src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecRF_VdwLJ_GeomP1P1_sse4_1_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 sse4_1_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_sse4_1_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomP1P1_VF_sse4_1_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_sse4_1_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           = sse41_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_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 195             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_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_sub_pd( _mm_mul_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             /* Calculate temporary vectorial force */
 212             tx               = _mm_mul_pd(fscal,dx00);
 213             ty               = _mm_mul_pd(fscal,dy00);
 214             tz               = _mm_mul_pd(fscal,dz00);
 215
 216             /* Update vectorial force */
 217             fix0             = _mm_add_pd(fix0,tx);
 218             fiy0             = _mm_add_pd(fiy0,ty);
 219             fiz0             = _mm_add_pd(fiz0,tz);
 220
 221             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 222
 223             /* Inner loop uses 44 flops */
 224         }
 225
 226         if(jidx<j_index_end)
 227         {
 228
 229             jnrA             = jjnr[jidx];
 230             j_coord_offsetA  = DIM*jnrA;
 231
 232             /* load j atom coordinates */
 233             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 234                                               &jx0,&jy0,&jz0);
 235
 236             /* Calculate displacement vector */
 237             dx00             = _mm_sub_pd(ix0,jx0);
 238             dy00             = _mm_sub_pd(iy0,jy0);
 239             dz00             = _mm_sub_pd(iz0,jz0);
 240
 241             /* Calculate squared distance and things based on it */
 242             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 243
 244             rinv00           = sse41_invsqrt_d(rsq00);
 245
 246             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 247
 248             /* Load parameters for j particles */
 249             jq0              = _mm_load_sd(charge+jnrA+0);
 250             vdwjidx0A        = 2*vdwtype[jnrA+0];
 251
 252             /**************************
 253              * CALCULATE INTERACTIONS *
 254              **************************/
 255
 256             /* Compute parameters for interactions between i and j atoms */
 257             qq00             = _mm_mul_pd(iq0,jq0);
 258             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 259
 260             /* REACTION-FIELD ELECTROSTATICS */
 261             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 262             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 263
 264             /* LENNARD-JONES DISPERSION/REPULSION */
 265
 266             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 267             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 268             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 269             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 270             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 271
 272             /* Update potential sum for this i atom from the interaction with this j atom. */
 273             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 274             velecsum         = _mm_add_pd(velecsum,velec);
 275             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 276             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 277
 278             fscal            = _mm_add_pd(felec,fvdw);
 279
 280             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 281
 282             /* Calculate temporary vectorial force */
 283             tx               = _mm_mul_pd(fscal,dx00);
 284             ty               = _mm_mul_pd(fscal,dy00);
 285             tz               = _mm_mul_pd(fscal,dz00);
 286
 287             /* Update vectorial force */
 288             fix0             = _mm_add_pd(fix0,tx);
 289             fiy0             = _mm_add_pd(fiy0,ty);
 290             fiz0             = _mm_add_pd(fiz0,tz);
 291
 292             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 293
 294             /* Inner loop uses 44 flops */
 295         }
 296
 297         /* End of innermost loop */
 298
 299         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 300                                               f+i_coord_offset,fshift+i_shift_offset);
 301
 302         ggid                        = gid[iidx];
 303         /* Update potential energies */
 304         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 305         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 306
 307         /* Increment number of inner iterations */
 308         inneriter                  += j_index_end - j_index_start;
 309
 310         /* Outer loop uses 9 flops */
 311     }
 312
 313     /* Increment number of outer iterations */
 314     outeriter        += nri;
 315
 316     /* Update outer/inner flops */
 317
 318     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*44);
 319 }
 320 /*
 321  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_sse4_1_double
 322  * Electrostatics interaction: ReactionField
 323  * VdW interaction:            LennardJones
 324  * Geometry:                   Particle-Particle
 325  * Calculate force/pot:        Force
 326  */
 327 void
 328 nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_sse4_1_double
 329                     (t_nblist                    * gmx_restrict       nlist,
 330                      rvec                        * gmx_restrict          xx,
 331                      rvec                        * gmx_restrict          ff,
 332                      struct t_forcerec           * gmx_restrict          fr,
 333                      t_mdatoms                   * gmx_restrict     mdatoms,
 334                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 335                      t_nrnb                      * gmx_restrict        nrnb)
 336 {
 337     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 338      * just 0 for non-waters.
 339      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 340      * jnr indices corresponding to data put in the four positions in the SIMD register.
 341      */
 342     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 343     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 344     int              jnrA,jnrB;
 345     int              j_coord_offsetA,j_coord_offsetB;
 346     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 347     real             rcutoff_scalar;
 348     real             *shiftvec,*fshift,*x,*f;
 349     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 350     int              vdwioffset0;
 351     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 352     int              vdwjidx0A,vdwjidx0B;
 353     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 354     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 355     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 356     real             *charge;
 357     int              nvdwtype;
 358     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 359     int              *vdwtype;
 360     real             *vdwparam;
 361     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 362     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 363     __m128d          dummy_mask,cutoff_mask;
 364     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 365     __m128d          one     = _mm_set1_pd(1.0);
 366     __m128d          two     = _mm_set1_pd(2.0);
 367     x                = xx[0];
 368     f                = ff[0];
 369
 370     nri              = nlist->nri;
 371     iinr             = nlist->iinr;
 372     jindex           = nlist->jindex;
 373     jjnr             = nlist->jjnr;
 374     shiftidx         = nlist->shift;
 375     gid              = nlist->gid;
 376     shiftvec         = fr->shift_vec[0];
 377     fshift           = fr->fshift[0];
 378     facel            = _mm_set1_pd(fr->ic->epsfac);
 379     charge           = mdatoms->chargeA;
 380     krf              = _mm_set1_pd(fr->ic->k_rf);
 381     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 382     crf              = _mm_set1_pd(fr->ic->c_rf);
 383     nvdwtype         = fr->ntype;
 384     vdwparam         = fr->nbfp;
 385     vdwtype          = mdatoms->typeA;
 386
 387     /* Avoid stupid compiler warnings */
 388     jnrA = jnrB = 0;
 389     j_coord_offsetA = 0;
 390     j_coord_offsetB = 0;
 391
 392     outeriter        = 0;
 393     inneriter        = 0;
 394
 395     /* Start outer loop over neighborlists */
 396     for(iidx=0; iidx<nri; iidx++)
 397     {
 398         /* Load shift vector for this list */
 399         i_shift_offset   = DIM*shiftidx[iidx];
 400
 401         /* Load limits for loop over neighbors */
 402         j_index_start    = jindex[iidx];
 403         j_index_end      = jindex[iidx+1];
 404
 405         /* Get outer coordinate index */
 406         inr              = iinr[iidx];
 407         i_coord_offset   = DIM*inr;
 408
 409         /* Load i particle coords and add shift vector */
 410         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 411
 412         fix0             = _mm_setzero_pd();
 413         fiy0             = _mm_setzero_pd();
 414         fiz0             = _mm_setzero_pd();
 415
 416         /* Load parameters for i particles */
 417         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 418         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 419
 420         /* Start inner kernel loop */
 421         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 422         {
 423
 424             /* Get j neighbor index, and coordinate index */
 425             jnrA             = jjnr[jidx];
 426             jnrB             = jjnr[jidx+1];
 427             j_coord_offsetA  = DIM*jnrA;
 428             j_coord_offsetB  = DIM*jnrB;
 429
 430             /* load j atom coordinates */
 431             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 432                                               &jx0,&jy0,&jz0);
 433
 434             /* Calculate displacement vector */
 435             dx00             = _mm_sub_pd(ix0,jx0);
 436             dy00             = _mm_sub_pd(iy0,jy0);
 437             dz00             = _mm_sub_pd(iz0,jz0);
 438
 439             /* Calculate squared distance and things based on it */
 440             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 441
 442             rinv00           = sse41_invsqrt_d(rsq00);
 443
 444             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 445
 446             /* Load parameters for j particles */
 447             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 448             vdwjidx0A        = 2*vdwtype[jnrA+0];
 449             vdwjidx0B        = 2*vdwtype[jnrB+0];
 450
 451             /**************************
 452              * CALCULATE INTERACTIONS *
 453              **************************/
 454
 455             /* Compute parameters for interactions between i and j atoms */
 456             qq00             = _mm_mul_pd(iq0,jq0);
 457             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 458                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 459
 460             /* REACTION-FIELD ELECTROSTATICS */
 461             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 462
 463             /* LENNARD-JONES DISPERSION/REPULSION */
 464
 465             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 466             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 467
 468             fscal            = _mm_add_pd(felec,fvdw);
 469
 470             /* Calculate temporary vectorial force */
 471             tx               = _mm_mul_pd(fscal,dx00);
 472             ty               = _mm_mul_pd(fscal,dy00);
 473             tz               = _mm_mul_pd(fscal,dz00);
 474
 475             /* Update vectorial force */
 476             fix0             = _mm_add_pd(fix0,tx);
 477             fiy0             = _mm_add_pd(fiy0,ty);
 478             fiz0             = _mm_add_pd(fiz0,tz);
 479
 480             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 481
 482             /* Inner loop uses 34 flops */
 483         }
 484
 485         if(jidx<j_index_end)
 486         {
 487
 488             jnrA             = jjnr[jidx];
 489             j_coord_offsetA  = DIM*jnrA;
 490
 491             /* load j atom coordinates */
 492             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 493                                               &jx0,&jy0,&jz0);
 494
 495             /* Calculate displacement vector */
 496             dx00             = _mm_sub_pd(ix0,jx0);
 497             dy00             = _mm_sub_pd(iy0,jy0);
 498             dz00             = _mm_sub_pd(iz0,jz0);
 499
 500             /* Calculate squared distance and things based on it */
 501             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 502
 503             rinv00           = sse41_invsqrt_d(rsq00);
 504
 505             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 506
 507             /* Load parameters for j particles */
 508             jq0              = _mm_load_sd(charge+jnrA+0);
 509             vdwjidx0A        = 2*vdwtype[jnrA+0];
 510
 511             /**************************
 512              * CALCULATE INTERACTIONS *
 513              **************************/
 514
 515             /* Compute parameters for interactions between i and j atoms */
 516             qq00             = _mm_mul_pd(iq0,jq0);
 517             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 518
 519             /* REACTION-FIELD ELECTROSTATICS */
 520             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 521
 522             /* LENNARD-JONES DISPERSION/REPULSION */
 523
 524             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 525             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 526
 527             fscal            = _mm_add_pd(felec,fvdw);
 528
 529             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 530
 531             /* Calculate temporary vectorial force */
 532             tx               = _mm_mul_pd(fscal,dx00);
 533             ty               = _mm_mul_pd(fscal,dy00);
 534             tz               = _mm_mul_pd(fscal,dz00);
 535
 536             /* Update vectorial force */
 537             fix0             = _mm_add_pd(fix0,tx);
 538             fiy0             = _mm_add_pd(fiy0,ty);
 539             fiz0             = _mm_add_pd(fiz0,tz);
 540
 541             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 542
 543             /* Inner loop uses 34 flops */
 544         }
 545
 546         /* End of innermost loop */
 547
 548         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 549                                               f+i_coord_offset,fshift+i_shift_offset);
 550
 551         /* Increment number of inner iterations */
 552         inneriter                  += j_index_end - j_index_start;
 553
 554         /* Outer loop uses 7 flops */
 555     }
 556
 557     /* Increment number of outer iterations */
 558     outeriter        += nri;
 559
 560     /* Update outer/inner flops */
 561
 562     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*34);
 563 }