src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRF_VdwLJ_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_ElecRF_VdwLJ_GeomP1P1_VF_avx_128_fma_double
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRF_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     krf              = _mm_set1_pd(fr->ic->k_rf);
 110     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 111     crf              = _mm_set1_pd(fr->ic->c_rf);
 112     nvdwtype         = fr->ntype;
 113     vdwparam         = fr->nbfp;
 114     vdwtype          = mdatoms->typeA;
 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         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 147         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 148
 149         /* Reset potential sums */
 150         velecsum         = _mm_setzero_pd();
 151         vvdwsum          = _mm_setzero_pd();
 152
 153         /* Start inner kernel loop */
 154         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 155         {
 156
 157             /* Get j neighbor index, and coordinate index */
 158             jnrA             = jjnr[jidx];
 159             jnrB             = jjnr[jidx+1];
 160             j_coord_offsetA  = DIM*jnrA;
 161             j_coord_offsetB  = DIM*jnrB;
 162
 163             /* load j atom coordinates */
 164             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 165                                               &jx0,&jy0,&jz0);
 166
 167             /* Calculate displacement vector */
 168             dx00             = _mm_sub_pd(ix0,jx0);
 169             dy00             = _mm_sub_pd(iy0,jy0);
 170             dz00             = _mm_sub_pd(iz0,jz0);
 171
 172             /* Calculate squared distance and things based on it */
 173             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 174
 175             rinv00           = avx128fma_invsqrt_d(rsq00);
 176
 177             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 178
 179             /* Load parameters for j particles */
 180             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 181             vdwjidx0A        = 2*vdwtype[jnrA+0];
 182             vdwjidx0B        = 2*vdwtype[jnrB+0];
 183
 184             /**************************
 185              * CALCULATE INTERACTIONS *
 186              **************************/
 187
 188             /* Compute parameters for interactions between i and j atoms */
 189             qq00             = _mm_mul_pd(iq0,jq0);
 190             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 191                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 192
 193             /* REACTION-FIELD ELECTROSTATICS */
 194             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 195             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 196
 197             /* LENNARD-JONES DISPERSION/REPULSION */
 198
 199             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 200             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 201             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 202             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 203             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 204
 205             /* Update potential sum for this i atom from the interaction with this j atom. */
 206             velecsum         = _mm_add_pd(velecsum,velec);
 207             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 208
 209             fscal            = _mm_add_pd(felec,fvdw);
 210
 211             /* Update vectorial force */
 212             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 213             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 214             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 215
 216             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 217                                                    _mm_mul_pd(dx00,fscal),
 218                                                    _mm_mul_pd(dy00,fscal),
 219                                                    _mm_mul_pd(dz00,fscal));
 220
 221             /* Inner loop uses 47 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             rinv00           = avx128fma_invsqrt_d(rsq00);
 243
 244             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 245
 246             /* Load parameters for j particles */
 247             jq0              = _mm_load_sd(charge+jnrA+0);
 248             vdwjidx0A        = 2*vdwtype[jnrA+0];
 249
 250             /**************************
 251              * CALCULATE INTERACTIONS *
 252              **************************/
 253
 254             /* Compute parameters for interactions between i and j atoms */
 255             qq00             = _mm_mul_pd(iq0,jq0);
 256             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 257
 258             /* REACTION-FIELD ELECTROSTATICS */
 259             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 260             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 261
 262             /* LENNARD-JONES DISPERSION/REPULSION */
 263
 264             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 265             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 266             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 267             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 268             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 269
 270             /* Update potential sum for this i atom from the interaction with this j atom. */
 271             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 272             velecsum         = _mm_add_pd(velecsum,velec);
 273             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 274             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 275
 276             fscal            = _mm_add_pd(felec,fvdw);
 277
 278             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 279
 280             /* Update vectorial force */
 281             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 282             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 283             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 284
 285             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 286                                                    _mm_mul_pd(dx00,fscal),
 287                                                    _mm_mul_pd(dy00,fscal),
 288                                                    _mm_mul_pd(dz00,fscal));
 289
 290             /* Inner loop uses 47 flops */
 291         }
 292
 293         /* End of innermost loop */
 294
 295         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 296                                               f+i_coord_offset,fshift+i_shift_offset);
 297
 298         ggid                        = gid[iidx];
 299         /* Update potential energies */
 300         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 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 9 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_ELEC_VDW_VF,outeriter*9 + inneriter*47);
 315 }
 316 /*
 317  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_avx_128_fma_double
 318  * Electrostatics interaction: ReactionField
 319  * VdW interaction:            LennardJones
 320  * Geometry:                   Particle-Particle
 321  * Calculate force/pot:        Force
 322  */
 323 void
 324 nb_kernel_ElecRF_VdwLJ_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     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 352     real             *charge;
 353     int              nvdwtype;
 354     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 355     int              *vdwtype;
 356     real             *vdwparam;
 357     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 358     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 359     __m128d          dummy_mask,cutoff_mask;
 360     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 361     __m128d          one     = _mm_set1_pd(1.0);
 362     __m128d          two     = _mm_set1_pd(2.0);
 363     x                = xx[0];
 364     f                = ff[0];
 365
 366     nri              = nlist->nri;
 367     iinr             = nlist->iinr;
 368     jindex           = nlist->jindex;
 369     jjnr             = nlist->jjnr;
 370     shiftidx         = nlist->shift;
 371     gid              = nlist->gid;
 372     shiftvec         = fr->shift_vec[0];
 373     fshift           = fr->fshift[0];
 374     facel            = _mm_set1_pd(fr->ic->epsfac);
 375     charge           = mdatoms->chargeA;
 376     krf              = _mm_set1_pd(fr->ic->k_rf);
 377     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 378     crf              = _mm_set1_pd(fr->ic->c_rf);
 379     nvdwtype         = fr->ntype;
 380     vdwparam         = fr->nbfp;
 381     vdwtype          = mdatoms->typeA;
 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         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 414         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 415
 416         /* Start inner kernel loop */
 417         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 418         {
 419
 420             /* Get j neighbor index, and coordinate index */
 421             jnrA             = jjnr[jidx];
 422             jnrB             = jjnr[jidx+1];
 423             j_coord_offsetA  = DIM*jnrA;
 424             j_coord_offsetB  = DIM*jnrB;
 425
 426             /* load j atom coordinates */
 427             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 428                                               &jx0,&jy0,&jz0);
 429
 430             /* Calculate displacement vector */
 431             dx00             = _mm_sub_pd(ix0,jx0);
 432             dy00             = _mm_sub_pd(iy0,jy0);
 433             dz00             = _mm_sub_pd(iz0,jz0);
 434
 435             /* Calculate squared distance and things based on it */
 436             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 437
 438             rinv00           = avx128fma_invsqrt_d(rsq00);
 439
 440             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 441
 442             /* Load parameters for j particles */
 443             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 444             vdwjidx0A        = 2*vdwtype[jnrA+0];
 445             vdwjidx0B        = 2*vdwtype[jnrB+0];
 446
 447             /**************************
 448              * CALCULATE INTERACTIONS *
 449              **************************/
 450
 451             /* Compute parameters for interactions between i and j atoms */
 452             qq00             = _mm_mul_pd(iq0,jq0);
 453             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 454                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 455
 456             /* REACTION-FIELD ELECTROSTATICS */
 457             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 458
 459             /* LENNARD-JONES DISPERSION/REPULSION */
 460
 461             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 462             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 463
 464             fscal            = _mm_add_pd(felec,fvdw);
 465
 466             /* Update vectorial force */
 467             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 468             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 469             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 470
 471             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 472                                                    _mm_mul_pd(dx00,fscal),
 473                                                    _mm_mul_pd(dy00,fscal),
 474                                                    _mm_mul_pd(dz00,fscal));
 475
 476             /* Inner loop uses 37 flops */
 477         }
 478
 479         if(jidx<j_index_end)
 480         {
 481
 482             jnrA             = jjnr[jidx];
 483             j_coord_offsetA  = DIM*jnrA;
 484
 485             /* load j atom coordinates */
 486             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 487                                               &jx0,&jy0,&jz0);
 488
 489             /* Calculate displacement vector */
 490             dx00             = _mm_sub_pd(ix0,jx0);
 491             dy00             = _mm_sub_pd(iy0,jy0);
 492             dz00             = _mm_sub_pd(iz0,jz0);
 493
 494             /* Calculate squared distance and things based on it */
 495             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 496
 497             rinv00           = avx128fma_invsqrt_d(rsq00);
 498
 499             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 500
 501             /* Load parameters for j particles */
 502             jq0              = _mm_load_sd(charge+jnrA+0);
 503             vdwjidx0A        = 2*vdwtype[jnrA+0];
 504
 505             /**************************
 506              * CALCULATE INTERACTIONS *
 507              **************************/
 508
 509             /* Compute parameters for interactions between i and j atoms */
 510             qq00             = _mm_mul_pd(iq0,jq0);
 511             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 512
 513             /* REACTION-FIELD ELECTROSTATICS */
 514             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 515
 516             /* LENNARD-JONES DISPERSION/REPULSION */
 517
 518             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 519             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 520
 521             fscal            = _mm_add_pd(felec,fvdw);
 522
 523             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 524
 525             /* Update vectorial force */
 526             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 527             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 528             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 529
 530             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 531                                                    _mm_mul_pd(dx00,fscal),
 532                                                    _mm_mul_pd(dy00,fscal),
 533                                                    _mm_mul_pd(dz00,fscal));
 534
 535             /* Inner loop uses 37 flops */
 536         }
 537
 538         /* End of innermost loop */
 539
 540         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 541                                               f+i_coord_offset,fshift+i_shift_offset);
 542
 543         /* Increment number of inner iterations */
 544         inneriter                  += j_index_end - j_index_start;
 545
 546         /* Outer loop uses 7 flops */
 547     }
 548
 549     /* Increment number of outer iterations */
 550     outeriter        += nri;
 551
 552     /* Update outer/inner flops */
 553
 554     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*37);
 555 }