src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecNone_VdwLJSh_GeomP1P1_avx_128_fma_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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_128_fma_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_128_fma_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSh_GeomP1P1_VF_avx_128_fma_double
  51  * Electrostatics interaction: None
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecNone_VdwLJSh_GeomP1P1_VF_avx_128_fma_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwjidx0A,vdwjidx0B;
  82     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  83     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  84     int              nvdwtype;
  85     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  86     int              *vdwtype;
  87     real             *vdwparam;
  88     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  89     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  90     __m128d          dummy_mask,cutoff_mask;
  91     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  92     __m128d          one     = _mm_set1_pd(1.0);
  93     __m128d          two     = _mm_set1_pd(2.0);
  94     x                = xx[0];
  95     f                = ff[0];
  96
  97     nri              = nlist->nri;
  98     iinr             = nlist->iinr;
  99     jindex           = nlist->jindex;
 100     jjnr             = nlist->jjnr;
 101     shiftidx         = nlist->shift;
 102     gid              = nlist->gid;
 103     shiftvec         = fr->shift_vec[0];
 104     fshift           = fr->fshift[0];
 105     nvdwtype         = fr->ntype;
 106     vdwparam         = fr->nbfp;
 107     vdwtype          = mdatoms->typeA;
 108
 109     rcutoff_scalar   = fr->ic->rvdw;
 110     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 111     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 112
 113     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 114     rvdw             = _mm_set1_pd(fr->ic->rvdw);
 115
 116     /* Avoid stupid compiler warnings */
 117     jnrA = jnrB = 0;
 118     j_coord_offsetA = 0;
 119     j_coord_offsetB = 0;
 120
 121     outeriter        = 0;
 122     inneriter        = 0;
 123
 124     /* Start outer loop over neighborlists */
 125     for(iidx=0; iidx<nri; iidx++)
 126     {
 127         /* Load shift vector for this list */
 128         i_shift_offset   = DIM*shiftidx[iidx];
 129
 130         /* Load limits for loop over neighbors */
 131         j_index_start    = jindex[iidx];
 132         j_index_end      = jindex[iidx+1];
 133
 134         /* Get outer coordinate index */
 135         inr              = iinr[iidx];
 136         i_coord_offset   = DIM*inr;
 137
 138         /* Load i particle coords and add shift vector */
 139         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 140
 141         fix0             = _mm_setzero_pd();
 142         fiy0             = _mm_setzero_pd();
 143         fiz0             = _mm_setzero_pd();
 144
 145         /* Load parameters for i particles */
 146         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 147
 148         /* Reset potential sums */
 149         vvdwsum          = _mm_setzero_pd();
 150
 151         /* Start inner kernel loop */
 152         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 153         {
 154
 155             /* Get j neighbor index, and coordinate index */
 156             jnrA             = jjnr[jidx];
 157             jnrB             = jjnr[jidx+1];
 158             j_coord_offsetA  = DIM*jnrA;
 159             j_coord_offsetB  = DIM*jnrB;
 160
 161             /* load j atom coordinates */
 162             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 163                                               &jx0,&jy0,&jz0);
 164
 165             /* Calculate displacement vector */
 166             dx00             = _mm_sub_pd(ix0,jx0);
 167             dy00             = _mm_sub_pd(iy0,jy0);
 168             dz00             = _mm_sub_pd(iz0,jz0);
 169
 170             /* Calculate squared distance and things based on it */
 171             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 172
 173             rinvsq00         = avx128fma_inv_d(rsq00);
 174
 175             /* Load parameters for j particles */
 176             vdwjidx0A        = 2*vdwtype[jnrA+0];
 177             vdwjidx0B        = 2*vdwtype[jnrB+0];
 178
 179             /**************************
 180              * CALCULATE INTERACTIONS *
 181              **************************/
 182
 183             if (gmx_mm_any_lt(rsq00,rcutoff2))
 184             {
 185
 186             /* Compute parameters for interactions between i and j atoms */
 187             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 188                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 189
 190             /* LENNARD-JONES DISPERSION/REPULSION */
 191
 192             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 193             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 194             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 195             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 196                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 197             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 198
 199             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 200
 201             /* Update potential sum for this i atom from the interaction with this j atom. */
 202             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 203             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 204
 205             fscal            = fvdw;
 206
 207             fscal            = _mm_and_pd(fscal,cutoff_mask);
 208
 209             /* Update vectorial force */
 210             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 211             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 212             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 213
 214             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 215                                                    _mm_mul_pd(dx00,fscal),
 216                                                    _mm_mul_pd(dy00,fscal),
 217                                                    _mm_mul_pd(dz00,fscal));
 218
 219             }
 220
 221             /* Inner loop uses 44 flops */
 222         }
 223
 224         if(jidx<j_index_end)
 225         {
 226
 227             jnrA             = jjnr[jidx];
 228             j_coord_offsetA  = DIM*jnrA;
 229
 230             /* load j atom coordinates */
 231             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 232                                               &jx0,&jy0,&jz0);
 233
 234             /* Calculate displacement vector */
 235             dx00             = _mm_sub_pd(ix0,jx0);
 236             dy00             = _mm_sub_pd(iy0,jy0);
 237             dz00             = _mm_sub_pd(iz0,jz0);
 238
 239             /* Calculate squared distance and things based on it */
 240             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 241
 242             rinvsq00         = avx128fma_inv_d(rsq00);
 243
 244             /* Load parameters for j particles */
 245             vdwjidx0A        = 2*vdwtype[jnrA+0];
 246
 247             /**************************
 248              * CALCULATE INTERACTIONS *
 249              **************************/
 250
 251             if (gmx_mm_any_lt(rsq00,rcutoff2))
 252             {
 253
 254             /* Compute parameters for interactions between i and j atoms */
 255             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 256
 257             /* LENNARD-JONES DISPERSION/REPULSION */
 258
 259             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 260             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 261             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 262             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 263                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 264             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 265
 266             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 267
 268             /* Update potential sum for this i atom from the interaction with this j atom. */
 269             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 270             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 271             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 272
 273             fscal            = fvdw;
 274
 275             fscal            = _mm_and_pd(fscal,cutoff_mask);
 276
 277             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 278
 279             /* Update vectorial force */
 280             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 281             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 282             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 283
 284             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 285                                                    _mm_mul_pd(dx00,fscal),
 286                                                    _mm_mul_pd(dy00,fscal),
 287                                                    _mm_mul_pd(dz00,fscal));
 288
 289             }
 290
 291             /* Inner loop uses 44 flops */
 292         }
 293
 294         /* End of innermost loop */
 295
 296         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 297                                               f+i_coord_offset,fshift+i_shift_offset);
 298
 299         ggid                        = gid[iidx];
 300         /* Update potential energies */
 301         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 302
 303         /* Increment number of inner iterations */
 304         inneriter                  += j_index_end - j_index_start;
 305
 306         /* Outer loop uses 7 flops */
 307     }
 308
 309     /* Increment number of outer iterations */
 310     outeriter        += nri;
 311
 312     /* Update outer/inner flops */
 313
 314     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*44);
 315 }
 316 /*
 317  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSh_GeomP1P1_F_avx_128_fma_double
 318  * Electrostatics interaction: None
 319  * VdW interaction:            LennardJones
 320  * Geometry:                   Particle-Particle
 321  * Calculate force/pot:        Force
 322  */
 323 void
 324 nb_kernel_ElecNone_VdwLJSh_GeomP1P1_F_avx_128_fma_double
 325                     (t_nblist                    * gmx_restrict       nlist,
 326                      rvec                        * gmx_restrict          xx,
 327                      rvec                        * gmx_restrict          ff,
 328                      struct t_forcerec           * gmx_restrict          fr,
 329                      t_mdatoms                   * gmx_restrict     mdatoms,
 330                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 331                      t_nrnb                      * gmx_restrict        nrnb)
 332 {
 333     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 334      * just 0 for non-waters.
 335      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 336      * jnr indices corresponding to data put in the four positions in the SIMD register.
 337      */
 338     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 339     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 340     int              jnrA,jnrB;
 341     int              j_coord_offsetA,j_coord_offsetB;
 342     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 343     real             rcutoff_scalar;
 344     real             *shiftvec,*fshift,*x,*f;
 345     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 346     int              vdwioffset0;
 347     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 348     int              vdwjidx0A,vdwjidx0B;
 349     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 350     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 351     int              nvdwtype;
 352     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 353     int              *vdwtype;
 354     real             *vdwparam;
 355     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 356     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 357     __m128d          dummy_mask,cutoff_mask;
 358     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 359     __m128d          one     = _mm_set1_pd(1.0);
 360     __m128d          two     = _mm_set1_pd(2.0);
 361     x                = xx[0];
 362     f                = ff[0];
 363
 364     nri              = nlist->nri;
 365     iinr             = nlist->iinr;
 366     jindex           = nlist->jindex;
 367     jjnr             = nlist->jjnr;
 368     shiftidx         = nlist->shift;
 369     gid              = nlist->gid;
 370     shiftvec         = fr->shift_vec[0];
 371     fshift           = fr->fshift[0];
 372     nvdwtype         = fr->ntype;
 373     vdwparam         = fr->nbfp;
 374     vdwtype          = mdatoms->typeA;
 375
 376     rcutoff_scalar   = fr->ic->rvdw;
 377     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 378     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 379
 380     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 381     rvdw             = _mm_set1_pd(fr->ic->rvdw);
 382
 383     /* Avoid stupid compiler warnings */
 384     jnrA = jnrB = 0;
 385     j_coord_offsetA = 0;
 386     j_coord_offsetB = 0;
 387
 388     outeriter        = 0;
 389     inneriter        = 0;
 390
 391     /* Start outer loop over neighborlists */
 392     for(iidx=0; iidx<nri; iidx++)
 393     {
 394         /* Load shift vector for this list */
 395         i_shift_offset   = DIM*shiftidx[iidx];
 396
 397         /* Load limits for loop over neighbors */
 398         j_index_start    = jindex[iidx];
 399         j_index_end      = jindex[iidx+1];
 400
 401         /* Get outer coordinate index */
 402         inr              = iinr[iidx];
 403         i_coord_offset   = DIM*inr;
 404
 405         /* Load i particle coords and add shift vector */
 406         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 407
 408         fix0             = _mm_setzero_pd();
 409         fiy0             = _mm_setzero_pd();
 410         fiz0             = _mm_setzero_pd();
 411
 412         /* Load parameters for i particles */
 413         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 414
 415         /* Start inner kernel loop */
 416         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 417         {
 418
 419             /* Get j neighbor index, and coordinate index */
 420             jnrA             = jjnr[jidx];
 421             jnrB             = jjnr[jidx+1];
 422             j_coord_offsetA  = DIM*jnrA;
 423             j_coord_offsetB  = DIM*jnrB;
 424
 425             /* load j atom coordinates */
 426             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 427                                               &jx0,&jy0,&jz0);
 428
 429             /* Calculate displacement vector */
 430             dx00             = _mm_sub_pd(ix0,jx0);
 431             dy00             = _mm_sub_pd(iy0,jy0);
 432             dz00             = _mm_sub_pd(iz0,jz0);
 433
 434             /* Calculate squared distance and things based on it */
 435             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 436
 437             rinvsq00         = avx128fma_inv_d(rsq00);
 438
 439             /* Load parameters for j particles */
 440             vdwjidx0A        = 2*vdwtype[jnrA+0];
 441             vdwjidx0B        = 2*vdwtype[jnrB+0];
 442
 443             /**************************
 444              * CALCULATE INTERACTIONS *
 445              **************************/
 446
 447             if (gmx_mm_any_lt(rsq00,rcutoff2))
 448             {
 449
 450             /* Compute parameters for interactions between i and j atoms */
 451             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 452                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 453
 454             /* LENNARD-JONES DISPERSION/REPULSION */
 455
 456             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 457             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 458
 459             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 460
 461             fscal            = fvdw;
 462
 463             fscal            = _mm_and_pd(fscal,cutoff_mask);
 464
 465             /* Update vectorial force */
 466             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 467             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 468             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 469
 470             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 471                                                    _mm_mul_pd(dx00,fscal),
 472                                                    _mm_mul_pd(dy00,fscal),
 473                                                    _mm_mul_pd(dz00,fscal));
 474
 475             }
 476
 477             /* Inner loop uses 33 flops */
 478         }
 479
 480         if(jidx<j_index_end)
 481         {
 482
 483             jnrA             = jjnr[jidx];
 484             j_coord_offsetA  = DIM*jnrA;
 485
 486             /* load j atom coordinates */
 487             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 488                                               &jx0,&jy0,&jz0);
 489
 490             /* Calculate displacement vector */
 491             dx00             = _mm_sub_pd(ix0,jx0);
 492             dy00             = _mm_sub_pd(iy0,jy0);
 493             dz00             = _mm_sub_pd(iz0,jz0);
 494
 495             /* Calculate squared distance and things based on it */
 496             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 497
 498             rinvsq00         = avx128fma_inv_d(rsq00);
 499
 500             /* Load parameters for j particles */
 501             vdwjidx0A        = 2*vdwtype[jnrA+0];
 502
 503             /**************************
 504              * CALCULATE INTERACTIONS *
 505              **************************/
 506
 507             if (gmx_mm_any_lt(rsq00,rcutoff2))
 508             {
 509
 510             /* Compute parameters for interactions between i and j atoms */
 511             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 512
 513             /* LENNARD-JONES DISPERSION/REPULSION */
 514
 515             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 516             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 517
 518             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 519
 520             fscal            = fvdw;
 521
 522             fscal            = _mm_and_pd(fscal,cutoff_mask);
 523
 524             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 525
 526             /* Update vectorial force */
 527             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 528             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 529             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 530
 531             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 532                                                    _mm_mul_pd(dx00,fscal),
 533                                                    _mm_mul_pd(dy00,fscal),
 534                                                    _mm_mul_pd(dz00,fscal));
 535
 536             }
 537
 538             /* Inner loop uses 33 flops */
 539         }
 540
 541         /* End of innermost loop */
 542
 543         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 544                                               f+i_coord_offset,fshift+i_shift_offset);
 545
 546         /* Increment number of inner iterations */
 547         inneriter                  += j_index_end - j_index_start;
 548
 549         /* Outer loop uses 6 flops */
 550     }
 551
 552     /* Increment number of outer iterations */
 553     outeriter        += nri;
 554
 555     /* Update outer/inner flops */
 556
 557     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*33);
 558 }