src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
  37  */
  38 #include "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_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_VF_avx_128_fma_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_avx_128_fma_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_macc_pd(krf,rsq00,rinv00),crf));
 209             felec            = _mm_mul_pd(qq00,_mm_msub_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_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 217                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),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             /* Update vectorial force */
 233             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 234             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 235             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 236
 237             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 238                                                    _mm_mul_pd(dx00,fscal),
 239                                                    _mm_mul_pd(dy00,fscal),
 240                                                    _mm_mul_pd(dz00,fscal));
 241
 242             }
 243
 244             /* Inner loop uses 57 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_macc_pd(krf,rsq00,rinv00),crf));
 286             felec            = _mm_mul_pd(qq00,_mm_msub_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_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 294                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),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             /* Update vectorial force */
 314             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 315             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 316             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 317
 318             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 319                                                    _mm_mul_pd(dx00,fscal),
 320                                                    _mm_mul_pd(dy00,fscal),
 321                                                    _mm_mul_pd(dz00,fscal));
 322
 323             }
 324
 325             /* Inner loop uses 57 flops */
 326         }
 327
 328         /* End of innermost loop */
 329
 330         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 331                                               f+i_coord_offset,fshift+i_shift_offset);
 332
 333         ggid                        = gid[iidx];
 334         /* Update potential energies */
 335         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 336         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 337
 338         /* Increment number of inner iterations */
 339         inneriter                  += j_index_end - j_index_start;
 340
 341         /* Outer loop uses 9 flops */
 342     }
 343
 344     /* Increment number of outer iterations */
 345     outeriter        += nri;
 346
 347     /* Update outer/inner flops */
 348
 349     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*57);
 350 }
 351 /*
 352  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_avx_128_fma_double
 353  * Electrostatics interaction: ReactionField
 354  * VdW interaction:            LennardJones
 355  * Geometry:                   Particle-Particle
 356  * Calculate force/pot:        Force
 357  */
 358 void
 359 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_avx_128_fma_double
 360                     (t_nblist                    * gmx_restrict       nlist,
 361                      rvec                        * gmx_restrict          xx,
 362                      rvec                        * gmx_restrict          ff,
 363                      t_forcerec                  * gmx_restrict          fr,
 364                      t_mdatoms                   * gmx_restrict     mdatoms,
 365                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 366                      t_nrnb                      * gmx_restrict        nrnb)
 367 {
 368     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 369      * just 0 for non-waters.
 370      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 371      * jnr indices corresponding to data put in the four positions in the SIMD register.
 372      */
 373     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 374     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 375     int              jnrA,jnrB;
 376     int              j_coord_offsetA,j_coord_offsetB;
 377     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 378     real             rcutoff_scalar;
 379     real             *shiftvec,*fshift,*x,*f;
 380     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 381     int              vdwioffset0;
 382     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 383     int              vdwjidx0A,vdwjidx0B;
 384     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 385     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 386     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 387     real             *charge;
 388     int              nvdwtype;
 389     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 390     int              *vdwtype;
 391     real             *vdwparam;
 392     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 393     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 394     __m128d          dummy_mask,cutoff_mask;
 395     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 396     __m128d          one     = _mm_set1_pd(1.0);
 397     __m128d          two     = _mm_set1_pd(2.0);
 398     x                = xx[0];
 399     f                = ff[0];
 400
 401     nri              = nlist->nri;
 402     iinr             = nlist->iinr;
 403     jindex           = nlist->jindex;
 404     jjnr             = nlist->jjnr;
 405     shiftidx         = nlist->shift;
 406     gid              = nlist->gid;
 407     shiftvec         = fr->shift_vec[0];
 408     fshift           = fr->fshift[0];
 409     facel            = _mm_set1_pd(fr->epsfac);
 410     charge           = mdatoms->chargeA;
 411     krf              = _mm_set1_pd(fr->ic->k_rf);
 412     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 413     crf              = _mm_set1_pd(fr->ic->c_rf);
 414     nvdwtype         = fr->ntype;
 415     vdwparam         = fr->nbfp;
 416     vdwtype          = mdatoms->typeA;
 417
 418     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 419     rcutoff_scalar   = fr->rcoulomb;
 420     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 421     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 422
 423     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 424     rvdw             = _mm_set1_pd(fr->rvdw);
 425
 426     /* Avoid stupid compiler warnings */
 427     jnrA = jnrB = 0;
 428     j_coord_offsetA = 0;
 429     j_coord_offsetB = 0;
 430
 431     outeriter        = 0;
 432     inneriter        = 0;
 433
 434     /* Start outer loop over neighborlists */
 435     for(iidx=0; iidx<nri; iidx++)
 436     {
 437         /* Load shift vector for this list */
 438         i_shift_offset   = DIM*shiftidx[iidx];
 439
 440         /* Load limits for loop over neighbors */
 441         j_index_start    = jindex[iidx];
 442         j_index_end      = jindex[iidx+1];
 443
 444         /* Get outer coordinate index */
 445         inr              = iinr[iidx];
 446         i_coord_offset   = DIM*inr;
 447
 448         /* Load i particle coords and add shift vector */
 449         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 450
 451         fix0             = _mm_setzero_pd();
 452         fiy0             = _mm_setzero_pd();
 453         fiz0             = _mm_setzero_pd();
 454
 455         /* Load parameters for i particles */
 456         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 457         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 458
 459         /* Start inner kernel loop */
 460         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 461         {
 462
 463             /* Get j neighbor index, and coordinate index */
 464             jnrA             = jjnr[jidx];
 465             jnrB             = jjnr[jidx+1];
 466             j_coord_offsetA  = DIM*jnrA;
 467             j_coord_offsetB  = DIM*jnrB;
 468
 469             /* load j atom coordinates */
 470             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 471                                               &jx0,&jy0,&jz0);
 472
 473             /* Calculate displacement vector */
 474             dx00             = _mm_sub_pd(ix0,jx0);
 475             dy00             = _mm_sub_pd(iy0,jy0);
 476             dz00             = _mm_sub_pd(iz0,jz0);
 477
 478             /* Calculate squared distance and things based on it */
 479             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 480
 481             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 482
 483             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 484
 485             /* Load parameters for j particles */
 486             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 487             vdwjidx0A        = 2*vdwtype[jnrA+0];
 488             vdwjidx0B        = 2*vdwtype[jnrB+0];
 489
 490             /**************************
 491              * CALCULATE INTERACTIONS *
 492              **************************/
 493
 494             if (gmx_mm_any_lt(rsq00,rcutoff2))
 495             {
 496
 497             /* Compute parameters for interactions between i and j atoms */
 498             qq00             = _mm_mul_pd(iq0,jq0);
 499             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 500                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 501
 502             /* REACTION-FIELD ELECTROSTATICS */
 503             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 504
 505             /* LENNARD-JONES DISPERSION/REPULSION */
 506
 507             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 508             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 509
 510             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 511
 512             fscal            = _mm_add_pd(felec,fvdw);
 513
 514             fscal            = _mm_and_pd(fscal,cutoff_mask);
 515
 516             /* Update vectorial force */
 517             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 518             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 519             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 520
 521             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 522                                                    _mm_mul_pd(dx00,fscal),
 523                                                    _mm_mul_pd(dy00,fscal),
 524                                                    _mm_mul_pd(dz00,fscal));
 525
 526             }
 527
 528             /* Inner loop uses 40 flops */
 529         }
 530
 531         if(jidx<j_index_end)
 532         {
 533
 534             jnrA             = jjnr[jidx];
 535             j_coord_offsetA  = DIM*jnrA;
 536
 537             /* load j atom coordinates */
 538             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 539                                               &jx0,&jy0,&jz0);
 540
 541             /* Calculate displacement vector */
 542             dx00             = _mm_sub_pd(ix0,jx0);
 543             dy00             = _mm_sub_pd(iy0,jy0);
 544             dz00             = _mm_sub_pd(iz0,jz0);
 545
 546             /* Calculate squared distance and things based on it */
 547             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 548
 549             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 550
 551             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 552
 553             /* Load parameters for j particles */
 554             jq0              = _mm_load_sd(charge+jnrA+0);
 555             vdwjidx0A        = 2*vdwtype[jnrA+0];
 556
 557             /**************************
 558              * CALCULATE INTERACTIONS *
 559              **************************/
 560
 561             if (gmx_mm_any_lt(rsq00,rcutoff2))
 562             {
 563
 564             /* Compute parameters for interactions between i and j atoms */
 565             qq00             = _mm_mul_pd(iq0,jq0);
 566             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 567
 568             /* REACTION-FIELD ELECTROSTATICS */
 569             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 570
 571             /* LENNARD-JONES DISPERSION/REPULSION */
 572
 573             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 574             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 575
 576             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 577
 578             fscal            = _mm_add_pd(felec,fvdw);
 579
 580             fscal            = _mm_and_pd(fscal,cutoff_mask);
 581
 582             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 583
 584             /* Update vectorial force */
 585             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 586             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 587             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 588
 589             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 590                                                    _mm_mul_pd(dx00,fscal),
 591                                                    _mm_mul_pd(dy00,fscal),
 592                                                    _mm_mul_pd(dz00,fscal));
 593
 594             }
 595
 596             /* Inner loop uses 40 flops */
 597         }
 598
 599         /* End of innermost loop */
 600
 601         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 602                                               f+i_coord_offset,fshift+i_shift_offset);
 603
 604         /* Increment number of inner iterations */
 605         inneriter                  += j_index_end - j_index_start;
 606
 607         /* Outer loop uses 7 flops */
 608     }
 609
 610     /* Increment number of outer iterations */
 611     outeriter        += nri;
 612
 613     /* Update outer/inner flops */
 614
 615     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*40);
 616 }