src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_sse2_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 sse2_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_double.h"
  48 #include "kernelutil_x86_sse2_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_VF_sse2_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_VF_sse2_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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 118     rcutoff_scalar   = fr->rcoulomb;
 119     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 120     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 121
 122     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 123     rvdw             = _mm_set1_pd(fr->rvdw);
 124
 125     /* Avoid stupid compiler warnings */
 126     jnrA = jnrB = 0;
 127     j_coord_offsetA = 0;
 128     j_coord_offsetB = 0;
 129
 130     outeriter        = 0;
 131     inneriter        = 0;
 132
 133     /* Start outer loop over neighborlists */
 134     for(iidx=0; iidx<nri; iidx++)
 135     {
 136         /* Load shift vector for this list */
 137         i_shift_offset   = DIM*shiftidx[iidx];
 138
 139         /* Load limits for loop over neighbors */
 140         j_index_start    = jindex[iidx];
 141         j_index_end      = jindex[iidx+1];
 142
 143         /* Get outer coordinate index */
 144         inr              = iinr[iidx];
 145         i_coord_offset   = DIM*inr;
 146
 147         /* Load i particle coords and add shift vector */
 148         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 149
 150         fix0             = _mm_setzero_pd();
 151         fiy0             = _mm_setzero_pd();
 152         fiz0             = _mm_setzero_pd();
 153
 154         /* Load parameters for i particles */
 155         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 156         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 157
 158         /* Reset potential sums */
 159         velecsum         = _mm_setzero_pd();
 160         vvdwsum          = _mm_setzero_pd();
 161
 162         /* Start inner kernel loop */
 163         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 164         {
 165
 166             /* Get j neighbor index, and coordinate index */
 167             jnrA             = jjnr[jidx];
 168             jnrB             = jjnr[jidx+1];
 169             j_coord_offsetA  = DIM*jnrA;
 170             j_coord_offsetB  = DIM*jnrB;
 171
 172             /* load j atom coordinates */
 173             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 174                                               &jx0,&jy0,&jz0);
 175
 176             /* Calculate displacement vector */
 177             dx00             = _mm_sub_pd(ix0,jx0);
 178             dy00             = _mm_sub_pd(iy0,jy0);
 179             dz00             = _mm_sub_pd(iz0,jz0);
 180
 181             /* Calculate squared distance and things based on it */
 182             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 183
 184             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 185
 186             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 187
 188             /* Load parameters for j particles */
 189             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 190             vdwjidx0A        = 2*vdwtype[jnrA+0];
 191             vdwjidx0B        = 2*vdwtype[jnrB+0];
 192
 193             /**************************
 194              * CALCULATE INTERACTIONS *
 195              **************************/
 196
 197             if (gmx_mm_any_lt(rsq00,rcutoff2))
 198             {
 199
 200             /* Compute parameters for interactions between i and j atoms */
 201             qq00             = _mm_mul_pd(iq0,jq0);
 202             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 203                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 204
 205             /* REACTION-FIELD ELECTROSTATICS */
 206             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 207             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 208
 209             /* LENNARD-JONES DISPERSION/REPULSION */
 210
 211             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 212             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 213             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 214             vvdw             = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 215                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 216             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 217
 218             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 219
 220             /* Update potential sum for this i atom from the interaction with this j atom. */
 221             velec            = _mm_and_pd(velec,cutoff_mask);
 222             velecsum         = _mm_add_pd(velecsum,velec);
 223             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 224             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 225
 226             fscal            = _mm_add_pd(felec,fvdw);
 227
 228             fscal            = _mm_and_pd(fscal,cutoff_mask);
 229
 230             /* Calculate temporary vectorial force */
 231             tx               = _mm_mul_pd(fscal,dx00);
 232             ty               = _mm_mul_pd(fscal,dy00);
 233             tz               = _mm_mul_pd(fscal,dz00);
 234
 235             /* Update vectorial force */
 236             fix0             = _mm_add_pd(fix0,tx);
 237             fiy0             = _mm_add_pd(fiy0,ty);
 238             fiz0             = _mm_add_pd(fiz0,tz);
 239
 240             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 241
 242             }
 243
 244             /* Inner loop uses 54 flops */
 245         }
 246
 247         if(jidx<j_index_end)
 248         {
 249
 250             jnrA             = jjnr[jidx];
 251             j_coord_offsetA  = DIM*jnrA;
 252
 253             /* load j atom coordinates */
 254             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 255                                               &jx0,&jy0,&jz0);
 256
 257             /* Calculate displacement vector */
 258             dx00             = _mm_sub_pd(ix0,jx0);
 259             dy00             = _mm_sub_pd(iy0,jy0);
 260             dz00             = _mm_sub_pd(iz0,jz0);
 261
 262             /* Calculate squared distance and things based on it */
 263             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 264
 265             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 266
 267             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 268
 269             /* Load parameters for j particles */
 270             jq0              = _mm_load_sd(charge+jnrA+0);
 271             vdwjidx0A        = 2*vdwtype[jnrA+0];
 272
 273             /**************************
 274              * CALCULATE INTERACTIONS *
 275              **************************/
 276
 277             if (gmx_mm_any_lt(rsq00,rcutoff2))
 278             {
 279
 280             /* Compute parameters for interactions between i and j atoms */
 281             qq00             = _mm_mul_pd(iq0,jq0);
 282             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 283
 284             /* REACTION-FIELD ELECTROSTATICS */
 285             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 286             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 287
 288             /* LENNARD-JONES DISPERSION/REPULSION */
 289
 290             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 291             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 292             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 293             vvdw             = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 294                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 295             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 296
 297             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 298
 299             /* Update potential sum for this i atom from the interaction with this j atom. */
 300             velec            = _mm_and_pd(velec,cutoff_mask);
 301             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 302             velecsum         = _mm_add_pd(velecsum,velec);
 303             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 304             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 305             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 306
 307             fscal            = _mm_add_pd(felec,fvdw);
 308
 309             fscal            = _mm_and_pd(fscal,cutoff_mask);
 310
 311             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 312
 313             /* Calculate temporary vectorial force */
 314             tx               = _mm_mul_pd(fscal,dx00);
 315             ty               = _mm_mul_pd(fscal,dy00);
 316             tz               = _mm_mul_pd(fscal,dz00);
 317
 318             /* Update vectorial force */
 319             fix0             = _mm_add_pd(fix0,tx);
 320             fiy0             = _mm_add_pd(fiy0,ty);
 321             fiz0             = _mm_add_pd(fiz0,tz);
 322
 323             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 324
 325             }
 326
 327             /* Inner loop uses 54 flops */
 328         }
 329
 330         /* End of innermost loop */
 331
 332         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 333                                               f+i_coord_offset,fshift+i_shift_offset);
 334
 335         ggid                        = gid[iidx];
 336         /* Update potential energies */
 337         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 338         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 339
 340         /* Increment number of inner iterations */
 341         inneriter                  += j_index_end - j_index_start;
 342
 343         /* Outer loop uses 9 flops */
 344     }
 345
 346     /* Increment number of outer iterations */
 347     outeriter        += nri;
 348
 349     /* Update outer/inner flops */
 350
 351     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*54);
 352 }
 353 /*
 354  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_sse2_double
 355  * Electrostatics interaction: ReactionField
 356  * VdW interaction:            LennardJones
 357  * Geometry:                   Particle-Particle
 358  * Calculate force/pot:        Force
 359  */
 360 void
 361 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_sse2_double
 362                     (t_nblist                    * gmx_restrict       nlist,
 363                      rvec                        * gmx_restrict          xx,
 364                      rvec                        * gmx_restrict          ff,
 365                      t_forcerec                  * gmx_restrict          fr,
 366                      t_mdatoms                   * gmx_restrict     mdatoms,
 367                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 368                      t_nrnb                      * gmx_restrict        nrnb)
 369 {
 370     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 371      * just 0 for non-waters.
 372      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 373      * jnr indices corresponding to data put in the four positions in the SIMD register.
 374      */
 375     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 376     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 377     int              jnrA,jnrB;
 378     int              j_coord_offsetA,j_coord_offsetB;
 379     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 380     real             rcutoff_scalar;
 381     real             *shiftvec,*fshift,*x,*f;
 382     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 383     int              vdwioffset0;
 384     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 385     int              vdwjidx0A,vdwjidx0B;
 386     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 387     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 388     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 389     real             *charge;
 390     int              nvdwtype;
 391     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 392     int              *vdwtype;
 393     real             *vdwparam;
 394     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 395     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 396     __m128d          dummy_mask,cutoff_mask;
 397     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 398     __m128d          one     = _mm_set1_pd(1.0);
 399     __m128d          two     = _mm_set1_pd(2.0);
 400     x                = xx[0];
 401     f                = ff[0];
 402
 403     nri              = nlist->nri;
 404     iinr             = nlist->iinr;
 405     jindex           = nlist->jindex;
 406     jjnr             = nlist->jjnr;
 407     shiftidx         = nlist->shift;
 408     gid              = nlist->gid;
 409     shiftvec         = fr->shift_vec[0];
 410     fshift           = fr->fshift[0];
 411     facel            = _mm_set1_pd(fr->epsfac);
 412     charge           = mdatoms->chargeA;
 413     krf              = _mm_set1_pd(fr->ic->k_rf);
 414     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 415     crf              = _mm_set1_pd(fr->ic->c_rf);
 416     nvdwtype         = fr->ntype;
 417     vdwparam         = fr->nbfp;
 418     vdwtype          = mdatoms->typeA;
 419
 420     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 421     rcutoff_scalar   = fr->rcoulomb;
 422     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 423     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 424
 425     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 426     rvdw             = _mm_set1_pd(fr->rvdw);
 427
 428     /* Avoid stupid compiler warnings */
 429     jnrA = jnrB = 0;
 430     j_coord_offsetA = 0;
 431     j_coord_offsetB = 0;
 432
 433     outeriter        = 0;
 434     inneriter        = 0;
 435
 436     /* Start outer loop over neighborlists */
 437     for(iidx=0; iidx<nri; iidx++)
 438     {
 439         /* Load shift vector for this list */
 440         i_shift_offset   = DIM*shiftidx[iidx];
 441
 442         /* Load limits for loop over neighbors */
 443         j_index_start    = jindex[iidx];
 444         j_index_end      = jindex[iidx+1];
 445
 446         /* Get outer coordinate index */
 447         inr              = iinr[iidx];
 448         i_coord_offset   = DIM*inr;
 449
 450         /* Load i particle coords and add shift vector */
 451         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 452
 453         fix0             = _mm_setzero_pd();
 454         fiy0             = _mm_setzero_pd();
 455         fiz0             = _mm_setzero_pd();
 456
 457         /* Load parameters for i particles */
 458         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 459         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 460
 461         /* Start inner kernel loop */
 462         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 463         {
 464
 465             /* Get j neighbor index, and coordinate index */
 466             jnrA             = jjnr[jidx];
 467             jnrB             = jjnr[jidx+1];
 468             j_coord_offsetA  = DIM*jnrA;
 469             j_coord_offsetB  = DIM*jnrB;
 470
 471             /* load j atom coordinates */
 472             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 473                                               &jx0,&jy0,&jz0);
 474
 475             /* Calculate displacement vector */
 476             dx00             = _mm_sub_pd(ix0,jx0);
 477             dy00             = _mm_sub_pd(iy0,jy0);
 478             dz00             = _mm_sub_pd(iz0,jz0);
 479
 480             /* Calculate squared distance and things based on it */
 481             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 482
 483             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 484
 485             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 486
 487             /* Load parameters for j particles */
 488             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 489             vdwjidx0A        = 2*vdwtype[jnrA+0];
 490             vdwjidx0B        = 2*vdwtype[jnrB+0];
 491
 492             /**************************
 493              * CALCULATE INTERACTIONS *
 494              **************************/
 495
 496             if (gmx_mm_any_lt(rsq00,rcutoff2))
 497             {
 498
 499             /* Compute parameters for interactions between i and j atoms */
 500             qq00             = _mm_mul_pd(iq0,jq0);
 501             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 502                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 503
 504             /* REACTION-FIELD ELECTROSTATICS */
 505             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 506
 507             /* LENNARD-JONES DISPERSION/REPULSION */
 508
 509             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 510             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 511
 512             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 513
 514             fscal            = _mm_add_pd(felec,fvdw);
 515
 516             fscal            = _mm_and_pd(fscal,cutoff_mask);
 517
 518             /* Calculate temporary vectorial force */
 519             tx               = _mm_mul_pd(fscal,dx00);
 520             ty               = _mm_mul_pd(fscal,dy00);
 521             tz               = _mm_mul_pd(fscal,dz00);
 522
 523             /* Update vectorial force */
 524             fix0             = _mm_add_pd(fix0,tx);
 525             fiy0             = _mm_add_pd(fiy0,ty);
 526             fiz0             = _mm_add_pd(fiz0,tz);
 527
 528             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 529
 530             }
 531
 532             /* Inner loop uses 37 flops */
 533         }
 534
 535         if(jidx<j_index_end)
 536         {
 537
 538             jnrA             = jjnr[jidx];
 539             j_coord_offsetA  = DIM*jnrA;
 540
 541             /* load j atom coordinates */
 542             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 543                                               &jx0,&jy0,&jz0);
 544
 545             /* Calculate displacement vector */
 546             dx00             = _mm_sub_pd(ix0,jx0);
 547             dy00             = _mm_sub_pd(iy0,jy0);
 548             dz00             = _mm_sub_pd(iz0,jz0);
 549
 550             /* Calculate squared distance and things based on it */
 551             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 552
 553             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 554
 555             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 556
 557             /* Load parameters for j particles */
 558             jq0              = _mm_load_sd(charge+jnrA+0);
 559             vdwjidx0A        = 2*vdwtype[jnrA+0];
 560
 561             /**************************
 562              * CALCULATE INTERACTIONS *
 563              **************************/
 564
 565             if (gmx_mm_any_lt(rsq00,rcutoff2))
 566             {
 567
 568             /* Compute parameters for interactions between i and j atoms */
 569             qq00             = _mm_mul_pd(iq0,jq0);
 570             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 571
 572             /* REACTION-FIELD ELECTROSTATICS */
 573             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 574
 575             /* LENNARD-JONES DISPERSION/REPULSION */
 576
 577             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 578             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 579
 580             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 581
 582             fscal            = _mm_add_pd(felec,fvdw);
 583
 584             fscal            = _mm_and_pd(fscal,cutoff_mask);
 585
 586             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 587
 588             /* Calculate temporary vectorial force */
 589             tx               = _mm_mul_pd(fscal,dx00);
 590             ty               = _mm_mul_pd(fscal,dy00);
 591             tz               = _mm_mul_pd(fscal,dz00);
 592
 593             /* Update vectorial force */
 594             fix0             = _mm_add_pd(fix0,tx);
 595             fiy0             = _mm_add_pd(fiy0,ty);
 596             fiz0             = _mm_add_pd(fiz0,tz);
 597
 598             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 599
 600             }
 601
 602             /* Inner loop uses 37 flops */
 603         }
 604
 605         /* End of innermost loop */
 606
 607         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 608                                               f+i_coord_offset,fshift+i_shift_offset);
 609
 610         /* Increment number of inner iterations */
 611         inneriter                  += j_index_end - j_index_start;
 612
 613         /* Outer loop uses 7 flops */
 614     }
 615
 616     /* Increment number of outer iterations */
 617     outeriter        += nri;
 618
 619     /* Update outer/inner flops */
 620
 621     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*37);
 622 }