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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  48 #include "kernelutil_x86_avx_128_fma_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_VF_avx_128_fma_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_avx_128_fma_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_macc_pd(krf,rsq00,rinv00),crf));
 207             felec            = _mm_mul_pd(qq00,_mm_msub_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_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 215                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),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             /* Update vectorial force */
 231             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 232             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 233             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 234
 235             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 236                                                    _mm_mul_pd(dx00,fscal),
 237                                                    _mm_mul_pd(dy00,fscal),
 238                                                    _mm_mul_pd(dz00,fscal));
 239
 240             }
 241
 242             /* Inner loop uses 57 flops */
 243         }
 244
 245         if(jidx<j_index_end)
 246         {
 247
 248             jnrA             = jjnr[jidx];
 249             j_coord_offsetA  = DIM*jnrA;
 250
 251             /* load j atom coordinates */
 252             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 253                                               &jx0,&jy0,&jz0);
 254
 255             /* Calculate displacement vector */
 256             dx00             = _mm_sub_pd(ix0,jx0);
 257             dy00             = _mm_sub_pd(iy0,jy0);
 258             dz00             = _mm_sub_pd(iz0,jz0);
 259
 260             /* Calculate squared distance and things based on it */
 261             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 262
 263             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 264
 265             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 266
 267             /* Load parameters for j particles */
 268             jq0              = _mm_load_sd(charge+jnrA+0);
 269             vdwjidx0A        = 2*vdwtype[jnrA+0];
 270
 271             /**************************
 272              * CALCULATE INTERACTIONS *
 273              **************************/
 274
 275             if (gmx_mm_any_lt(rsq00,rcutoff2))
 276             {
 277
 278             /* Compute parameters for interactions between i and j atoms */
 279             qq00             = _mm_mul_pd(iq0,jq0);
 280             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 281
 282             /* REACTION-FIELD ELECTROSTATICS */
 283             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 284             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 285
 286             /* LENNARD-JONES DISPERSION/REPULSION */
 287
 288             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 289             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 290             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 291             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 292                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 293             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 294
 295             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 296
 297             /* Update potential sum for this i atom from the interaction with this j atom. */
 298             velec            = _mm_and_pd(velec,cutoff_mask);
 299             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 300             velecsum         = _mm_add_pd(velecsum,velec);
 301             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 302             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 303             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 304
 305             fscal            = _mm_add_pd(felec,fvdw);
 306
 307             fscal            = _mm_and_pd(fscal,cutoff_mask);
 308
 309             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 310
 311             /* Update vectorial force */
 312             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 313             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 314             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 315
 316             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 317                                                    _mm_mul_pd(dx00,fscal),
 318                                                    _mm_mul_pd(dy00,fscal),
 319                                                    _mm_mul_pd(dz00,fscal));
 320
 321             }
 322
 323             /* Inner loop uses 57 flops */
 324         }
 325
 326         /* End of innermost loop */
 327
 328         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 329                                               f+i_coord_offset,fshift+i_shift_offset);
 330
 331         ggid                        = gid[iidx];
 332         /* Update potential energies */
 333         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 334         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 335
 336         /* Increment number of inner iterations */
 337         inneriter                  += j_index_end - j_index_start;
 338
 339         /* Outer loop uses 9 flops */
 340     }
 341
 342     /* Increment number of outer iterations */
 343     outeriter        += nri;
 344
 345     /* Update outer/inner flops */
 346
 347     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*57);
 348 }
 349 /*
 350  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_avx_128_fma_double
 351  * Electrostatics interaction: ReactionField
 352  * VdW interaction:            LennardJones
 353  * Geometry:                   Particle-Particle
 354  * Calculate force/pot:        Force
 355  */
 356 void
 357 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_avx_128_fma_double
 358                     (t_nblist                    * gmx_restrict       nlist,
 359                      rvec                        * gmx_restrict          xx,
 360                      rvec                        * gmx_restrict          ff,
 361                      t_forcerec                  * gmx_restrict          fr,
 362                      t_mdatoms                   * gmx_restrict     mdatoms,
 363                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 364                      t_nrnb                      * gmx_restrict        nrnb)
 365 {
 366     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 367      * just 0 for non-waters.
 368      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 369      * jnr indices corresponding to data put in the four positions in the SIMD register.
 370      */
 371     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 372     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 373     int              jnrA,jnrB;
 374     int              j_coord_offsetA,j_coord_offsetB;
 375     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 376     real             rcutoff_scalar;
 377     real             *shiftvec,*fshift,*x,*f;
 378     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 379     int              vdwioffset0;
 380     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 381     int              vdwjidx0A,vdwjidx0B;
 382     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 383     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 384     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 385     real             *charge;
 386     int              nvdwtype;
 387     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 388     int              *vdwtype;
 389     real             *vdwparam;
 390     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 391     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 392     __m128d          dummy_mask,cutoff_mask;
 393     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 394     __m128d          one     = _mm_set1_pd(1.0);
 395     __m128d          two     = _mm_set1_pd(2.0);
 396     x                = xx[0];
 397     f                = ff[0];
 398
 399     nri              = nlist->nri;
 400     iinr             = nlist->iinr;
 401     jindex           = nlist->jindex;
 402     jjnr             = nlist->jjnr;
 403     shiftidx         = nlist->shift;
 404     gid              = nlist->gid;
 405     shiftvec         = fr->shift_vec[0];
 406     fshift           = fr->fshift[0];
 407     facel            = _mm_set1_pd(fr->epsfac);
 408     charge           = mdatoms->chargeA;
 409     krf              = _mm_set1_pd(fr->ic->k_rf);
 410     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 411     crf              = _mm_set1_pd(fr->ic->c_rf);
 412     nvdwtype         = fr->ntype;
 413     vdwparam         = fr->nbfp;
 414     vdwtype          = mdatoms->typeA;
 415
 416     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 417     rcutoff_scalar   = fr->rcoulomb;
 418     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 419     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 420
 421     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 422     rvdw             = _mm_set1_pd(fr->rvdw);
 423
 424     /* Avoid stupid compiler warnings */
 425     jnrA = jnrB = 0;
 426     j_coord_offsetA = 0;
 427     j_coord_offsetB = 0;
 428
 429     outeriter        = 0;
 430     inneriter        = 0;
 431
 432     /* Start outer loop over neighborlists */
 433     for(iidx=0; iidx<nri; iidx++)
 434     {
 435         /* Load shift vector for this list */
 436         i_shift_offset   = DIM*shiftidx[iidx];
 437
 438         /* Load limits for loop over neighbors */
 439         j_index_start    = jindex[iidx];
 440         j_index_end      = jindex[iidx+1];
 441
 442         /* Get outer coordinate index */
 443         inr              = iinr[iidx];
 444         i_coord_offset   = DIM*inr;
 445
 446         /* Load i particle coords and add shift vector */
 447         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 448
 449         fix0             = _mm_setzero_pd();
 450         fiy0             = _mm_setzero_pd();
 451         fiz0             = _mm_setzero_pd();
 452
 453         /* Load parameters for i particles */
 454         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 455         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 456
 457         /* Start inner kernel loop */
 458         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 459         {
 460
 461             /* Get j neighbor index, and coordinate index */
 462             jnrA             = jjnr[jidx];
 463             jnrB             = jjnr[jidx+1];
 464             j_coord_offsetA  = DIM*jnrA;
 465             j_coord_offsetB  = DIM*jnrB;
 466
 467             /* load j atom coordinates */
 468             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 469                                               &jx0,&jy0,&jz0);
 470
 471             /* Calculate displacement vector */
 472             dx00             = _mm_sub_pd(ix0,jx0);
 473             dy00             = _mm_sub_pd(iy0,jy0);
 474             dz00             = _mm_sub_pd(iz0,jz0);
 475
 476             /* Calculate squared distance and things based on it */
 477             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 478
 479             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 480
 481             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 482
 483             /* Load parameters for j particles */
 484             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 485             vdwjidx0A        = 2*vdwtype[jnrA+0];
 486             vdwjidx0B        = 2*vdwtype[jnrB+0];
 487
 488             /**************************
 489              * CALCULATE INTERACTIONS *
 490              **************************/
 491
 492             if (gmx_mm_any_lt(rsq00,rcutoff2))
 493             {
 494
 495             /* Compute parameters for interactions between i and j atoms */
 496             qq00             = _mm_mul_pd(iq0,jq0);
 497             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 498                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 499
 500             /* REACTION-FIELD ELECTROSTATICS */
 501             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 502
 503             /* LENNARD-JONES DISPERSION/REPULSION */
 504
 505             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 506             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 507
 508             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 509
 510             fscal            = _mm_add_pd(felec,fvdw);
 511
 512             fscal            = _mm_and_pd(fscal,cutoff_mask);
 513
 514             /* Update vectorial force */
 515             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 516             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 517             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 518
 519             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 520                                                    _mm_mul_pd(dx00,fscal),
 521                                                    _mm_mul_pd(dy00,fscal),
 522                                                    _mm_mul_pd(dz00,fscal));
 523
 524             }
 525
 526             /* Inner loop uses 40 flops */
 527         }
 528
 529         if(jidx<j_index_end)
 530         {
 531
 532             jnrA             = jjnr[jidx];
 533             j_coord_offsetA  = DIM*jnrA;
 534
 535             /* load j atom coordinates */
 536             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 537                                               &jx0,&jy0,&jz0);
 538
 539             /* Calculate displacement vector */
 540             dx00             = _mm_sub_pd(ix0,jx0);
 541             dy00             = _mm_sub_pd(iy0,jy0);
 542             dz00             = _mm_sub_pd(iz0,jz0);
 543
 544             /* Calculate squared distance and things based on it */
 545             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 546
 547             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 548
 549             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 550
 551             /* Load parameters for j particles */
 552             jq0              = _mm_load_sd(charge+jnrA+0);
 553             vdwjidx0A        = 2*vdwtype[jnrA+0];
 554
 555             /**************************
 556              * CALCULATE INTERACTIONS *
 557              **************************/
 558
 559             if (gmx_mm_any_lt(rsq00,rcutoff2))
 560             {
 561
 562             /* Compute parameters for interactions between i and j atoms */
 563             qq00             = _mm_mul_pd(iq0,jq0);
 564             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 565
 566             /* REACTION-FIELD ELECTROSTATICS */
 567             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 568
 569             /* LENNARD-JONES DISPERSION/REPULSION */
 570
 571             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 572             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 573
 574             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 575
 576             fscal            = _mm_add_pd(felec,fvdw);
 577
 578             fscal            = _mm_and_pd(fscal,cutoff_mask);
 579
 580             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 581
 582             /* Update vectorial force */
 583             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 584             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 585             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 586
 587             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 588                                                    _mm_mul_pd(dx00,fscal),
 589                                                    _mm_mul_pd(dy00,fscal),
 590                                                    _mm_mul_pd(dz00,fscal));
 591
 592             }
 593
 594             /* Inner loop uses 40 flops */
 595         }
 596
 597         /* End of innermost loop */
 598
 599         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 600                                               f+i_coord_offset,fshift+i_shift_offset);
 601
 602         /* Increment number of inner iterations */
 603         inneriter                  += j_index_end - j_index_start;
 604
 605         /* Outer loop uses 7 flops */
 606     }
 607
 608     /* Increment number of outer iterations */
 609     outeriter        += nri;
 610
 611     /* Update outer/inner flops */
 612
 613     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*40);
 614 }