src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_sse4_1_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 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/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_double.h"
  50 #include "kernelutil_x86_sse4_1_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_VF_sse4_1_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_VF_sse4_1_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B;
  85     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  88     real             *charge;
  89     int              nvdwtype;
  90     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  91     int              *vdwtype;
  92     real             *vdwparam;
  93     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  94     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  95     __m128d          dummy_mask,cutoff_mask;
  96     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  97     __m128d          one     = _mm_set1_pd(1.0);
  98     __m128d          two     = _mm_set1_pd(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     facel            = _mm_set1_pd(fr->epsfac);
 111     charge           = mdatoms->chargeA;
 112     krf              = _mm_set1_pd(fr->ic->k_rf);
 113     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 114     crf              = _mm_set1_pd(fr->ic->c_rf);
 115     nvdwtype         = fr->ntype;
 116     vdwparam         = fr->nbfp;
 117     vdwtype          = mdatoms->typeA;
 118
 119     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 120     rcutoff_scalar   = fr->rcoulomb;
 121     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 122     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 123
 124     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 125     rvdw             = _mm_set1_pd(fr->rvdw);
 126
 127     /* Avoid stupid compiler warnings */
 128     jnrA = jnrB = 0;
 129     j_coord_offsetA = 0;
 130     j_coord_offsetB = 0;
 131
 132     outeriter        = 0;
 133     inneriter        = 0;
 134
 135     /* Start outer loop over neighborlists */
 136     for(iidx=0; iidx<nri; iidx++)
 137     {
 138         /* Load shift vector for this list */
 139         i_shift_offset   = DIM*shiftidx[iidx];
 140
 141         /* Load limits for loop over neighbors */
 142         j_index_start    = jindex[iidx];
 143         j_index_end      = jindex[iidx+1];
 144
 145         /* Get outer coordinate index */
 146         inr              = iinr[iidx];
 147         i_coord_offset   = DIM*inr;
 148
 149         /* Load i particle coords and add shift vector */
 150         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 151
 152         fix0             = _mm_setzero_pd();
 153         fiy0             = _mm_setzero_pd();
 154         fiz0             = _mm_setzero_pd();
 155
 156         /* Load parameters for i particles */
 157         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 158         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 159
 160         /* Reset potential sums */
 161         velecsum         = _mm_setzero_pd();
 162         vvdwsum          = _mm_setzero_pd();
 163
 164         /* Start inner kernel loop */
 165         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 166         {
 167
 168             /* Get j neighbor index, and coordinate index */
 169             jnrA             = jjnr[jidx];
 170             jnrB             = jjnr[jidx+1];
 171             j_coord_offsetA  = DIM*jnrA;
 172             j_coord_offsetB  = DIM*jnrB;
 173
 174             /* load j atom coordinates */
 175             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 176                                               &jx0,&jy0,&jz0);
 177
 178             /* Calculate displacement vector */
 179             dx00             = _mm_sub_pd(ix0,jx0);
 180             dy00             = _mm_sub_pd(iy0,jy0);
 181             dz00             = _mm_sub_pd(iz0,jz0);
 182
 183             /* Calculate squared distance and things based on it */
 184             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 185
 186             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 187
 188             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 189
 190             /* Load parameters for j particles */
 191             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 192             vdwjidx0A        = 2*vdwtype[jnrA+0];
 193             vdwjidx0B        = 2*vdwtype[jnrB+0];
 194
 195             /**************************
 196              * CALCULATE INTERACTIONS *
 197              **************************/
 198
 199             if (gmx_mm_any_lt(rsq00,rcutoff2))
 200             {
 201
 202             /* Compute parameters for interactions between i and j atoms */
 203             qq00             = _mm_mul_pd(iq0,jq0);
 204             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 205                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 206
 207             /* REACTION-FIELD ELECTROSTATICS */
 208             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 209             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 210
 211             /* LENNARD-JONES DISPERSION/REPULSION */
 212
 213             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 214             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 215             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 216             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) ,
 217                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 218             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 219
 220             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 221
 222             /* Update potential sum for this i atom from the interaction with this j atom. */
 223             velec            = _mm_and_pd(velec,cutoff_mask);
 224             velecsum         = _mm_add_pd(velecsum,velec);
 225             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 226             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 227
 228             fscal            = _mm_add_pd(felec,fvdw);
 229
 230             fscal            = _mm_and_pd(fscal,cutoff_mask);
 231
 232             /* Calculate temporary vectorial force */
 233             tx               = _mm_mul_pd(fscal,dx00);
 234             ty               = _mm_mul_pd(fscal,dy00);
 235             tz               = _mm_mul_pd(fscal,dz00);
 236
 237             /* Update vectorial force */
 238             fix0             = _mm_add_pd(fix0,tx);
 239             fiy0             = _mm_add_pd(fiy0,ty);
 240             fiz0             = _mm_add_pd(fiz0,tz);
 241
 242             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 243
 244             }
 245
 246             /* Inner loop uses 54 flops */
 247         }
 248
 249         if(jidx<j_index_end)
 250         {
 251
 252             jnrA             = jjnr[jidx];
 253             j_coord_offsetA  = DIM*jnrA;
 254
 255             /* load j atom coordinates */
 256             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 257                                               &jx0,&jy0,&jz0);
 258
 259             /* Calculate displacement vector */
 260             dx00             = _mm_sub_pd(ix0,jx0);
 261             dy00             = _mm_sub_pd(iy0,jy0);
 262             dz00             = _mm_sub_pd(iz0,jz0);
 263
 264             /* Calculate squared distance and things based on it */
 265             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 266
 267             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 268
 269             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 270
 271             /* Load parameters for j particles */
 272             jq0              = _mm_load_sd(charge+jnrA+0);
 273             vdwjidx0A        = 2*vdwtype[jnrA+0];
 274
 275             /**************************
 276              * CALCULATE INTERACTIONS *
 277              **************************/
 278
 279             if (gmx_mm_any_lt(rsq00,rcutoff2))
 280             {
 281
 282             /* Compute parameters for interactions between i and j atoms */
 283             qq00             = _mm_mul_pd(iq0,jq0);
 284             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 285
 286             /* REACTION-FIELD ELECTROSTATICS */
 287             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 288             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 289
 290             /* LENNARD-JONES DISPERSION/REPULSION */
 291
 292             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 293             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 294             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 295             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) ,
 296                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 297             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 298
 299             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 300
 301             /* Update potential sum for this i atom from the interaction with this j atom. */
 302             velec            = _mm_and_pd(velec,cutoff_mask);
 303             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 304             velecsum         = _mm_add_pd(velecsum,velec);
 305             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 306             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 307             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 308
 309             fscal            = _mm_add_pd(felec,fvdw);
 310
 311             fscal            = _mm_and_pd(fscal,cutoff_mask);
 312
 313             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 314
 315             /* Calculate temporary vectorial force */
 316             tx               = _mm_mul_pd(fscal,dx00);
 317             ty               = _mm_mul_pd(fscal,dy00);
 318             tz               = _mm_mul_pd(fscal,dz00);
 319
 320             /* Update vectorial force */
 321             fix0             = _mm_add_pd(fix0,tx);
 322             fiy0             = _mm_add_pd(fiy0,ty);
 323             fiz0             = _mm_add_pd(fiz0,tz);
 324
 325             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 326
 327             }
 328
 329             /* Inner loop uses 54 flops */
 330         }
 331
 332         /* End of innermost loop */
 333
 334         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 335                                               f+i_coord_offset,fshift+i_shift_offset);
 336
 337         ggid                        = gid[iidx];
 338         /* Update potential energies */
 339         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 340         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 341
 342         /* Increment number of inner iterations */
 343         inneriter                  += j_index_end - j_index_start;
 344
 345         /* Outer loop uses 9 flops */
 346     }
 347
 348     /* Increment number of outer iterations */
 349     outeriter        += nri;
 350
 351     /* Update outer/inner flops */
 352
 353     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*54);
 354 }
 355 /*
 356  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_sse4_1_double
 357  * Electrostatics interaction: ReactionField
 358  * VdW interaction:            LennardJones
 359  * Geometry:                   Particle-Particle
 360  * Calculate force/pot:        Force
 361  */
 362 void
 363 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_sse4_1_double
 364                     (t_nblist                    * gmx_restrict       nlist,
 365                      rvec                        * gmx_restrict          xx,
 366                      rvec                        * gmx_restrict          ff,
 367                      t_forcerec                  * gmx_restrict          fr,
 368                      t_mdatoms                   * gmx_restrict     mdatoms,
 369                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 370                      t_nrnb                      * gmx_restrict        nrnb)
 371 {
 372     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 373      * just 0 for non-waters.
 374      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 375      * jnr indices corresponding to data put in the four positions in the SIMD register.
 376      */
 377     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 378     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 379     int              jnrA,jnrB;
 380     int              j_coord_offsetA,j_coord_offsetB;
 381     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 382     real             rcutoff_scalar;
 383     real             *shiftvec,*fshift,*x,*f;
 384     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 385     int              vdwioffset0;
 386     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 387     int              vdwjidx0A,vdwjidx0B;
 388     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 389     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 390     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 391     real             *charge;
 392     int              nvdwtype;
 393     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 394     int              *vdwtype;
 395     real             *vdwparam;
 396     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 397     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 398     __m128d          dummy_mask,cutoff_mask;
 399     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 400     __m128d          one     = _mm_set1_pd(1.0);
 401     __m128d          two     = _mm_set1_pd(2.0);
 402     x                = xx[0];
 403     f                = ff[0];
 404
 405     nri              = nlist->nri;
 406     iinr             = nlist->iinr;
 407     jindex           = nlist->jindex;
 408     jjnr             = nlist->jjnr;
 409     shiftidx         = nlist->shift;
 410     gid              = nlist->gid;
 411     shiftvec         = fr->shift_vec[0];
 412     fshift           = fr->fshift[0];
 413     facel            = _mm_set1_pd(fr->epsfac);
 414     charge           = mdatoms->chargeA;
 415     krf              = _mm_set1_pd(fr->ic->k_rf);
 416     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 417     crf              = _mm_set1_pd(fr->ic->c_rf);
 418     nvdwtype         = fr->ntype;
 419     vdwparam         = fr->nbfp;
 420     vdwtype          = mdatoms->typeA;
 421
 422     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 423     rcutoff_scalar   = fr->rcoulomb;
 424     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 425     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 426
 427     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 428     rvdw             = _mm_set1_pd(fr->rvdw);
 429
 430     /* Avoid stupid compiler warnings */
 431     jnrA = jnrB = 0;
 432     j_coord_offsetA = 0;
 433     j_coord_offsetB = 0;
 434
 435     outeriter        = 0;
 436     inneriter        = 0;
 437
 438     /* Start outer loop over neighborlists */
 439     for(iidx=0; iidx<nri; iidx++)
 440     {
 441         /* Load shift vector for this list */
 442         i_shift_offset   = DIM*shiftidx[iidx];
 443
 444         /* Load limits for loop over neighbors */
 445         j_index_start    = jindex[iidx];
 446         j_index_end      = jindex[iidx+1];
 447
 448         /* Get outer coordinate index */
 449         inr              = iinr[iidx];
 450         i_coord_offset   = DIM*inr;
 451
 452         /* Load i particle coords and add shift vector */
 453         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 454
 455         fix0             = _mm_setzero_pd();
 456         fiy0             = _mm_setzero_pd();
 457         fiz0             = _mm_setzero_pd();
 458
 459         /* Load parameters for i particles */
 460         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 461         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 462
 463         /* Start inner kernel loop */
 464         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 465         {
 466
 467             /* Get j neighbor index, and coordinate index */
 468             jnrA             = jjnr[jidx];
 469             jnrB             = jjnr[jidx+1];
 470             j_coord_offsetA  = DIM*jnrA;
 471             j_coord_offsetB  = DIM*jnrB;
 472
 473             /* load j atom coordinates */
 474             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 475                                               &jx0,&jy0,&jz0);
 476
 477             /* Calculate displacement vector */
 478             dx00             = _mm_sub_pd(ix0,jx0);
 479             dy00             = _mm_sub_pd(iy0,jy0);
 480             dz00             = _mm_sub_pd(iz0,jz0);
 481
 482             /* Calculate squared distance and things based on it */
 483             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 484
 485             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 486
 487             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 488
 489             /* Load parameters for j particles */
 490             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 491             vdwjidx0A        = 2*vdwtype[jnrA+0];
 492             vdwjidx0B        = 2*vdwtype[jnrB+0];
 493
 494             /**************************
 495              * CALCULATE INTERACTIONS *
 496              **************************/
 497
 498             if (gmx_mm_any_lt(rsq00,rcutoff2))
 499             {
 500
 501             /* Compute parameters for interactions between i and j atoms */
 502             qq00             = _mm_mul_pd(iq0,jq0);
 503             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 504                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 505
 506             /* REACTION-FIELD ELECTROSTATICS */
 507             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 508
 509             /* LENNARD-JONES DISPERSION/REPULSION */
 510
 511             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 512             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 513
 514             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 515
 516             fscal            = _mm_add_pd(felec,fvdw);
 517
 518             fscal            = _mm_and_pd(fscal,cutoff_mask);
 519
 520             /* Calculate temporary vectorial force */
 521             tx               = _mm_mul_pd(fscal,dx00);
 522             ty               = _mm_mul_pd(fscal,dy00);
 523             tz               = _mm_mul_pd(fscal,dz00);
 524
 525             /* Update vectorial force */
 526             fix0             = _mm_add_pd(fix0,tx);
 527             fiy0             = _mm_add_pd(fiy0,ty);
 528             fiz0             = _mm_add_pd(fiz0,tz);
 529
 530             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 531
 532             }
 533
 534             /* Inner loop uses 37 flops */
 535         }
 536
 537         if(jidx<j_index_end)
 538         {
 539
 540             jnrA             = jjnr[jidx];
 541             j_coord_offsetA  = DIM*jnrA;
 542
 543             /* load j atom coordinates */
 544             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 545                                               &jx0,&jy0,&jz0);
 546
 547             /* Calculate displacement vector */
 548             dx00             = _mm_sub_pd(ix0,jx0);
 549             dy00             = _mm_sub_pd(iy0,jy0);
 550             dz00             = _mm_sub_pd(iz0,jz0);
 551
 552             /* Calculate squared distance and things based on it */
 553             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 554
 555             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 556
 557             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 558
 559             /* Load parameters for j particles */
 560             jq0              = _mm_load_sd(charge+jnrA+0);
 561             vdwjidx0A        = 2*vdwtype[jnrA+0];
 562
 563             /**************************
 564              * CALCULATE INTERACTIONS *
 565              **************************/
 566
 567             if (gmx_mm_any_lt(rsq00,rcutoff2))
 568             {
 569
 570             /* Compute parameters for interactions between i and j atoms */
 571             qq00             = _mm_mul_pd(iq0,jq0);
 572             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 573
 574             /* REACTION-FIELD ELECTROSTATICS */
 575             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 576
 577             /* LENNARD-JONES DISPERSION/REPULSION */
 578
 579             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 580             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 581
 582             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 583
 584             fscal            = _mm_add_pd(felec,fvdw);
 585
 586             fscal            = _mm_and_pd(fscal,cutoff_mask);
 587
 588             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 589
 590             /* Calculate temporary vectorial force */
 591             tx               = _mm_mul_pd(fscal,dx00);
 592             ty               = _mm_mul_pd(fscal,dy00);
 593             tz               = _mm_mul_pd(fscal,dz00);
 594
 595             /* Update vectorial force */
 596             fix0             = _mm_add_pd(fix0,tx);
 597             fiy0             = _mm_add_pd(fiy0,ty);
 598             fiz0             = _mm_add_pd(fiz0,tz);
 599
 600             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 601
 602             }
 603
 604             /* Inner loop uses 37 flops */
 605         }
 606
 607         /* End of innermost loop */
 608
 609         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 610                                               f+i_coord_offset,fshift+i_shift_offset);
 611
 612         /* Increment number of inner iterations */
 613         inneriter                  += j_index_end - j_index_start;
 614
 615         /* Outer loop uses 7 flops */
 616     }
 617
 618     /* Increment number of outer iterations */
 619     outeriter        += nri;
 620
 621     /* Update outer/inner flops */
 622
 623     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*37);
 624 }