src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRF_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, 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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  48 #include "kernelutil_x86_avx_128_fma_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomP1P1_VF_avx_128_fma_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRF_VdwLJ_GeomP1P1_VF_avx_128_fma_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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwjidx0A,vdwjidx0B;
  83     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  84     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  85     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  86     real             *charge;
  87     int              nvdwtype;
  88     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  89     int              *vdwtype;
  90     real             *vdwparam;
  91     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  92     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  93     __m128d          dummy_mask,cutoff_mask;
  94     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  95     __m128d          one     = _mm_set1_pd(1.0);
  96     __m128d          two     = _mm_set1_pd(2.0);
  97     x                = xx[0];
  98     f                = ff[0];
  99
 100     nri              = nlist->nri;
 101     iinr             = nlist->iinr;
 102     jindex           = nlist->jindex;
 103     jjnr             = nlist->jjnr;
 104     shiftidx         = nlist->shift;
 105     gid              = nlist->gid;
 106     shiftvec         = fr->shift_vec[0];
 107     fshift           = fr->fshift[0];
 108     facel            = _mm_set1_pd(fr->epsfac);
 109     charge           = mdatoms->chargeA;
 110     krf              = _mm_set1_pd(fr->ic->k_rf);
 111     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 112     crf              = _mm_set1_pd(fr->ic->c_rf);
 113     nvdwtype         = fr->ntype;
 114     vdwparam         = fr->nbfp;
 115     vdwtype          = mdatoms->typeA;
 116
 117     /* Avoid stupid compiler warnings */
 118     jnrA = jnrB = 0;
 119     j_coord_offsetA = 0;
 120     j_coord_offsetB = 0;
 121
 122     outeriter        = 0;
 123     inneriter        = 0;
 124
 125     /* Start outer loop over neighborlists */
 126     for(iidx=0; iidx<nri; iidx++)
 127     {
 128         /* Load shift vector for this list */
 129         i_shift_offset   = DIM*shiftidx[iidx];
 130
 131         /* Load limits for loop over neighbors */
 132         j_index_start    = jindex[iidx];
 133         j_index_end      = jindex[iidx+1];
 134
 135         /* Get outer coordinate index */
 136         inr              = iinr[iidx];
 137         i_coord_offset   = DIM*inr;
 138
 139         /* Load i particle coords and add shift vector */
 140         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 141
 142         fix0             = _mm_setzero_pd();
 143         fiy0             = _mm_setzero_pd();
 144         fiz0             = _mm_setzero_pd();
 145
 146         /* Load parameters for i particles */
 147         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 148         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 149
 150         /* Reset potential sums */
 151         velecsum         = _mm_setzero_pd();
 152         vvdwsum          = _mm_setzero_pd();
 153
 154         /* Start inner kernel loop */
 155         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 156         {
 157
 158             /* Get j neighbor index, and coordinate index */
 159             jnrA             = jjnr[jidx];
 160             jnrB             = jjnr[jidx+1];
 161             j_coord_offsetA  = DIM*jnrA;
 162             j_coord_offsetB  = DIM*jnrB;
 163
 164             /* load j atom coordinates */
 165             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 166                                               &jx0,&jy0,&jz0);
 167
 168             /* Calculate displacement vector */
 169             dx00             = _mm_sub_pd(ix0,jx0);
 170             dy00             = _mm_sub_pd(iy0,jy0);
 171             dz00             = _mm_sub_pd(iz0,jz0);
 172
 173             /* Calculate squared distance and things based on it */
 174             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 175
 176             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 177
 178             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 179
 180             /* Load parameters for j particles */
 181             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 182             vdwjidx0A        = 2*vdwtype[jnrA+0];
 183             vdwjidx0B        = 2*vdwtype[jnrB+0];
 184
 185             /**************************
 186              * CALCULATE INTERACTIONS *
 187              **************************/
 188
 189             /* Compute parameters for interactions between i and j atoms */
 190             qq00             = _mm_mul_pd(iq0,jq0);
 191             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 192                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 193
 194             /* REACTION-FIELD ELECTROSTATICS */
 195             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 196             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 197
 198             /* LENNARD-JONES DISPERSION/REPULSION */
 199
 200             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 201             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 202             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 203             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 204             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 205
 206             /* Update potential sum for this i atom from the interaction with this j atom. */
 207             velecsum         = _mm_add_pd(velecsum,velec);
 208             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 209
 210             fscal            = _mm_add_pd(felec,fvdw);
 211
 212             /* Update vectorial force */
 213             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 214             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 215             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 216
 217             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 218                                                    _mm_mul_pd(dx00,fscal),
 219                                                    _mm_mul_pd(dy00,fscal),
 220                                                    _mm_mul_pd(dz00,fscal));
 221
 222             /* Inner loop uses 47 flops */
 223         }
 224
 225         if(jidx<j_index_end)
 226         {
 227
 228             jnrA             = jjnr[jidx];
 229             j_coord_offsetA  = DIM*jnrA;
 230
 231             /* load j atom coordinates */
 232             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 233                                               &jx0,&jy0,&jz0);
 234
 235             /* Calculate displacement vector */
 236             dx00             = _mm_sub_pd(ix0,jx0);
 237             dy00             = _mm_sub_pd(iy0,jy0);
 238             dz00             = _mm_sub_pd(iz0,jz0);
 239
 240             /* Calculate squared distance and things based on it */
 241             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 242
 243             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 244
 245             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 246
 247             /* Load parameters for j particles */
 248             jq0              = _mm_load_sd(charge+jnrA+0);
 249             vdwjidx0A        = 2*vdwtype[jnrA+0];
 250
 251             /**************************
 252              * CALCULATE INTERACTIONS *
 253              **************************/
 254
 255             /* Compute parameters for interactions between i and j atoms */
 256             qq00             = _mm_mul_pd(iq0,jq0);
 257             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 258
 259             /* REACTION-FIELD ELECTROSTATICS */
 260             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 261             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 262
 263             /* LENNARD-JONES DISPERSION/REPULSION */
 264
 265             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 266             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 267             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 268             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 269             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 270
 271             /* Update potential sum for this i atom from the interaction with this j atom. */
 272             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 273             velecsum         = _mm_add_pd(velecsum,velec);
 274             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 275             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 276
 277             fscal            = _mm_add_pd(felec,fvdw);
 278
 279             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 280
 281             /* Update vectorial force */
 282             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 283             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 284             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 285
 286             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 287                                                    _mm_mul_pd(dx00,fscal),
 288                                                    _mm_mul_pd(dy00,fscal),
 289                                                    _mm_mul_pd(dz00,fscal));
 290
 291             /* Inner loop uses 47 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(velecsum,kernel_data->energygrp_elec+ggid);
 302         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 303
 304         /* Increment number of inner iterations */
 305         inneriter                  += j_index_end - j_index_start;
 306
 307         /* Outer loop uses 9 flops */
 308     }
 309
 310     /* Increment number of outer iterations */
 311     outeriter        += nri;
 312
 313     /* Update outer/inner flops */
 314
 315     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*47);
 316 }
 317 /*
 318  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_avx_128_fma_double
 319  * Electrostatics interaction: ReactionField
 320  * VdW interaction:            LennardJones
 321  * Geometry:                   Particle-Particle
 322  * Calculate force/pot:        Force
 323  */
 324 void
 325 nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_avx_128_fma_double
 326                     (t_nblist                    * gmx_restrict       nlist,
 327                      rvec                        * gmx_restrict          xx,
 328                      rvec                        * gmx_restrict          ff,
 329                      t_forcerec                  * gmx_restrict          fr,
 330                      t_mdatoms                   * gmx_restrict     mdatoms,
 331                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 332                      t_nrnb                      * gmx_restrict        nrnb)
 333 {
 334     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 335      * just 0 for non-waters.
 336      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 337      * jnr indices corresponding to data put in the four positions in the SIMD register.
 338      */
 339     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 340     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 341     int              jnrA,jnrB;
 342     int              j_coord_offsetA,j_coord_offsetB;
 343     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 344     real             rcutoff_scalar;
 345     real             *shiftvec,*fshift,*x,*f;
 346     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 347     int              vdwioffset0;
 348     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 349     int              vdwjidx0A,vdwjidx0B;
 350     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 351     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 352     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 353     real             *charge;
 354     int              nvdwtype;
 355     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 356     int              *vdwtype;
 357     real             *vdwparam;
 358     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 359     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 360     __m128d          dummy_mask,cutoff_mask;
 361     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 362     __m128d          one     = _mm_set1_pd(1.0);
 363     __m128d          two     = _mm_set1_pd(2.0);
 364     x                = xx[0];
 365     f                = ff[0];
 366
 367     nri              = nlist->nri;
 368     iinr             = nlist->iinr;
 369     jindex           = nlist->jindex;
 370     jjnr             = nlist->jjnr;
 371     shiftidx         = nlist->shift;
 372     gid              = nlist->gid;
 373     shiftvec         = fr->shift_vec[0];
 374     fshift           = fr->fshift[0];
 375     facel            = _mm_set1_pd(fr->epsfac);
 376     charge           = mdatoms->chargeA;
 377     krf              = _mm_set1_pd(fr->ic->k_rf);
 378     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 379     crf              = _mm_set1_pd(fr->ic->c_rf);
 380     nvdwtype         = fr->ntype;
 381     vdwparam         = fr->nbfp;
 382     vdwtype          = mdatoms->typeA;
 383
 384     /* Avoid stupid compiler warnings */
 385     jnrA = jnrB = 0;
 386     j_coord_offsetA = 0;
 387     j_coord_offsetB = 0;
 388
 389     outeriter        = 0;
 390     inneriter        = 0;
 391
 392     /* Start outer loop over neighborlists */
 393     for(iidx=0; iidx<nri; iidx++)
 394     {
 395         /* Load shift vector for this list */
 396         i_shift_offset   = DIM*shiftidx[iidx];
 397
 398         /* Load limits for loop over neighbors */
 399         j_index_start    = jindex[iidx];
 400         j_index_end      = jindex[iidx+1];
 401
 402         /* Get outer coordinate index */
 403         inr              = iinr[iidx];
 404         i_coord_offset   = DIM*inr;
 405
 406         /* Load i particle coords and add shift vector */
 407         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 408
 409         fix0             = _mm_setzero_pd();
 410         fiy0             = _mm_setzero_pd();
 411         fiz0             = _mm_setzero_pd();
 412
 413         /* Load parameters for i particles */
 414         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 415         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 416
 417         /* Start inner kernel loop */
 418         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 419         {
 420
 421             /* Get j neighbor index, and coordinate index */
 422             jnrA             = jjnr[jidx];
 423             jnrB             = jjnr[jidx+1];
 424             j_coord_offsetA  = DIM*jnrA;
 425             j_coord_offsetB  = DIM*jnrB;
 426
 427             /* load j atom coordinates */
 428             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 429                                               &jx0,&jy0,&jz0);
 430
 431             /* Calculate displacement vector */
 432             dx00             = _mm_sub_pd(ix0,jx0);
 433             dy00             = _mm_sub_pd(iy0,jy0);
 434             dz00             = _mm_sub_pd(iz0,jz0);
 435
 436             /* Calculate squared distance and things based on it */
 437             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 438
 439             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 440
 441             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 442
 443             /* Load parameters for j particles */
 444             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 445             vdwjidx0A        = 2*vdwtype[jnrA+0];
 446             vdwjidx0B        = 2*vdwtype[jnrB+0];
 447
 448             /**************************
 449              * CALCULATE INTERACTIONS *
 450              **************************/
 451
 452             /* Compute parameters for interactions between i and j atoms */
 453             qq00             = _mm_mul_pd(iq0,jq0);
 454             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 455                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 456
 457             /* REACTION-FIELD ELECTROSTATICS */
 458             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 459
 460             /* LENNARD-JONES DISPERSION/REPULSION */
 461
 462             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 463             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 464
 465             fscal            = _mm_add_pd(felec,fvdw);
 466
 467             /* Update vectorial force */
 468             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 469             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 470             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 471
 472             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 473                                                    _mm_mul_pd(dx00,fscal),
 474                                                    _mm_mul_pd(dy00,fscal),
 475                                                    _mm_mul_pd(dz00,fscal));
 476
 477             /* Inner loop uses 37 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             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 499
 500             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 501
 502             /* Load parameters for j particles */
 503             jq0              = _mm_load_sd(charge+jnrA+0);
 504             vdwjidx0A        = 2*vdwtype[jnrA+0];
 505
 506             /**************************
 507              * CALCULATE INTERACTIONS *
 508              **************************/
 509
 510             /* Compute parameters for interactions between i and j atoms */
 511             qq00             = _mm_mul_pd(iq0,jq0);
 512             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 513
 514             /* REACTION-FIELD ELECTROSTATICS */
 515             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 516
 517             /* LENNARD-JONES DISPERSION/REPULSION */
 518
 519             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 520             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 521
 522             fscal            = _mm_add_pd(felec,fvdw);
 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             /* Inner loop uses 37 flops */
 537         }
 538
 539         /* End of innermost loop */
 540
 541         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 542                                               f+i_coord_offset,fshift+i_shift_offset);
 543
 544         /* Increment number of inner iterations */
 545         inneriter                  += j_index_end - j_index_start;
 546
 547         /* Outer loop uses 7 flops */
 548     }
 549
 550     /* Increment number of outer iterations */
 551     outeriter        += nri;
 552
 553     /* Update outer/inner flops */
 554
 555     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*37);
 556 }