src/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_sse2_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse2_double.h"
  34 #include "kernelutil_x86_sse2_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_VF_sse2_double
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_VF_sse2_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB;
  61     int              j_coord_offsetA,j_coord_offsetB;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwjidx0A,vdwjidx0B;
  69     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  70     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  71     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  72     real             *charge;
  73     int              nvdwtype;
  74     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  75     int              *vdwtype;
  76     real             *vdwparam;
  77     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  78     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  79     __m128d          dummy_mask,cutoff_mask;
  80     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  81     __m128d          one     = _mm_set1_pd(1.0);
  82     __m128d          two     = _mm_set1_pd(2.0);
  83     x                = xx[0];
  84     f                = ff[0];
  85
  86     nri              = nlist->nri;
  87     iinr             = nlist->iinr;
  88     jindex           = nlist->jindex;
  89     jjnr             = nlist->jjnr;
  90     shiftidx         = nlist->shift;
  91     gid              = nlist->gid;
  92     shiftvec         = fr->shift_vec[0];
  93     fshift           = fr->fshift[0];
  94     facel            = _mm_set1_pd(fr->epsfac);
  95     charge           = mdatoms->chargeA;
  96     krf              = _mm_set1_pd(fr->ic->k_rf);
  97     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
  98     crf              = _mm_set1_pd(fr->ic->c_rf);
  99     nvdwtype         = fr->ntype;
 100     vdwparam         = fr->nbfp;
 101     vdwtype          = mdatoms->typeA;
 102
 103     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 104     rcutoff_scalar   = fr->rcoulomb;
 105     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 106     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 107
 108     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 109     rvdw             = _mm_set1_pd(fr->rvdw);
 110
 111     /* Avoid stupid compiler warnings */
 112     jnrA = jnrB = 0;
 113     j_coord_offsetA = 0;
 114     j_coord_offsetB = 0;
 115
 116     outeriter        = 0;
 117     inneriter        = 0;
 118
 119     /* Start outer loop over neighborlists */
 120     for(iidx=0; iidx<nri; iidx++)
 121     {
 122         /* Load shift vector for this list */
 123         i_shift_offset   = DIM*shiftidx[iidx];
 124
 125         /* Load limits for loop over neighbors */
 126         j_index_start    = jindex[iidx];
 127         j_index_end      = jindex[iidx+1];
 128
 129         /* Get outer coordinate index */
 130         inr              = iinr[iidx];
 131         i_coord_offset   = DIM*inr;
 132
 133         /* Load i particle coords and add shift vector */
 134         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 135
 136         fix0             = _mm_setzero_pd();
 137         fiy0             = _mm_setzero_pd();
 138         fiz0             = _mm_setzero_pd();
 139
 140         /* Load parameters for i particles */
 141         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 142         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 143
 144         /* Reset potential sums */
 145         velecsum         = _mm_setzero_pd();
 146         vvdwsum          = _mm_setzero_pd();
 147
 148         /* Start inner kernel loop */
 149         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 150         {
 151
 152             /* Get j neighbor index, and coordinate index */
 153             jnrA             = jjnr[jidx];
 154             jnrB             = jjnr[jidx+1];
 155             j_coord_offsetA  = DIM*jnrA;
 156             j_coord_offsetB  = DIM*jnrB;
 157
 158             /* load j atom coordinates */
 159             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 160                                               &jx0,&jy0,&jz0);
 161
 162             /* Calculate displacement vector */
 163             dx00             = _mm_sub_pd(ix0,jx0);
 164             dy00             = _mm_sub_pd(iy0,jy0);
 165             dz00             = _mm_sub_pd(iz0,jz0);
 166
 167             /* Calculate squared distance and things based on it */
 168             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 169
 170             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 171
 172             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 173
 174             /* Load parameters for j particles */
 175             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 176             vdwjidx0A        = 2*vdwtype[jnrA+0];
 177             vdwjidx0B        = 2*vdwtype[jnrB+0];
 178
 179             /**************************
 180              * CALCULATE INTERACTIONS *
 181              **************************/
 182
 183             if (gmx_mm_any_lt(rsq00,rcutoff2))
 184             {
 185
 186             /* Compute parameters for interactions between i and j atoms */
 187             qq00             = _mm_mul_pd(iq0,jq0);
 188             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 189                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 190
 191             /* REACTION-FIELD ELECTROSTATICS */
 192             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 193             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 194
 195             /* LENNARD-JONES DISPERSION/REPULSION */
 196
 197             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 198             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 199             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 200             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) ,
 201                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 202             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 203
 204             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 205
 206             /* Update potential sum for this i atom from the interaction with this j atom. */
 207             velec            = _mm_and_pd(velec,cutoff_mask);
 208             velecsum         = _mm_add_pd(velecsum,velec);
 209             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 210             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 211
 212             fscal            = _mm_add_pd(felec,fvdw);
 213
 214             fscal            = _mm_and_pd(fscal,cutoff_mask);
 215
 216             /* Calculate temporary vectorial force */
 217             tx               = _mm_mul_pd(fscal,dx00);
 218             ty               = _mm_mul_pd(fscal,dy00);
 219             tz               = _mm_mul_pd(fscal,dz00);
 220
 221             /* Update vectorial force */
 222             fix0             = _mm_add_pd(fix0,tx);
 223             fiy0             = _mm_add_pd(fiy0,ty);
 224             fiz0             = _mm_add_pd(fiz0,tz);
 225
 226             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 227
 228             }
 229
 230             /* Inner loop uses 54 flops */
 231         }
 232
 233         if(jidx<j_index_end)
 234         {
 235
 236             jnrA             = jjnr[jidx];
 237             j_coord_offsetA  = DIM*jnrA;
 238
 239             /* load j atom coordinates */
 240             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 241                                               &jx0,&jy0,&jz0);
 242
 243             /* Calculate displacement vector */
 244             dx00             = _mm_sub_pd(ix0,jx0);
 245             dy00             = _mm_sub_pd(iy0,jy0);
 246             dz00             = _mm_sub_pd(iz0,jz0);
 247
 248             /* Calculate squared distance and things based on it */
 249             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 250
 251             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 252
 253             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 254
 255             /* Load parameters for j particles */
 256             jq0              = _mm_load_sd(charge+jnrA+0);
 257             vdwjidx0A        = 2*vdwtype[jnrA+0];
 258
 259             /**************************
 260              * CALCULATE INTERACTIONS *
 261              **************************/
 262
 263             if (gmx_mm_any_lt(rsq00,rcutoff2))
 264             {
 265
 266             /* Compute parameters for interactions between i and j atoms */
 267             qq00             = _mm_mul_pd(iq0,jq0);
 268             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 269
 270             /* REACTION-FIELD ELECTROSTATICS */
 271             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 272             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 273
 274             /* LENNARD-JONES DISPERSION/REPULSION */
 275
 276             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 277             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 278             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 279             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) ,
 280                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 281             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 282
 283             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 284
 285             /* Update potential sum for this i atom from the interaction with this j atom. */
 286             velec            = _mm_and_pd(velec,cutoff_mask);
 287             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 288             velecsum         = _mm_add_pd(velecsum,velec);
 289             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 290             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 291             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 292
 293             fscal            = _mm_add_pd(felec,fvdw);
 294
 295             fscal            = _mm_and_pd(fscal,cutoff_mask);
 296
 297             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 298
 299             /* Calculate temporary vectorial force */
 300             tx               = _mm_mul_pd(fscal,dx00);
 301             ty               = _mm_mul_pd(fscal,dy00);
 302             tz               = _mm_mul_pd(fscal,dz00);
 303
 304             /* Update vectorial force */
 305             fix0             = _mm_add_pd(fix0,tx);
 306             fiy0             = _mm_add_pd(fiy0,ty);
 307             fiz0             = _mm_add_pd(fiz0,tz);
 308
 309             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 310
 311             }
 312
 313             /* Inner loop uses 54 flops */
 314         }
 315
 316         /* End of innermost loop */
 317
 318         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 319                                               f+i_coord_offset,fshift+i_shift_offset);
 320
 321         ggid                        = gid[iidx];
 322         /* Update potential energies */
 323         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 324         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 325
 326         /* Increment number of inner iterations */
 327         inneriter                  += j_index_end - j_index_start;
 328
 329         /* Outer loop uses 9 flops */
 330     }
 331
 332     /* Increment number of outer iterations */
 333     outeriter        += nri;
 334
 335     /* Update outer/inner flops */
 336
 337     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*54);
 338 }
 339 /*
 340  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_sse2_double
 341  * Electrostatics interaction: ReactionField
 342  * VdW interaction:            LennardJones
 343  * Geometry:                   Particle-Particle
 344  * Calculate force/pot:        Force
 345  */
 346 void
 347 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_sse2_double
 348                     (t_nblist * gmx_restrict                nlist,
 349                      rvec * gmx_restrict                    xx,
 350                      rvec * gmx_restrict                    ff,
 351                      t_forcerec * gmx_restrict              fr,
 352                      t_mdatoms * gmx_restrict               mdatoms,
 353                      nb_kernel_data_t * gmx_restrict        kernel_data,
 354                      t_nrnb * gmx_restrict                  nrnb)
 355 {
 356     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 357      * just 0 for non-waters.
 358      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 359      * jnr indices corresponding to data put in the four positions in the SIMD register.
 360      */
 361     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 362     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 363     int              jnrA,jnrB;
 364     int              j_coord_offsetA,j_coord_offsetB;
 365     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 366     real             rcutoff_scalar;
 367     real             *shiftvec,*fshift,*x,*f;
 368     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 369     int              vdwioffset0;
 370     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 371     int              vdwjidx0A,vdwjidx0B;
 372     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 373     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 374     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 375     real             *charge;
 376     int              nvdwtype;
 377     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 378     int              *vdwtype;
 379     real             *vdwparam;
 380     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 381     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 382     __m128d          dummy_mask,cutoff_mask;
 383     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 384     __m128d          one     = _mm_set1_pd(1.0);
 385     __m128d          two     = _mm_set1_pd(2.0);
 386     x                = xx[0];
 387     f                = ff[0];
 388
 389     nri              = nlist->nri;
 390     iinr             = nlist->iinr;
 391     jindex           = nlist->jindex;
 392     jjnr             = nlist->jjnr;
 393     shiftidx         = nlist->shift;
 394     gid              = nlist->gid;
 395     shiftvec         = fr->shift_vec[0];
 396     fshift           = fr->fshift[0];
 397     facel            = _mm_set1_pd(fr->epsfac);
 398     charge           = mdatoms->chargeA;
 399     krf              = _mm_set1_pd(fr->ic->k_rf);
 400     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 401     crf              = _mm_set1_pd(fr->ic->c_rf);
 402     nvdwtype         = fr->ntype;
 403     vdwparam         = fr->nbfp;
 404     vdwtype          = mdatoms->typeA;
 405
 406     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 407     rcutoff_scalar   = fr->rcoulomb;
 408     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 409     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 410
 411     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 412     rvdw             = _mm_set1_pd(fr->rvdw);
 413
 414     /* Avoid stupid compiler warnings */
 415     jnrA = jnrB = 0;
 416     j_coord_offsetA = 0;
 417     j_coord_offsetB = 0;
 418
 419     outeriter        = 0;
 420     inneriter        = 0;
 421
 422     /* Start outer loop over neighborlists */
 423     for(iidx=0; iidx<nri; iidx++)
 424     {
 425         /* Load shift vector for this list */
 426         i_shift_offset   = DIM*shiftidx[iidx];
 427
 428         /* Load limits for loop over neighbors */
 429         j_index_start    = jindex[iidx];
 430         j_index_end      = jindex[iidx+1];
 431
 432         /* Get outer coordinate index */
 433         inr              = iinr[iidx];
 434         i_coord_offset   = DIM*inr;
 435
 436         /* Load i particle coords and add shift vector */
 437         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 438
 439         fix0             = _mm_setzero_pd();
 440         fiy0             = _mm_setzero_pd();
 441         fiz0             = _mm_setzero_pd();
 442
 443         /* Load parameters for i particles */
 444         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 445         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 446
 447         /* Start inner kernel loop */
 448         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 449         {
 450
 451             /* Get j neighbor index, and coordinate index */
 452             jnrA             = jjnr[jidx];
 453             jnrB             = jjnr[jidx+1];
 454             j_coord_offsetA  = DIM*jnrA;
 455             j_coord_offsetB  = DIM*jnrB;
 456
 457             /* load j atom coordinates */
 458             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 459                                               &jx0,&jy0,&jz0);
 460
 461             /* Calculate displacement vector */
 462             dx00             = _mm_sub_pd(ix0,jx0);
 463             dy00             = _mm_sub_pd(iy0,jy0);
 464             dz00             = _mm_sub_pd(iz0,jz0);
 465
 466             /* Calculate squared distance and things based on it */
 467             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 468
 469             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 470
 471             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 472
 473             /* Load parameters for j particles */
 474             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 475             vdwjidx0A        = 2*vdwtype[jnrA+0];
 476             vdwjidx0B        = 2*vdwtype[jnrB+0];
 477
 478             /**************************
 479              * CALCULATE INTERACTIONS *
 480              **************************/
 481
 482             if (gmx_mm_any_lt(rsq00,rcutoff2))
 483             {
 484
 485             /* Compute parameters for interactions between i and j atoms */
 486             qq00             = _mm_mul_pd(iq0,jq0);
 487             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 488                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 489
 490             /* REACTION-FIELD ELECTROSTATICS */
 491             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 492
 493             /* LENNARD-JONES DISPERSION/REPULSION */
 494
 495             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 496             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 497
 498             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 499
 500             fscal            = _mm_add_pd(felec,fvdw);
 501
 502             fscal            = _mm_and_pd(fscal,cutoff_mask);
 503
 504             /* Calculate temporary vectorial force */
 505             tx               = _mm_mul_pd(fscal,dx00);
 506             ty               = _mm_mul_pd(fscal,dy00);
 507             tz               = _mm_mul_pd(fscal,dz00);
 508
 509             /* Update vectorial force */
 510             fix0             = _mm_add_pd(fix0,tx);
 511             fiy0             = _mm_add_pd(fiy0,ty);
 512             fiz0             = _mm_add_pd(fiz0,tz);
 513
 514             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 515
 516             }
 517
 518             /* Inner loop uses 37 flops */
 519         }
 520
 521         if(jidx<j_index_end)
 522         {
 523
 524             jnrA             = jjnr[jidx];
 525             j_coord_offsetA  = DIM*jnrA;
 526
 527             /* load j atom coordinates */
 528             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 529                                               &jx0,&jy0,&jz0);
 530
 531             /* Calculate displacement vector */
 532             dx00             = _mm_sub_pd(ix0,jx0);
 533             dy00             = _mm_sub_pd(iy0,jy0);
 534             dz00             = _mm_sub_pd(iz0,jz0);
 535
 536             /* Calculate squared distance and things based on it */
 537             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 538
 539             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 540
 541             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 542
 543             /* Load parameters for j particles */
 544             jq0              = _mm_load_sd(charge+jnrA+0);
 545             vdwjidx0A        = 2*vdwtype[jnrA+0];
 546
 547             /**************************
 548              * CALCULATE INTERACTIONS *
 549              **************************/
 550
 551             if (gmx_mm_any_lt(rsq00,rcutoff2))
 552             {
 553
 554             /* Compute parameters for interactions between i and j atoms */
 555             qq00             = _mm_mul_pd(iq0,jq0);
 556             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 557
 558             /* REACTION-FIELD ELECTROSTATICS */
 559             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 560
 561             /* LENNARD-JONES DISPERSION/REPULSION */
 562
 563             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 564             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 565
 566             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 567
 568             fscal            = _mm_add_pd(felec,fvdw);
 569
 570             fscal            = _mm_and_pd(fscal,cutoff_mask);
 571
 572             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 573
 574             /* Calculate temporary vectorial force */
 575             tx               = _mm_mul_pd(fscal,dx00);
 576             ty               = _mm_mul_pd(fscal,dy00);
 577             tz               = _mm_mul_pd(fscal,dz00);
 578
 579             /* Update vectorial force */
 580             fix0             = _mm_add_pd(fix0,tx);
 581             fiy0             = _mm_add_pd(fiy0,ty);
 582             fiz0             = _mm_add_pd(fiz0,tz);
 583
 584             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 585
 586             }
 587
 588             /* Inner loop uses 37 flops */
 589         }
 590
 591         /* End of innermost loop */
 592
 593         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 594                                               f+i_coord_offset,fshift+i_shift_offset);
 595
 596         /* Increment number of inner iterations */
 597         inneriter                  += j_index_end - j_index_start;
 598
 599         /* Outer loop uses 7 flops */
 600     }
 601
 602     /* Increment number of outer iterations */
 603     outeriter        += nri;
 604
 605     /* Update outer/inner flops */
 606
 607     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*37);
 608 }