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, 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 "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "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_VdwLJ_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_VdwLJ_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     /* Avoid stupid compiler warnings */
 116     jnrA = jnrB = jnrC = jnrD = 0;
 117     j_coord_offsetA = 0;
 118     j_coord_offsetB = 0;
 119     j_coord_offsetC = 0;
 120     j_coord_offsetD = 0;
 121
 122     outeriter        = 0;
 123     inneriter        = 0;
 124
 125     for(iidx=0;iidx<4*DIM;iidx++)
 126     {
 127         scratch[iidx] = 0.0;
 128     }
 129
 130     /* Start outer loop over neighborlists */
 131     for(iidx=0; iidx<nri; iidx++)
 132     {
 133         /* Load shift vector for this list */
 134         i_shift_offset   = DIM*shiftidx[iidx];
 135
 136         /* Load limits for loop over neighbors */
 137         j_index_start    = jindex[iidx];
 138         j_index_end      = jindex[iidx+1];
 139
 140         /* Get outer coordinate index */
 141         inr              = iinr[iidx];
 142         i_coord_offset   = DIM*inr;
 143
 144         /* Load i particle coords and add shift vector */
 145         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 146
 147         fix0             = _mm256_setzero_pd();
 148         fiy0             = _mm256_setzero_pd();
 149         fiz0             = _mm256_setzero_pd();
 150
 151         /* Load parameters for i particles */
 152         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 153
 154         /* Reset potential sums */
 155         vvdwsum          = _mm256_setzero_pd();
 156
 157         /* Start inner kernel loop */
 158         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 159         {
 160
 161             /* Get j neighbor index, and coordinate index */
 162             jnrA             = jjnr[jidx];
 163             jnrB             = jjnr[jidx+1];
 164             jnrC             = jjnr[jidx+2];
 165             jnrD             = jjnr[jidx+3];
 166             j_coord_offsetA  = DIM*jnrA;
 167             j_coord_offsetB  = DIM*jnrB;
 168             j_coord_offsetC  = DIM*jnrC;
 169             j_coord_offsetD  = DIM*jnrD;
 170
 171             /* load j atom coordinates */
 172             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 173                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 174                                                  &jx0,&jy0,&jz0);
 175
 176             /* Calculate displacement vector */
 177             dx00             = _mm256_sub_pd(ix0,jx0);
 178             dy00             = _mm256_sub_pd(iy0,jy0);
 179             dz00             = _mm256_sub_pd(iz0,jz0);
 180
 181             /* Calculate squared distance and things based on it */
 182             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 183
 184             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 185
 186             /* Load parameters for j particles */
 187             vdwjidx0A        = 2*vdwtype[jnrA+0];
 188             vdwjidx0B        = 2*vdwtype[jnrB+0];
 189             vdwjidx0C        = 2*vdwtype[jnrC+0];
 190             vdwjidx0D        = 2*vdwtype[jnrD+0];
 191
 192             /**************************
 193              * CALCULATE INTERACTIONS *
 194              **************************/
 195
 196             /* Compute parameters for interactions between i and j atoms */
 197             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 198                                             vdwioffsetptr0+vdwjidx0B,
 199                                             vdwioffsetptr0+vdwjidx0C,
 200                                             vdwioffsetptr0+vdwjidx0D,
 201                                             &c6_00,&c12_00);
 202
 203             /* LENNARD-JONES DISPERSION/REPULSION */
 204
 205             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 206             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 207             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 208             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 209             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 210
 211             /* Update potential sum for this i atom from the interaction with this j atom. */
 212             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 213
 214             fscal            = fvdw;
 215
 216             /* Calculate temporary vectorial force */
 217             tx               = _mm256_mul_pd(fscal,dx00);
 218             ty               = _mm256_mul_pd(fscal,dy00);
 219             tz               = _mm256_mul_pd(fscal,dz00);
 220
 221             /* Update vectorial force */
 222             fix0             = _mm256_add_pd(fix0,tx);
 223             fiy0             = _mm256_add_pd(fiy0,ty);
 224             fiz0             = _mm256_add_pd(fiz0,tz);
 225
 226             fjptrA             = f+j_coord_offsetA;
 227             fjptrB             = f+j_coord_offsetB;
 228             fjptrC             = f+j_coord_offsetC;
 229             fjptrD             = f+j_coord_offsetD;
 230             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 231
 232             /* Inner loop uses 32 flops */
 233         }
 234
 235         if(jidx<j_index_end)
 236         {
 237
 238             /* Get j neighbor index, and coordinate index */
 239             jnrlistA         = jjnr[jidx];
 240             jnrlistB         = jjnr[jidx+1];
 241             jnrlistC         = jjnr[jidx+2];
 242             jnrlistD         = jjnr[jidx+3];
 243             /* Sign of each element will be negative for non-real atoms.
 244              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 245              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 246              */
 247             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 248
 249             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 250             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 251             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 252
 253             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 254             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 255             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 256             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 257             j_coord_offsetA  = DIM*jnrA;
 258             j_coord_offsetB  = DIM*jnrB;
 259             j_coord_offsetC  = DIM*jnrC;
 260             j_coord_offsetD  = DIM*jnrD;
 261
 262             /* load j atom coordinates */
 263             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 264                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 265                                                  &jx0,&jy0,&jz0);
 266
 267             /* Calculate displacement vector */
 268             dx00             = _mm256_sub_pd(ix0,jx0);
 269             dy00             = _mm256_sub_pd(iy0,jy0);
 270             dz00             = _mm256_sub_pd(iz0,jz0);
 271
 272             /* Calculate squared distance and things based on it */
 273             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 274
 275             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 276
 277             /* Load parameters for j particles */
 278             vdwjidx0A        = 2*vdwtype[jnrA+0];
 279             vdwjidx0B        = 2*vdwtype[jnrB+0];
 280             vdwjidx0C        = 2*vdwtype[jnrC+0];
 281             vdwjidx0D        = 2*vdwtype[jnrD+0];
 282
 283             /**************************
 284              * CALCULATE INTERACTIONS *
 285              **************************/
 286
 287             /* Compute parameters for interactions between i and j atoms */
 288             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 289                                             vdwioffsetptr0+vdwjidx0B,
 290                                             vdwioffsetptr0+vdwjidx0C,
 291                                             vdwioffsetptr0+vdwjidx0D,
 292                                             &c6_00,&c12_00);
 293
 294             /* LENNARD-JONES DISPERSION/REPULSION */
 295
 296             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 297             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 298             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 299             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 300             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 301
 302             /* Update potential sum for this i atom from the interaction with this j atom. */
 303             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 304             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 305
 306             fscal            = fvdw;
 307
 308             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 309
 310             /* Calculate temporary vectorial force */
 311             tx               = _mm256_mul_pd(fscal,dx00);
 312             ty               = _mm256_mul_pd(fscal,dy00);
 313             tz               = _mm256_mul_pd(fscal,dz00);
 314
 315             /* Update vectorial force */
 316             fix0             = _mm256_add_pd(fix0,tx);
 317             fiy0             = _mm256_add_pd(fiy0,ty);
 318             fiz0             = _mm256_add_pd(fiz0,tz);
 319
 320             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 321             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 322             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 323             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 324             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 325
 326             /* Inner loop uses 32 flops */
 327         }
 328
 329         /* End of innermost loop */
 330
 331         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 332                                                  f+i_coord_offset,fshift+i_shift_offset);
 333
 334         ggid                        = gid[iidx];
 335         /* Update potential energies */
 336         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 337
 338         /* Increment number of inner iterations */
 339         inneriter                  += j_index_end - j_index_start;
 340
 341         /* Outer loop uses 7 flops */
 342     }
 343
 344     /* Increment number of outer iterations */
 345     outeriter        += nri;
 346
 347     /* Update outer/inner flops */
 348
 349     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*32);
 350 }
 351 /*
 352  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJ_GeomP1P1_F_avx_256_double
 353  * Electrostatics interaction: None
 354  * VdW interaction:            LennardJones
 355  * Geometry:                   Particle-Particle
 356  * Calculate force/pot:        Force
 357  */
 358 void
 359 nb_kernel_ElecNone_VdwLJ_GeomP1P1_F_avx_256_double
 360                     (t_nblist                    * gmx_restrict       nlist,
 361                      rvec                        * gmx_restrict          xx,
 362                      rvec                        * gmx_restrict          ff,
 363                      t_forcerec                  * gmx_restrict          fr,
 364                      t_mdatoms                   * gmx_restrict     mdatoms,
 365                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 366                      t_nrnb                      * gmx_restrict        nrnb)
 367 {
 368     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 369      * just 0 for non-waters.
 370      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 371      * jnr indices corresponding to data put in the four positions in the SIMD register.
 372      */
 373     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 374     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 375     int              jnrA,jnrB,jnrC,jnrD;
 376     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 377     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 378     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 379     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 380     real             rcutoff_scalar;
 381     real             *shiftvec,*fshift,*x,*f;
 382     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 383     real             scratch[4*DIM];
 384     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 385     real *           vdwioffsetptr0;
 386     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 387     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 388     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 389     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 390     int              nvdwtype;
 391     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 392     int              *vdwtype;
 393     real             *vdwparam;
 394     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 395     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 396     __m256d          dummy_mask,cutoff_mask;
 397     __m128           tmpmask0,tmpmask1;
 398     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 399     __m256d          one     = _mm256_set1_pd(1.0);
 400     __m256d          two     = _mm256_set1_pd(2.0);
 401     x                = xx[0];
 402     f                = ff[0];
 403
 404     nri              = nlist->nri;
 405     iinr             = nlist->iinr;
 406     jindex           = nlist->jindex;
 407     jjnr             = nlist->jjnr;
 408     shiftidx         = nlist->shift;
 409     gid              = nlist->gid;
 410     shiftvec         = fr->shift_vec[0];
 411     fshift           = fr->fshift[0];
 412     nvdwtype         = fr->ntype;
 413     vdwparam         = fr->nbfp;
 414     vdwtype          = mdatoms->typeA;
 415
 416     /* Avoid stupid compiler warnings */
 417     jnrA = jnrB = jnrC = jnrD = 0;
 418     j_coord_offsetA = 0;
 419     j_coord_offsetB = 0;
 420     j_coord_offsetC = 0;
 421     j_coord_offsetD = 0;
 422
 423     outeriter        = 0;
 424     inneriter        = 0;
 425
 426     for(iidx=0;iidx<4*DIM;iidx++)
 427     {
 428         scratch[iidx] = 0.0;
 429     }
 430
 431     /* Start outer loop over neighborlists */
 432     for(iidx=0; iidx<nri; iidx++)
 433     {
 434         /* Load shift vector for this list */
 435         i_shift_offset   = DIM*shiftidx[iidx];
 436
 437         /* Load limits for loop over neighbors */
 438         j_index_start    = jindex[iidx];
 439         j_index_end      = jindex[iidx+1];
 440
 441         /* Get outer coordinate index */
 442         inr              = iinr[iidx];
 443         i_coord_offset   = DIM*inr;
 444
 445         /* Load i particle coords and add shift vector */
 446         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 447
 448         fix0             = _mm256_setzero_pd();
 449         fiy0             = _mm256_setzero_pd();
 450         fiz0             = _mm256_setzero_pd();
 451
 452         /* Load parameters for i particles */
 453         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 454
 455         /* Start inner kernel loop */
 456         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 457         {
 458
 459             /* Get j neighbor index, and coordinate index */
 460             jnrA             = jjnr[jidx];
 461             jnrB             = jjnr[jidx+1];
 462             jnrC             = jjnr[jidx+2];
 463             jnrD             = jjnr[jidx+3];
 464             j_coord_offsetA  = DIM*jnrA;
 465             j_coord_offsetB  = DIM*jnrB;
 466             j_coord_offsetC  = DIM*jnrC;
 467             j_coord_offsetD  = DIM*jnrD;
 468
 469             /* load j atom coordinates */
 470             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 471                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 472                                                  &jx0,&jy0,&jz0);
 473
 474             /* Calculate displacement vector */
 475             dx00             = _mm256_sub_pd(ix0,jx0);
 476             dy00             = _mm256_sub_pd(iy0,jy0);
 477             dz00             = _mm256_sub_pd(iz0,jz0);
 478
 479             /* Calculate squared distance and things based on it */
 480             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 481
 482             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 483
 484             /* Load parameters for j particles */
 485             vdwjidx0A        = 2*vdwtype[jnrA+0];
 486             vdwjidx0B        = 2*vdwtype[jnrB+0];
 487             vdwjidx0C        = 2*vdwtype[jnrC+0];
 488             vdwjidx0D        = 2*vdwtype[jnrD+0];
 489
 490             /**************************
 491              * CALCULATE INTERACTIONS *
 492              **************************/
 493
 494             /* Compute parameters for interactions between i and j atoms */
 495             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 496                                             vdwioffsetptr0+vdwjidx0B,
 497                                             vdwioffsetptr0+vdwjidx0C,
 498                                             vdwioffsetptr0+vdwjidx0D,
 499                                             &c6_00,&c12_00);
 500
 501             /* LENNARD-JONES DISPERSION/REPULSION */
 502
 503             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 504             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 505
 506             fscal            = fvdw;
 507
 508             /* Calculate temporary vectorial force */
 509             tx               = _mm256_mul_pd(fscal,dx00);
 510             ty               = _mm256_mul_pd(fscal,dy00);
 511             tz               = _mm256_mul_pd(fscal,dz00);
 512
 513             /* Update vectorial force */
 514             fix0             = _mm256_add_pd(fix0,tx);
 515             fiy0             = _mm256_add_pd(fiy0,ty);
 516             fiz0             = _mm256_add_pd(fiz0,tz);
 517
 518             fjptrA             = f+j_coord_offsetA;
 519             fjptrB             = f+j_coord_offsetB;
 520             fjptrC             = f+j_coord_offsetC;
 521             fjptrD             = f+j_coord_offsetD;
 522             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 523
 524             /* Inner loop uses 27 flops */
 525         }
 526
 527         if(jidx<j_index_end)
 528         {
 529
 530             /* Get j neighbor index, and coordinate index */
 531             jnrlistA         = jjnr[jidx];
 532             jnrlistB         = jjnr[jidx+1];
 533             jnrlistC         = jjnr[jidx+2];
 534             jnrlistD         = jjnr[jidx+3];
 535             /* Sign of each element will be negative for non-real atoms.
 536              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 537              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 538              */
 539             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 540
 541             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 542             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 543             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 544
 545             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 546             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 547             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 548             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 549             j_coord_offsetA  = DIM*jnrA;
 550             j_coord_offsetB  = DIM*jnrB;
 551             j_coord_offsetC  = DIM*jnrC;
 552             j_coord_offsetD  = DIM*jnrD;
 553
 554             /* load j atom coordinates */
 555             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 556                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 557                                                  &jx0,&jy0,&jz0);
 558
 559             /* Calculate displacement vector */
 560             dx00             = _mm256_sub_pd(ix0,jx0);
 561             dy00             = _mm256_sub_pd(iy0,jy0);
 562             dz00             = _mm256_sub_pd(iz0,jz0);
 563
 564             /* Calculate squared distance and things based on it */
 565             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 566
 567             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 568
 569             /* Load parameters for j particles */
 570             vdwjidx0A        = 2*vdwtype[jnrA+0];
 571             vdwjidx0B        = 2*vdwtype[jnrB+0];
 572             vdwjidx0C        = 2*vdwtype[jnrC+0];
 573             vdwjidx0D        = 2*vdwtype[jnrD+0];
 574
 575             /**************************
 576              * CALCULATE INTERACTIONS *
 577              **************************/
 578
 579             /* Compute parameters for interactions between i and j atoms */
 580             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 581                                             vdwioffsetptr0+vdwjidx0B,
 582                                             vdwioffsetptr0+vdwjidx0C,
 583                                             vdwioffsetptr0+vdwjidx0D,
 584                                             &c6_00,&c12_00);
 585
 586             /* LENNARD-JONES DISPERSION/REPULSION */
 587
 588             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 589             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 590
 591             fscal            = fvdw;
 592
 593             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 594
 595             /* Calculate temporary vectorial force */
 596             tx               = _mm256_mul_pd(fscal,dx00);
 597             ty               = _mm256_mul_pd(fscal,dy00);
 598             tz               = _mm256_mul_pd(fscal,dz00);
 599
 600             /* Update vectorial force */
 601             fix0             = _mm256_add_pd(fix0,tx);
 602             fiy0             = _mm256_add_pd(fiy0,ty);
 603             fiz0             = _mm256_add_pd(fiz0,tz);
 604
 605             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 606             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 607             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 608             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 609             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 610
 611             /* Inner loop uses 27 flops */
 612         }
 613
 614         /* End of innermost loop */
 615
 616         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 617                                                  f+i_coord_offset,fshift+i_shift_offset);
 618
 619         /* Increment number of inner iterations */
 620         inneriter                  += j_index_end - j_index_start;
 621
 622         /* Outer loop uses 6 flops */
 623     }
 624
 625     /* Increment number of outer iterations */
 626     outeriter        += nri;
 627
 628     /* Update outer/inner flops */
 629
 630     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*27);
 631 }