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) 2012,2013,2014,2015,2017, 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 "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_sse2_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_VF_sse2_double
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_VF_sse2_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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 117     rcutoff_scalar   = fr->ic->rcoulomb;
 118     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 119     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 120
 121     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 122     rvdw             = _mm_set1_pd(fr->ic->rvdw);
 123
 124     /* Avoid stupid compiler warnings */
 125     jnrA = jnrB = 0;
 126     j_coord_offsetA = 0;
 127     j_coord_offsetB = 0;
 128
 129     outeriter        = 0;
 130     inneriter        = 0;
 131
 132     /* Start outer loop over neighborlists */
 133     for(iidx=0; iidx<nri; iidx++)
 134     {
 135         /* Load shift vector for this list */
 136         i_shift_offset   = DIM*shiftidx[iidx];
 137
 138         /* Load limits for loop over neighbors */
 139         j_index_start    = jindex[iidx];
 140         j_index_end      = jindex[iidx+1];
 141
 142         /* Get outer coordinate index */
 143         inr              = iinr[iidx];
 144         i_coord_offset   = DIM*inr;
 145
 146         /* Load i particle coords and add shift vector */
 147         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 148
 149         fix0             = _mm_setzero_pd();
 150         fiy0             = _mm_setzero_pd();
 151         fiz0             = _mm_setzero_pd();
 152
 153         /* Load parameters for i particles */
 154         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 155         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 156
 157         /* Reset potential sums */
 158         velecsum         = _mm_setzero_pd();
 159         vvdwsum          = _mm_setzero_pd();
 160
 161         /* Start inner kernel loop */
 162         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 163         {
 164
 165             /* Get j neighbor index, and coordinate index */
 166             jnrA             = jjnr[jidx];
 167             jnrB             = jjnr[jidx+1];
 168             j_coord_offsetA  = DIM*jnrA;
 169             j_coord_offsetB  = DIM*jnrB;
 170
 171             /* load j atom coordinates */
 172             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 173                                               &jx0,&jy0,&jz0);
 174
 175             /* Calculate displacement vector */
 176             dx00             = _mm_sub_pd(ix0,jx0);
 177             dy00             = _mm_sub_pd(iy0,jy0);
 178             dz00             = _mm_sub_pd(iz0,jz0);
 179
 180             /* Calculate squared distance and things based on it */
 181             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 182
 183             rinv00           = sse2_invsqrt_d(rsq00);
 184
 185             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 186
 187             /* Load parameters for j particles */
 188             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 189             vdwjidx0A        = 2*vdwtype[jnrA+0];
 190             vdwjidx0B        = 2*vdwtype[jnrB+0];
 191
 192             /**************************
 193              * CALCULATE INTERACTIONS *
 194              **************************/
 195
 196             if (gmx_mm_any_lt(rsq00,rcutoff2))
 197             {
 198
 199             /* Compute parameters for interactions between i and j atoms */
 200             qq00             = _mm_mul_pd(iq0,jq0);
 201             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 202                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 203
 204             /* REACTION-FIELD ELECTROSTATICS */
 205             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 206             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 207
 208             /* LENNARD-JONES DISPERSION/REPULSION */
 209
 210             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 211             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 212             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 213             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) ,
 214                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 215             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 216
 217             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 218
 219             /* Update potential sum for this i atom from the interaction with this j atom. */
 220             velec            = _mm_and_pd(velec,cutoff_mask);
 221             velecsum         = _mm_add_pd(velecsum,velec);
 222             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 223             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 224
 225             fscal            = _mm_add_pd(felec,fvdw);
 226
 227             fscal            = _mm_and_pd(fscal,cutoff_mask);
 228
 229             /* Calculate temporary vectorial force */
 230             tx               = _mm_mul_pd(fscal,dx00);
 231             ty               = _mm_mul_pd(fscal,dy00);
 232             tz               = _mm_mul_pd(fscal,dz00);
 233
 234             /* Update vectorial force */
 235             fix0             = _mm_add_pd(fix0,tx);
 236             fiy0             = _mm_add_pd(fiy0,ty);
 237             fiz0             = _mm_add_pd(fiz0,tz);
 238
 239             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 240
 241             }
 242
 243             /* Inner loop uses 54 flops */
 244         }
 245
 246         if(jidx<j_index_end)
 247         {
 248
 249             jnrA             = jjnr[jidx];
 250             j_coord_offsetA  = DIM*jnrA;
 251
 252             /* load j atom coordinates */
 253             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 254                                               &jx0,&jy0,&jz0);
 255
 256             /* Calculate displacement vector */
 257             dx00             = _mm_sub_pd(ix0,jx0);
 258             dy00             = _mm_sub_pd(iy0,jy0);
 259             dz00             = _mm_sub_pd(iz0,jz0);
 260
 261             /* Calculate squared distance and things based on it */
 262             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 263
 264             rinv00           = sse2_invsqrt_d(rsq00);
 265
 266             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 267
 268             /* Load parameters for j particles */
 269             jq0              = _mm_load_sd(charge+jnrA+0);
 270             vdwjidx0A        = 2*vdwtype[jnrA+0];
 271
 272             /**************************
 273              * CALCULATE INTERACTIONS *
 274              **************************/
 275
 276             if (gmx_mm_any_lt(rsq00,rcutoff2))
 277             {
 278
 279             /* Compute parameters for interactions between i and j atoms */
 280             qq00             = _mm_mul_pd(iq0,jq0);
 281             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 282
 283             /* REACTION-FIELD ELECTROSTATICS */
 284             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 285             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 286
 287             /* LENNARD-JONES DISPERSION/REPULSION */
 288
 289             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 290             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 291             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 292             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) ,
 293                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 294             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 295
 296             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 297
 298             /* Update potential sum for this i atom from the interaction with this j atom. */
 299             velec            = _mm_and_pd(velec,cutoff_mask);
 300             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 301             velecsum         = _mm_add_pd(velecsum,velec);
 302             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 303             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 304             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 305
 306             fscal            = _mm_add_pd(felec,fvdw);
 307
 308             fscal            = _mm_and_pd(fscal,cutoff_mask);
 309
 310             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 311
 312             /* Calculate temporary vectorial force */
 313             tx               = _mm_mul_pd(fscal,dx00);
 314             ty               = _mm_mul_pd(fscal,dy00);
 315             tz               = _mm_mul_pd(fscal,dz00);
 316
 317             /* Update vectorial force */
 318             fix0             = _mm_add_pd(fix0,tx);
 319             fiy0             = _mm_add_pd(fiy0,ty);
 320             fiz0             = _mm_add_pd(fiz0,tz);
 321
 322             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 323
 324             }
 325
 326             /* Inner loop uses 54 flops */
 327         }
 328
 329         /* End of innermost loop */
 330
 331         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 332                                               f+i_coord_offset,fshift+i_shift_offset);
 333
 334         ggid                        = gid[iidx];
 335         /* Update potential energies */
 336         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 337         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 338
 339         /* Increment number of inner iterations */
 340         inneriter                  += j_index_end - j_index_start;
 341
 342         /* Outer loop uses 9 flops */
 343     }
 344
 345     /* Increment number of outer iterations */
 346     outeriter        += nri;
 347
 348     /* Update outer/inner flops */
 349
 350     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*54);
 351 }
 352 /*
 353  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_sse2_double
 354  * Electrostatics interaction: ReactionField
 355  * VdW interaction:            LennardJones
 356  * Geometry:                   Particle-Particle
 357  * Calculate force/pot:        Force
 358  */
 359 void
 360 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_sse2_double
 361                     (t_nblist                    * gmx_restrict       nlist,
 362                      rvec                        * gmx_restrict          xx,
 363                      rvec                        * gmx_restrict          ff,
 364                      struct t_forcerec           * gmx_restrict          fr,
 365                      t_mdatoms                   * gmx_restrict     mdatoms,
 366                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 367                      t_nrnb                      * gmx_restrict        nrnb)
 368 {
 369     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 370      * just 0 for non-waters.
 371      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 372      * jnr indices corresponding to data put in the four positions in the SIMD register.
 373      */
 374     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 375     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 376     int              jnrA,jnrB;
 377     int              j_coord_offsetA,j_coord_offsetB;
 378     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 379     real             rcutoff_scalar;
 380     real             *shiftvec,*fshift,*x,*f;
 381     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 382     int              vdwioffset0;
 383     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 384     int              vdwjidx0A,vdwjidx0B;
 385     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 386     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 387     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 388     real             *charge;
 389     int              nvdwtype;
 390     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 391     int              *vdwtype;
 392     real             *vdwparam;
 393     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 394     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 395     __m128d          dummy_mask,cutoff_mask;
 396     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 397     __m128d          one     = _mm_set1_pd(1.0);
 398     __m128d          two     = _mm_set1_pd(2.0);
 399     x                = xx[0];
 400     f                = ff[0];
 401
 402     nri              = nlist->nri;
 403     iinr             = nlist->iinr;
 404     jindex           = nlist->jindex;
 405     jjnr             = nlist->jjnr;
 406     shiftidx         = nlist->shift;
 407     gid              = nlist->gid;
 408     shiftvec         = fr->shift_vec[0];
 409     fshift           = fr->fshift[0];
 410     facel            = _mm_set1_pd(fr->ic->epsfac);
 411     charge           = mdatoms->chargeA;
 412     krf              = _mm_set1_pd(fr->ic->k_rf);
 413     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 414     crf              = _mm_set1_pd(fr->ic->c_rf);
 415     nvdwtype         = fr->ntype;
 416     vdwparam         = fr->nbfp;
 417     vdwtype          = mdatoms->typeA;
 418
 419     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 420     rcutoff_scalar   = fr->ic->rcoulomb;
 421     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 422     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 423
 424     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 425     rvdw             = _mm_set1_pd(fr->ic->rvdw);
 426
 427     /* Avoid stupid compiler warnings */
 428     jnrA = jnrB = 0;
 429     j_coord_offsetA = 0;
 430     j_coord_offsetB = 0;
 431
 432     outeriter        = 0;
 433     inneriter        = 0;
 434
 435     /* Start outer loop over neighborlists */
 436     for(iidx=0; iidx<nri; iidx++)
 437     {
 438         /* Load shift vector for this list */
 439         i_shift_offset   = DIM*shiftidx[iidx];
 440
 441         /* Load limits for loop over neighbors */
 442         j_index_start    = jindex[iidx];
 443         j_index_end      = jindex[iidx+1];
 444
 445         /* Get outer coordinate index */
 446         inr              = iinr[iidx];
 447         i_coord_offset   = DIM*inr;
 448
 449         /* Load i particle coords and add shift vector */
 450         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 451
 452         fix0             = _mm_setzero_pd();
 453         fiy0             = _mm_setzero_pd();
 454         fiz0             = _mm_setzero_pd();
 455
 456         /* Load parameters for i particles */
 457         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 458         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 459
 460         /* Start inner kernel loop */
 461         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 462         {
 463
 464             /* Get j neighbor index, and coordinate index */
 465             jnrA             = jjnr[jidx];
 466             jnrB             = jjnr[jidx+1];
 467             j_coord_offsetA  = DIM*jnrA;
 468             j_coord_offsetB  = DIM*jnrB;
 469
 470             /* load j atom coordinates */
 471             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 472                                               &jx0,&jy0,&jz0);
 473
 474             /* Calculate displacement vector */
 475             dx00             = _mm_sub_pd(ix0,jx0);
 476             dy00             = _mm_sub_pd(iy0,jy0);
 477             dz00             = _mm_sub_pd(iz0,jz0);
 478
 479             /* Calculate squared distance and things based on it */
 480             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 481
 482             rinv00           = sse2_invsqrt_d(rsq00);
 483
 484             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 485
 486             /* Load parameters for j particles */
 487             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 488             vdwjidx0A        = 2*vdwtype[jnrA+0];
 489             vdwjidx0B        = 2*vdwtype[jnrB+0];
 490
 491             /**************************
 492              * CALCULATE INTERACTIONS *
 493              **************************/
 494
 495             if (gmx_mm_any_lt(rsq00,rcutoff2))
 496             {
 497
 498             /* Compute parameters for interactions between i and j atoms */
 499             qq00             = _mm_mul_pd(iq0,jq0);
 500             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 501                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 502
 503             /* REACTION-FIELD ELECTROSTATICS */
 504             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 505
 506             /* LENNARD-JONES DISPERSION/REPULSION */
 507
 508             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 509             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 510
 511             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 512
 513             fscal            = _mm_add_pd(felec,fvdw);
 514
 515             fscal            = _mm_and_pd(fscal,cutoff_mask);
 516
 517             /* Calculate temporary vectorial force */
 518             tx               = _mm_mul_pd(fscal,dx00);
 519             ty               = _mm_mul_pd(fscal,dy00);
 520             tz               = _mm_mul_pd(fscal,dz00);
 521
 522             /* Update vectorial force */
 523             fix0             = _mm_add_pd(fix0,tx);
 524             fiy0             = _mm_add_pd(fiy0,ty);
 525             fiz0             = _mm_add_pd(fiz0,tz);
 526
 527             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 528
 529             }
 530
 531             /* Inner loop uses 37 flops */
 532         }
 533
 534         if(jidx<j_index_end)
 535         {
 536
 537             jnrA             = jjnr[jidx];
 538             j_coord_offsetA  = DIM*jnrA;
 539
 540             /* load j atom coordinates */
 541             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 542                                               &jx0,&jy0,&jz0);
 543
 544             /* Calculate displacement vector */
 545             dx00             = _mm_sub_pd(ix0,jx0);
 546             dy00             = _mm_sub_pd(iy0,jy0);
 547             dz00             = _mm_sub_pd(iz0,jz0);
 548
 549             /* Calculate squared distance and things based on it */
 550             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 551
 552             rinv00           = sse2_invsqrt_d(rsq00);
 553
 554             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 555
 556             /* Load parameters for j particles */
 557             jq0              = _mm_load_sd(charge+jnrA+0);
 558             vdwjidx0A        = 2*vdwtype[jnrA+0];
 559
 560             /**************************
 561              * CALCULATE INTERACTIONS *
 562              **************************/
 563
 564             if (gmx_mm_any_lt(rsq00,rcutoff2))
 565             {
 566
 567             /* Compute parameters for interactions between i and j atoms */
 568             qq00             = _mm_mul_pd(iq0,jq0);
 569             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 570
 571             /* REACTION-FIELD ELECTROSTATICS */
 572             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 573
 574             /* LENNARD-JONES DISPERSION/REPULSION */
 575
 576             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 577             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 578
 579             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 580
 581             fscal            = _mm_add_pd(felec,fvdw);
 582
 583             fscal            = _mm_and_pd(fscal,cutoff_mask);
 584
 585             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 586
 587             /* Calculate temporary vectorial force */
 588             tx               = _mm_mul_pd(fscal,dx00);
 589             ty               = _mm_mul_pd(fscal,dy00);
 590             tz               = _mm_mul_pd(fscal,dz00);
 591
 592             /* Update vectorial force */
 593             fix0             = _mm_add_pd(fix0,tx);
 594             fiy0             = _mm_add_pd(fiy0,ty);
 595             fiz0             = _mm_add_pd(fiz0,tz);
 596
 597             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 598
 599             }
 600
 601             /* Inner loop uses 37 flops */
 602         }
 603
 604         /* End of innermost loop */
 605
 606         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 607                                               f+i_coord_offset,fshift+i_shift_offset);
 608
 609         /* Increment number of inner iterations */
 610         inneriter                  += j_index_end - j_index_start;
 611
 612         /* Outer loop uses 7 flops */
 613     }
 614
 615     /* Increment number of outer iterations */
 616     outeriter        += nri;
 617
 618     /* Update outer/inner flops */
 619
 620     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*37);
 621 }