src/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecNone_VdwLJSh_GeomP1P1_sse4_1_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_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_sse4_1_double.h"
  34 #include "kernelutil_x86_sse4_1_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSh_GeomP1P1_VF_sse4_1_double
  38  * Electrostatics interaction: None
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecNone_VdwLJSh_GeomP1P1_VF_sse4_1_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     int              nvdwtype;
  72     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  73     int              *vdwtype;
  74     real             *vdwparam;
  75     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  76     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  77     __m128d          dummy_mask,cutoff_mask;
  78     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  79     __m128d          one     = _mm_set1_pd(1.0);
  80     __m128d          two     = _mm_set1_pd(2.0);
  81     x                = xx[0];
  82     f                = ff[0];
  83
  84     nri              = nlist->nri;
  85     iinr             = nlist->iinr;
  86     jindex           = nlist->jindex;
  87     jjnr             = nlist->jjnr;
  88     shiftidx         = nlist->shift;
  89     gid              = nlist->gid;
  90     shiftvec         = fr->shift_vec[0];
  91     fshift           = fr->fshift[0];
  92     nvdwtype         = fr->ntype;
  93     vdwparam         = fr->nbfp;
  94     vdwtype          = mdatoms->typeA;
  95
  96     rcutoff_scalar   = fr->rvdw;
  97     rcutoff          = _mm_set1_pd(rcutoff_scalar);
  98     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
  99
 100     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 101     rvdw             = _mm_set1_pd(fr->rvdw);
 102
 103     /* Avoid stupid compiler warnings */
 104     jnrA = jnrB = 0;
 105     j_coord_offsetA = 0;
 106     j_coord_offsetB = 0;
 107
 108     outeriter        = 0;
 109     inneriter        = 0;
 110
 111     /* Start outer loop over neighborlists */
 112     for(iidx=0; iidx<nri; iidx++)
 113     {
 114         /* Load shift vector for this list */
 115         i_shift_offset   = DIM*shiftidx[iidx];
 116
 117         /* Load limits for loop over neighbors */
 118         j_index_start    = jindex[iidx];
 119         j_index_end      = jindex[iidx+1];
 120
 121         /* Get outer coordinate index */
 122         inr              = iinr[iidx];
 123         i_coord_offset   = DIM*inr;
 124
 125         /* Load i particle coords and add shift vector */
 126         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 127
 128         fix0             = _mm_setzero_pd();
 129         fiy0             = _mm_setzero_pd();
 130         fiz0             = _mm_setzero_pd();
 131
 132         /* Load parameters for i particles */
 133         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 134
 135         /* Reset potential sums */
 136         vvdwsum          = _mm_setzero_pd();
 137
 138         /* Start inner kernel loop */
 139         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 140         {
 141
 142             /* Get j neighbor index, and coordinate index */
 143             jnrA             = jjnr[jidx];
 144             jnrB             = jjnr[jidx+1];
 145             j_coord_offsetA  = DIM*jnrA;
 146             j_coord_offsetB  = DIM*jnrB;
 147
 148             /* load j atom coordinates */
 149             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 150                                               &jx0,&jy0,&jz0);
 151
 152             /* Calculate displacement vector */
 153             dx00             = _mm_sub_pd(ix0,jx0);
 154             dy00             = _mm_sub_pd(iy0,jy0);
 155             dz00             = _mm_sub_pd(iz0,jz0);
 156
 157             /* Calculate squared distance and things based on it */
 158             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 159
 160             rinvsq00         = gmx_mm_inv_pd(rsq00);
 161
 162             /* Load parameters for j particles */
 163             vdwjidx0A        = 2*vdwtype[jnrA+0];
 164             vdwjidx0B        = 2*vdwtype[jnrB+0];
 165
 166             /**************************
 167              * CALCULATE INTERACTIONS *
 168              **************************/
 169
 170             if (gmx_mm_any_lt(rsq00,rcutoff2))
 171             {
 172
 173             /* Compute parameters for interactions between i and j atoms */
 174             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 175                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 176
 177             /* LENNARD-JONES DISPERSION/REPULSION */
 178
 179             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 180             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 181             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 182             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) ,
 183                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 184             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 185
 186             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 187
 188             /* Update potential sum for this i atom from the interaction with this j atom. */
 189             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 190             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 191
 192             fscal            = fvdw;
 193
 194             fscal            = _mm_and_pd(fscal,cutoff_mask);
 195
 196             /* Calculate temporary vectorial force */
 197             tx               = _mm_mul_pd(fscal,dx00);
 198             ty               = _mm_mul_pd(fscal,dy00);
 199             tz               = _mm_mul_pd(fscal,dz00);
 200
 201             /* Update vectorial force */
 202             fix0             = _mm_add_pd(fix0,tx);
 203             fiy0             = _mm_add_pd(fiy0,ty);
 204             fiz0             = _mm_add_pd(fiz0,tz);
 205
 206             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 207
 208             }
 209
 210             /* Inner loop uses 41 flops */
 211         }
 212
 213         if(jidx<j_index_end)
 214         {
 215
 216             jnrA             = jjnr[jidx];
 217             j_coord_offsetA  = DIM*jnrA;
 218
 219             /* load j atom coordinates */
 220             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 221                                               &jx0,&jy0,&jz0);
 222
 223             /* Calculate displacement vector */
 224             dx00             = _mm_sub_pd(ix0,jx0);
 225             dy00             = _mm_sub_pd(iy0,jy0);
 226             dz00             = _mm_sub_pd(iz0,jz0);
 227
 228             /* Calculate squared distance and things based on it */
 229             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 230
 231             rinvsq00         = gmx_mm_inv_pd(rsq00);
 232
 233             /* Load parameters for j particles */
 234             vdwjidx0A        = 2*vdwtype[jnrA+0];
 235
 236             /**************************
 237              * CALCULATE INTERACTIONS *
 238              **************************/
 239
 240             if (gmx_mm_any_lt(rsq00,rcutoff2))
 241             {
 242
 243             /* Compute parameters for interactions between i and j atoms */
 244             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 245
 246             /* LENNARD-JONES DISPERSION/REPULSION */
 247
 248             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 249             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 250             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 251             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) ,
 252                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 253             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 254
 255             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 256
 257             /* Update potential sum for this i atom from the interaction with this j atom. */
 258             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 259             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 260             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 261
 262             fscal            = fvdw;
 263
 264             fscal            = _mm_and_pd(fscal,cutoff_mask);
 265
 266             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 267
 268             /* Calculate temporary vectorial force */
 269             tx               = _mm_mul_pd(fscal,dx00);
 270             ty               = _mm_mul_pd(fscal,dy00);
 271             tz               = _mm_mul_pd(fscal,dz00);
 272
 273             /* Update vectorial force */
 274             fix0             = _mm_add_pd(fix0,tx);
 275             fiy0             = _mm_add_pd(fiy0,ty);
 276             fiz0             = _mm_add_pd(fiz0,tz);
 277
 278             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 279
 280             }
 281
 282             /* Inner loop uses 41 flops */
 283         }
 284
 285         /* End of innermost loop */
 286
 287         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 288                                               f+i_coord_offset,fshift+i_shift_offset);
 289
 290         ggid                        = gid[iidx];
 291         /* Update potential energies */
 292         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 293
 294         /* Increment number of inner iterations */
 295         inneriter                  += j_index_end - j_index_start;
 296
 297         /* Outer loop uses 7 flops */
 298     }
 299
 300     /* Increment number of outer iterations */
 301     outeriter        += nri;
 302
 303     /* Update outer/inner flops */
 304
 305     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*41);
 306 }
 307 /*
 308  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSh_GeomP1P1_F_sse4_1_double
 309  * Electrostatics interaction: None
 310  * VdW interaction:            LennardJones
 311  * Geometry:                   Particle-Particle
 312  * Calculate force/pot:        Force
 313  */
 314 void
 315 nb_kernel_ElecNone_VdwLJSh_GeomP1P1_F_sse4_1_double
 316                     (t_nblist * gmx_restrict                nlist,
 317                      rvec * gmx_restrict                    xx,
 318                      rvec * gmx_restrict                    ff,
 319                      t_forcerec * gmx_restrict              fr,
 320                      t_mdatoms * gmx_restrict               mdatoms,
 321                      nb_kernel_data_t * gmx_restrict        kernel_data,
 322                      t_nrnb * gmx_restrict                  nrnb)
 323 {
 324     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 325      * just 0 for non-waters.
 326      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 327      * jnr indices corresponding to data put in the four positions in the SIMD register.
 328      */
 329     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 330     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 331     int              jnrA,jnrB;
 332     int              j_coord_offsetA,j_coord_offsetB;
 333     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 334     real             rcutoff_scalar;
 335     real             *shiftvec,*fshift,*x,*f;
 336     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 337     int              vdwioffset0;
 338     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 339     int              vdwjidx0A,vdwjidx0B;
 340     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 341     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 342     int              nvdwtype;
 343     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 344     int              *vdwtype;
 345     real             *vdwparam;
 346     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 347     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 348     __m128d          dummy_mask,cutoff_mask;
 349     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 350     __m128d          one     = _mm_set1_pd(1.0);
 351     __m128d          two     = _mm_set1_pd(2.0);
 352     x                = xx[0];
 353     f                = ff[0];
 354
 355     nri              = nlist->nri;
 356     iinr             = nlist->iinr;
 357     jindex           = nlist->jindex;
 358     jjnr             = nlist->jjnr;
 359     shiftidx         = nlist->shift;
 360     gid              = nlist->gid;
 361     shiftvec         = fr->shift_vec[0];
 362     fshift           = fr->fshift[0];
 363     nvdwtype         = fr->ntype;
 364     vdwparam         = fr->nbfp;
 365     vdwtype          = mdatoms->typeA;
 366
 367     rcutoff_scalar   = fr->rvdw;
 368     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 369     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 370
 371     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 372     rvdw             = _mm_set1_pd(fr->rvdw);
 373
 374     /* Avoid stupid compiler warnings */
 375     jnrA = jnrB = 0;
 376     j_coord_offsetA = 0;
 377     j_coord_offsetB = 0;
 378
 379     outeriter        = 0;
 380     inneriter        = 0;
 381
 382     /* Start outer loop over neighborlists */
 383     for(iidx=0; iidx<nri; iidx++)
 384     {
 385         /* Load shift vector for this list */
 386         i_shift_offset   = DIM*shiftidx[iidx];
 387
 388         /* Load limits for loop over neighbors */
 389         j_index_start    = jindex[iidx];
 390         j_index_end      = jindex[iidx+1];
 391
 392         /* Get outer coordinate index */
 393         inr              = iinr[iidx];
 394         i_coord_offset   = DIM*inr;
 395
 396         /* Load i particle coords and add shift vector */
 397         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 398
 399         fix0             = _mm_setzero_pd();
 400         fiy0             = _mm_setzero_pd();
 401         fiz0             = _mm_setzero_pd();
 402
 403         /* Load parameters for i particles */
 404         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 405
 406         /* Start inner kernel loop */
 407         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 408         {
 409
 410             /* Get j neighbor index, and coordinate index */
 411             jnrA             = jjnr[jidx];
 412             jnrB             = jjnr[jidx+1];
 413             j_coord_offsetA  = DIM*jnrA;
 414             j_coord_offsetB  = DIM*jnrB;
 415
 416             /* load j atom coordinates */
 417             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 418                                               &jx0,&jy0,&jz0);
 419
 420             /* Calculate displacement vector */
 421             dx00             = _mm_sub_pd(ix0,jx0);
 422             dy00             = _mm_sub_pd(iy0,jy0);
 423             dz00             = _mm_sub_pd(iz0,jz0);
 424
 425             /* Calculate squared distance and things based on it */
 426             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 427
 428             rinvsq00         = gmx_mm_inv_pd(rsq00);
 429
 430             /* Load parameters for j particles */
 431             vdwjidx0A        = 2*vdwtype[jnrA+0];
 432             vdwjidx0B        = 2*vdwtype[jnrB+0];
 433
 434             /**************************
 435              * CALCULATE INTERACTIONS *
 436              **************************/
 437
 438             if (gmx_mm_any_lt(rsq00,rcutoff2))
 439             {
 440
 441             /* Compute parameters for interactions between i and j atoms */
 442             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 443                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 444
 445             /* LENNARD-JONES DISPERSION/REPULSION */
 446
 447             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 448             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 449
 450             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 451
 452             fscal            = fvdw;
 453
 454             fscal            = _mm_and_pd(fscal,cutoff_mask);
 455
 456             /* Calculate temporary vectorial force */
 457             tx               = _mm_mul_pd(fscal,dx00);
 458             ty               = _mm_mul_pd(fscal,dy00);
 459             tz               = _mm_mul_pd(fscal,dz00);
 460
 461             /* Update vectorial force */
 462             fix0             = _mm_add_pd(fix0,tx);
 463             fiy0             = _mm_add_pd(fiy0,ty);
 464             fiz0             = _mm_add_pd(fiz0,tz);
 465
 466             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 467
 468             }
 469
 470             /* Inner loop uses 30 flops */
 471         }
 472
 473         if(jidx<j_index_end)
 474         {
 475
 476             jnrA             = jjnr[jidx];
 477             j_coord_offsetA  = DIM*jnrA;
 478
 479             /* load j atom coordinates */
 480             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 481                                               &jx0,&jy0,&jz0);
 482
 483             /* Calculate displacement vector */
 484             dx00             = _mm_sub_pd(ix0,jx0);
 485             dy00             = _mm_sub_pd(iy0,jy0);
 486             dz00             = _mm_sub_pd(iz0,jz0);
 487
 488             /* Calculate squared distance and things based on it */
 489             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 490
 491             rinvsq00         = gmx_mm_inv_pd(rsq00);
 492
 493             /* Load parameters for j particles */
 494             vdwjidx0A        = 2*vdwtype[jnrA+0];
 495
 496             /**************************
 497              * CALCULATE INTERACTIONS *
 498              **************************/
 499
 500             if (gmx_mm_any_lt(rsq00,rcutoff2))
 501             {
 502
 503             /* Compute parameters for interactions between i and j atoms */
 504             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 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            = fvdw;
 514
 515             fscal            = _mm_and_pd(fscal,cutoff_mask);
 516
 517             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 518
 519             /* Calculate temporary vectorial force */
 520             tx               = _mm_mul_pd(fscal,dx00);
 521             ty               = _mm_mul_pd(fscal,dy00);
 522             tz               = _mm_mul_pd(fscal,dz00);
 523
 524             /* Update vectorial force */
 525             fix0             = _mm_add_pd(fix0,tx);
 526             fiy0             = _mm_add_pd(fiy0,ty);
 527             fiz0             = _mm_add_pd(fiz0,tz);
 528
 529             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 530
 531             }
 532
 533             /* Inner loop uses 30 flops */
 534         }
 535
 536         /* End of innermost loop */
 537
 538         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 539                                               f+i_coord_offset,fshift+i_shift_offset);
 540
 541         /* Increment number of inner iterations */
 542         inneriter                  += j_index_end - j_index_start;
 543
 544         /* Outer loop uses 6 flops */
 545     }
 546
 547     /* Increment number of outer iterations */
 548     outeriter        += nri;
 549
 550     /* Update outer/inner flops */
 551
 552     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*30);
 553 }