA             = jjnr[jidx];
 499             jnrB             = jjnr[jidx+1];
 500             jnrC             = jjnr[jidx+2];
 501             jnrD             = jjnr[jidx+3];
 502             j_coord_offsetA  = DIM*jnrA;
 503             j_coord_offsetB  = DIM*jnrB;
 504             j_coord_offsetC  = DIM*jnrC;
 505             j_coord_offsetD  = DIM*jnrD;
 506
 507             /* load j atom coordinates */
 508             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 509                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 510                                                  &jx0,&jy0,&jz0);
 511
 512             /* Calculate displacement vector */
 513             dx00             = _mm256_sub_pd(ix0,jx0);
 514             dy00             = _mm256_sub_pd(iy0,jy0);
 515             dz00             = _mm256_sub_pd(iz0,jz0);
 516
 517             /* Calculate squared distance and things based on it */
 518             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 519
 520             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 521
 522             /* Load parameters for j particles */
 523             vdwjidx0A        = 2*vdwtype[jnrA+0];
 524             vdwjidx0B        = 2*vdwtype[jnrB+0];
 525             vdwjidx0C        = 2*vdwtype[jnrC+0];
 526             vdwjidx0D        = 2*vdwtype[jnrD+0];
 527
 528             /**************************
 529              * CALCULATE INTERACTIONS *
 530              **************************/
 531
 532             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 533             {
 534
 535             /* Compute parameters for interactions between i and j atoms */
 536             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 537                                             vdwioffsetptr0+vdwjidx0B,
 538                                             vdwioffsetptr0+vdwjidx0C,
 539                                             vdwioffsetptr0+vdwjidx0D,
 540                                             &c6_00,&c12_00);
 541
 542             /* LENNARD-JONES DISPERSION/REPULSION */
 543
 544             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 545             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 546
 547             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 548
 549             fscal            = fvdw;
 550
 551             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 552
 553             /* Calculate temporary vectorial force */
 554             tx               = _mm256_mul_pd(fscal,dx00);
 555             ty               = _mm256_mul_pd(fscal,dy00);
 556             tz               = _mm256_mul_pd(fscal,dz00);
 557
 558             /* Update vectorial force */
 559             fix0             = _mm256_add_pd(fix0,tx);
 560             fiy0             = _mm256_add_pd(fiy0,ty);
 561             fiz0             = _mm256_add_pd(fiz0,tz);
 562
 563             fjptrA             = f+j_coord_offsetA;
 564             fjptrB             = f+j_coord_offsetB;
 565             fjptrC             = f+j_coord_offsetC;
 566             fjptrD             = f+j_coord_offsetD;
 567             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 568
 569             }
 570
 571             /* Inner loop uses 30 flops */
 572         }
 573
 574         if(jidx<j_index_end)
 575         {
 576
 577             /* Get j neighbor index, and coordinate index */
 578             jnrlistA         = jjnr[jidx];
 579             jnrlistB         = jjnr[jidx+1];
 580             jnrlistC         = jjnr[jidx+2];
 581             jnrlistD         = jjnr[jidx+3];
 582             /* Sign of each element will be negative for non-real atoms.
 583              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 584              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 585              */
 586             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 587
 588             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 589             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 590             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 591
 592             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 593             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 594             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 595             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 596             j_coord_offsetA  = DIM*jnrA;
 597             j_coord_offsetB  = DIM*jnrB;
 598             j_coord_offsetC  = DIM*jnrC;
 599             j_coord_offsetD  = DIM*jnrD;
 600
 601             /* load j atom coordinates */
 602             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 603                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 604                                                  &jx0,&jy0,&jz0);
 605
 606             /* Calculate displacement vector */
 607             dx00             = _mm256_sub_pd(ix0,jx0);
 608             dy00             = _mm256_sub_pd(iy0,jy0);
 609             dz00             = _mm256_sub_pd(iz0,jz0);
 610
 611             /* Calculate squared distance and things based on it */
 612             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 613
 614             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 615
 616             /* Load parameters for j particles */
 617             vdwjidx0A        = 2*vdwtype[jnrA+0];
 618             vdwjidx0B        = 2*vdwtype[jnrB+0];
 619             vdwjidx0C        = 2*vdwtype[jnrC+0];
 620             vdwjidx0D        = 2*vdwtype[jnrD+0];
 621
 622             /**************************
 623              * CALCULATE INTERACTIONS *
 624              **************************/
 625
 626             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 627             {
 628
 629             /* Compute parameters for interactions between i and j atoms */
 630             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 631                                             vdwioffsetptr0+vdwjidx0B,
 632                                             vdwioffsetptr0+vdwjidx0C,
 633                                             vdwioffsetptr0+vdwjidx0D,
 634                                             &c6_00,&c12_00);
 635
 636             /* LENNARD-JONES DISPERSION/REPULSION */
 637
 638             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 639             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 640
 641             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 642
 643             fscal            = fvdw;
 644
 645             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 646
 647             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 648
 649             /* Calculate temporary vectorial force */
 650             tx               = _mm256_mul_pd(fscal,dx00);
 651             ty               = _mm256_mul_pd(fscal,dy00);
 652             tz               = _mm256_mul_pd(fscal,dz00);
 653
 654             /* Update vectorial force */
 655             fix0             = _mm256_add_pd(fix0,tx);
 656             fiy0             = _mm256_add_pd(fiy0,ty);
 657             fiz0             = _mm256_add_pd(fiz0,tz);
 658
 659             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 660             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 661             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 662             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 663             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 664
 665             }
 666
 667             /* Inner loop uses 30 flops */
 668         }
 669
 670         /* End of innermost loop */
 671
 672         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 673                                                  f+i_coord_offset,fshift+i_shift_offset);
 674
 675         /* Increment number of inner iterations */
 676         inneriter                  += j_index_end - j_index_start;
 677
 678         /* Outer loop uses 6 flops */
 679     }
 680
 681     /* Increment number of outer iterations */
 682     outeriter        += nri;
 683
 684     /* Update outer/inner flops */
 685
 686     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*30);
 687 }