src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCoul_VdwLJ_GeomP1P1_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_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_ElecCoul_VdwLJ_GeomP1P1_VF_avx_128_fma_double
  51  * Electrostatics interaction: Coulomb
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCoul_VdwLJ_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     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  85     real             *charge;
  86     int              nvdwtype;
  87     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  88     int              *vdwtype;
  89     real             *vdwparam;
  90     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  91     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  92     __m128d          dummy_mask,cutoff_mask;
  93     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  94     __m128d          one     = _mm_set1_pd(1.0);
  95     __m128d          two     = _mm_set1_pd(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm_set1_pd(fr->ic->epsfac);
 108     charge           = mdatoms->chargeA;
 109     nvdwtype         = fr->ntype;
 110     vdwparam         = fr->nbfp;
 111     vdwtype          = mdatoms->typeA;
 112
 113     /* Avoid stupid compiler warnings */
 114     jnrA = jnrB = 0;
 115     j_coord_offsetA = 0;
 116     j_coord_offsetB = 0;
 117
 118     outeriter        = 0;
 119     inneriter        = 0;
 120
 121     /* Start outer loop over neighborlists */
 122     for(iidx=0; iidx<nri; iidx++)
 123     {
 124         /* Load shift vector for this list */
 125         i_shift_offset   = DIM*shiftidx[iidx];
 126
 127         /* Load limits for loop over neighbors */
 128         j_index_start    = jindex[iidx];
 129         j_index_end      = jindex[iidx+1];
 130
 131         /* Get outer coordinate index */
 132         inr              = iinr[iidx];
 133         i_coord_offset   = DIM*inr;
 134
 135         /* Load i particle coords and add shift vector */
 136         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 137
 138         fix0             = _mm_setzero_pd();
 139         fiy0             = _mm_setzero_pd();
 140         fiz0             = _mm_setzero_pd();
 141
 142         /* Load parameters for i particles */
 143         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 144         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 145
 146         /* Reset potential sums */
 147         velecsum         = _mm_setzero_pd();
 148         vvdwsum          = _mm_setzero_pd();
 149
 150         /* Start inner kernel loop */
 151         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 152         {
 153
 154             /* Get j neighbor index, and coordinate index */
 155             jnrA             = jjnr[jidx];
 156             jnrB             = jjnr[jidx+1];
 157             j_coord_offsetA  = DIM*jnrA;
 158             j_coord_offsetB  = DIM*jnrB;
 159
 160             /* load j atom coordinates */
 161             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 162                                               &jx0,&jy0,&jz0);
 163
 164             /* Calculate displacement vector */
 165             dx00             = _mm_sub_pd(ix0,jx0);
 166             dy00             = _mm_sub_pd(iy0,jy0);
 167             dz00             = _mm_sub_pd(iz0,jz0);
 168
 169             /* Calculate squared distance and things based on it */
 170             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 171
 172             rinv00           = avx128fma_invsqrt_d(rsq00);
 173
 174             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 175
 176             /* Load parameters for j particles */
 177             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 178             vdwjidx0A        = 2*vdwtype[jnrA+0];
 179             vdwjidx0B        = 2*vdwtype[jnrB+0];
 180
 181             /**************************
 182              * CALCULATE INTERACTIONS *
 183              **************************/
 184
 185             /* Compute parameters for interactions between i and j atoms */
 186             qq00             = _mm_mul_pd(iq0,jq0);
 187             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 188                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 189
 190             /* COULOMB ELECTROSTATICS */
 191             velec            = _mm_mul_pd(qq00,rinv00);
 192             felec            = _mm_mul_pd(velec,rinvsq00);
 193
 194             /* LENNARD-JONES DISPERSION/REPULSION */
 195
 196             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 197             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 198             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 199             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 200             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 201
 202             /* Update potential sum for this i atom from the interaction with this j atom. */
 203             velecsum         = _mm_add_pd(velecsum,velec);
 204             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 205
 206             fscal            = _mm_add_pd(felec,fvdw);
 207
 208             /* Update vectorial force */
 209             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 210             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 211             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 212
 213             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 214                                                    _mm_mul_pd(dx00,fscal),
 215                                                    _mm_mul_pd(dy00,fscal),
 216                                                    _mm_mul_pd(dz00,fscal));
 217
 218             /* Inner loop uses 43 flops */
 219         }
 220
 221         if(jidx<j_index_end)
 222         {
 223
 224             jnrA             = jjnr[jidx];
 225             j_coord_offsetA  = DIM*jnrA;
 226
 227             /* load j atom coordinates */
 228             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 229                                               &jx0,&jy0,&jz0);
 230
 231             /* Calculate displacement vector */
 232             dx00             = _mm_sub_pd(ix0,jx0);
 233             dy00             = _mm_sub_pd(iy0,jy0);
 234             dz00             = _mm_sub_pd(iz0,jz0);
 235
 236             /* Calculate squared distance and things based on it */
 237             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 238
 239             rinv00           = avx128fma_invsqrt_d(rsq00);
 240
 241             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 242
 243             /* Load parameters for j particles */
 244             jq0              = _mm_load_sd(charge+jnrA+0);
 245             vdwjidx0A        = 2*vdwtype[jnrA+0];
 246
 247             /**************************
 248              * CALCULATE INTERACTIONS *
 249              **************************/
 250
 251             /* Compute parameters for interactions between i and j atoms */
 252             qq00             = _mm_mul_pd(iq0,jq0);
 253             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 254
 255             /* COULOMB ELECTROSTATICS */
 256             velec            = _mm_mul_pd(qq00,rinv00);
 257             felec            = _mm_mul_pd(velec,rinvsq00);
 258
 259             /* LENNARD-JONES DISPERSION/REPULSION */
 260
 261             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 262             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 263             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 264             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 265             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 266
 267             /* Update potential sum for this i atom from the interaction with this j atom. */
 268             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 269             velecsum         = _mm_add_pd(velecsum,velec);
 270             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 271             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 272
 273             fscal            = _mm_add_pd(felec,fvdw);
 274
 275             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 276
 277             /* Update vectorial force */
 278             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 279             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 280             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 281
 282             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 283                                                    _mm_mul_pd(dx00,fscal),
 284                                                    _mm_mul_pd(dy00,fscal),
 285                                                    _mm_mul_pd(dz00,fscal));
 286
 287             /* Inner loop uses 43 flops */
 288         }
 289
 290         /* End of innermost loop */
 291
 292         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 293                                               f+i_coord_offset,fshift+i_shift_offset);
 294
 295         ggid                        = gid[iidx];
 296         /* Update potential energies */
 297         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 298         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 299
 300         /* Increment number of inner iterations */
 301         inneriter                  += j_index_end - j_index_start;
 302
 303         /* Outer loop uses 9 flops */
 304     }
 305
 306     /* Increment number of outer iterations */
 307     outeriter        += nri;
 308
 309     /* Update outer/inner flops */
 310
 311     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*43);
 312 }
 313 /*
 314  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomP1P1_F_avx_128_fma_double
 315  * Electrostatics interaction: Coulomb
 316  * VdW interaction:            LennardJones
 317  * Geometry:                   Particle-Particle
 318  * Calculate force/pot:        Force
 319  */
 320 void
 321 nb_kernel_ElecCoul_VdwLJ_GeomP1P1_F_avx_128_fma_double
 322                     (t_nblist                    * gmx_restrict       nlist,
 323                      rvec                        * gmx_restrict          xx,
 324                      rvec                        * gmx_restrict          ff,
 325                      struct t_forcerec           * gmx_restrict          fr,
 326                      t_mdatoms                   * gmx_restrict     mdatoms,
 327                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 328                      t_nrnb                      * gmx_restrict        nrnb)
 329 {
 330     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 331      * just 0 for non-waters.
 332      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 333      * jnr indices corresponding to data put in the four positions in the SIMD register.
 334      */
 335     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 336     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 337     int              jnrA,jnrB;
 338     int              j_coord_offsetA,j_coord_offsetB;
 339     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 340     real             rcutoff_scalar;
 341     real             *shiftvec,*fshift,*x,*f;
 342     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 343     int              vdwioffset0;
 344     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 345     int              vdwjidx0A,vdwjidx0B;
 346     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 347     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 348     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 349     real             *charge;
 350     int              nvdwtype;
 351     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 352     int              *vdwtype;
 353     real             *vdwparam;
 354     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 355     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 356     __m128d          dummy_mask,cutoff_mask;
 357     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 358     __m128d          one     = _mm_set1_pd(1.0);
 359     __m128d          two     = _mm_set1_pd(2.0);
 360     x                = xx[0];
 361     f                = ff[0];
 362
 363     nri              = nlist->nri;
 364     iinr             = nlist->iinr;
 365     jindex           = nlist->jindex;
 366     jjnr             = nlist->jjnr;
 367     shiftidx         = nlist->shift;
 368     gid              = nlist->gid;
 369     shiftvec         = fr->shift_vec[0];
 370     fshift           = fr->fshift[0];
 371     facel            = _mm_set1_pd(fr->ic->epsfac);
 372     charge           = mdatoms->chargeA;
 373     nvdwtype         = fr->ntype;
 374     vdwparam         = fr->nbfp;
 375     vdwtype          = mdatoms->typeA;
 376
 377     /* Avoid stupid compiler warnings */
 378     jnrA = jnrB = 0;
 379     j_coord_offsetA = 0;
 380     j_coord_offsetB = 0;
 381
 382     outeriter        = 0;
 383     inneriter        = 0;
 384
 385     /* Start outer loop over neighborlists */
 386     for(iidx=0; iidx<nri; iidx++)
 387     {
 388         /* Load shift vector for this list */
 389         i_shift_offset   = DIM*shiftidx[iidx];
 390
 391         /* Load limits for loop over neighbors */
 392         j_index_start    = jindex[iidx];
 393         j_index_end      = jindex[iidx+1];
 394
 395         /* Get outer coordinate index */
 396         inr              = iinr[iidx];
 397         i_coord_offset   = DIM*inr;
 398
 399         /* Load i particle coords and add shift vector */
 400         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 401
 402         fix0             = _mm_setzero_pd();
 403         fiy0             = _mm_setzero_pd();
 404         fiz0             = _mm_setzero_pd();
 405
 406         /* Load parameters for i particles */
 407         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 408         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 409
 410         /* Start inner kernel loop */
 411         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 412         {
 413
 414             /* Get j neighbor index, and coordinate index */
 415             jnrA             = jjnr[jidx];
 416             jnrB             = jjnr[jidx+1];
 417             j_coord_offsetA  = DIM*jnrA;
 418             j_coord_offsetB  = DIM*jnrB;
 419
 420             /* load j atom coordinates */
 421             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 422                                               &jx0,&jy0,&jz0);
 423
 424             /* Calculate displacement vector */
 425             dx00             = _mm_sub_pd(ix0,jx0);
 426             dy00             = _mm_sub_pd(iy0,jy0);
 427             dz00             = _mm_sub_pd(iz0,jz0);
 428
 429             /* Calculate squared distance and things based on it */
 430             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 431
 432             rinv00           = avx128fma_invsqrt_d(rsq00);
 433
 434             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 435
 436             /* Load parameters for j particles */
 437             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 438             vdwjidx0A        = 2*vdwtype[jnrA+0];
 439             vdwjidx0B        = 2*vdwtype[jnrB+0];
 440
 441             /**************************
 442              * CALCULATE INTERACTIONS *
 443              **************************/
 444
 445             /* Compute parameters for interactions between i and j atoms */
 446             qq00             = _mm_mul_pd(iq0,jq0);
 447             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 448                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 449
 450             /* COULOMB ELECTROSTATICS */
 451             velec            = _mm_mul_pd(qq00,rinv00);
 452             felec            = _mm_mul_pd(velec,rinvsq00);
 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             fscal            = _mm_add_pd(felec,fvdw);
 460
 461             /* Update vectorial force */
 462             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 463             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 464             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 465
 466             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 467                                                    _mm_mul_pd(dx00,fscal),
 468                                                    _mm_mul_pd(dy00,fscal),
 469                                                    _mm_mul_pd(dz00,fscal));
 470
 471             /* Inner loop uses 37 flops */
 472         }
 473
 474         if(jidx<j_index_end)
 475         {
 476
 477             jnrA             = jjnr[jidx];
 478             j_coord_offsetA  = DIM*jnrA;
 479
 480             /* load j atom coordinates */
 481             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 482                                               &jx0,&jy0,&jz0);
 483
 484             /* Calculate displacement vector */
 485             dx00             = _mm_sub_pd(ix0,jx0);
 486             dy00             = _mm_sub_pd(iy0,jy0);
 487             dz00             = _mm_sub_pd(iz0,jz0);
 488
 489             /* Calculate squared distance and things based on it */
 490             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 491
 492             rinv00           = avx128fma_invsqrt_d(rsq00);
 493
 494             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 495
 496             /* Load parameters for j particles */
 497             jq0              = _mm_load_sd(charge+jnrA+0);
 498             vdwjidx0A        = 2*vdwtype[jnrA+0];
 499
 500             /**************************
 501              * CALCULATE INTERACTIONS *
 502              **************************/
 503
 504             /* Compute parameters for interactions between i and j atoms */
 505             qq00             = _mm_mul_pd(iq0,jq0);
 506             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 507
 508             /* COULOMB ELECTROSTATICS */
 509             velec            = _mm_mul_pd(qq00,rinv00);
 510             felec            = _mm_mul_pd(velec,rinvsq00);
 511
 512             /* LENNARD-JONES DISPERSION/REPULSION */
 513
 514             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 515             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 516
 517             fscal            = _mm_add_pd(felec,fvdw);
 518
 519             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 520
 521             /* Update vectorial force */
 522             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 523             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 524             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 525
 526             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 527                                                    _mm_mul_pd(dx00,fscal),
 528                                                    _mm_mul_pd(dy00,fscal),
 529                                                    _mm_mul_pd(dz00,fscal));
 530
 531             /* Inner loop uses 37 flops */
 532         }
 533
 534         /* End of innermost loop */
 535
 536         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 537                                               f+i_coord_offset,fshift+i_shift_offset);
 538
 539         /* Increment number of inner iterations */
 540         inneriter                  += j_index_end - j_index_start;
 541
 542         /* Outer loop uses 7 flops */
 543     }
 544
 545     /* Increment number of outer iterations */
 546     outeriter        += nri;
 547
 548     /* Update outer/inner flops */
 549
 550     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*37);
 551 }