src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_sse4_1_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 sse4_1_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_double.h"
  50 #include "kernelutil_x86_sse4_1_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_VF_sse4_1_double
  54  * Electrostatics interaction: None
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_VF_sse4_1_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     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           c6grid_00;
  94     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
  95     real             *vdwgridparam;
  96     __m128d           one_half = _mm_set1_pd(0.5);
  97     __m128d           minus_one = _mm_set1_pd(-1.0);
  98     __m128d          dummy_mask,cutoff_mask;
  99     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 100     __m128d          one     = _mm_set1_pd(1.0);
 101     __m128d          two     = _mm_set1_pd(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     nvdwtype         = fr->ntype;
 114     vdwparam         = fr->nbfp;
 115     vdwtype          = mdatoms->typeA;
 116     vdwgridparam     = fr->ljpme_c6grid;
 117     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 118     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 119     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 120
 121     rcutoff_scalar   = fr->rvdw;
 122     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 123     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 124
 125     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 126     rvdw             = _mm_set1_pd(fr->rvdw);
 127
 128     /* Avoid stupid compiler warnings */
 129     jnrA = jnrB = 0;
 130     j_coord_offsetA = 0;
 131     j_coord_offsetB = 0;
 132
 133     outeriter        = 0;
 134     inneriter        = 0;
 135
 136     /* Start outer loop over neighborlists */
 137     for(iidx=0; iidx<nri; iidx++)
 138     {
 139         /* Load shift vector for this list */
 140         i_shift_offset   = DIM*shiftidx[iidx];
 141
 142         /* Load limits for loop over neighbors */
 143         j_index_start    = jindex[iidx];
 144         j_index_end      = jindex[iidx+1];
 145
 146         /* Get outer coordinate index */
 147         inr              = iinr[iidx];
 148         i_coord_offset   = DIM*inr;
 149
 150         /* Load i particle coords and add shift vector */
 151         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 152
 153         fix0             = _mm_setzero_pd();
 154         fiy0             = _mm_setzero_pd();
 155         fiz0             = _mm_setzero_pd();
 156
 157         /* Load parameters for i particles */
 158         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 159
 160         /* Reset potential sums */
 161         vvdwsum          = _mm_setzero_pd();
 162
 163         /* Start inner kernel loop */
 164         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 165         {
 166
 167             /* Get j neighbor index, and coordinate index */
 168             jnrA             = jjnr[jidx];
 169             jnrB             = jjnr[jidx+1];
 170             j_coord_offsetA  = DIM*jnrA;
 171             j_coord_offsetB  = DIM*jnrB;
 172
 173             /* load j atom coordinates */
 174             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 175                                               &jx0,&jy0,&jz0);
 176
 177             /* Calculate displacement vector */
 178             dx00             = _mm_sub_pd(ix0,jx0);
 179             dy00             = _mm_sub_pd(iy0,jy0);
 180             dz00             = _mm_sub_pd(iz0,jz0);
 181
 182             /* Calculate squared distance and things based on it */
 183             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 184
 185             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 186
 187             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 188
 189             /* Load parameters for j particles */
 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             r00              = _mm_mul_pd(rsq00,rinv00);
 201
 202             /* Compute parameters for interactions between i and j atoms */
 203             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 204                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 205             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 206                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 207
 208             /* Analytical LJ-PME */
 209             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 210             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 211             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 212             exponent         = gmx_simd_exp_d(ewcljrsq);
 213             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 214             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 215             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 216             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 217             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 218             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),
 219                                _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_add_pd(_mm_mul_pd(c6_00,sh_vdw_invrcut6),_mm_mul_pd(c6grid_00,sh_lj_ewald))),one_sixth));
 220             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 221             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,_mm_sub_pd(vvdw6,_mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6)))),rinvsq00);
 222
 223             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 224
 225             /* Update potential sum for this i atom from the interaction with this j atom. */
 226             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 227             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 228
 229             fscal            = fvdw;
 230
 231             fscal            = _mm_and_pd(fscal,cutoff_mask);
 232
 233             /* Calculate temporary vectorial force */
 234             tx               = _mm_mul_pd(fscal,dx00);
 235             ty               = _mm_mul_pd(fscal,dy00);
 236             tz               = _mm_mul_pd(fscal,dz00);
 237
 238             /* Update vectorial force */
 239             fix0             = _mm_add_pd(fix0,tx);
 240             fiy0             = _mm_add_pd(fiy0,ty);
 241             fiz0             = _mm_add_pd(fiz0,tz);
 242
 243             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 244
 245             }
 246
 247             /* Inner loop uses 61 flops */
 248         }
 249
 250         if(jidx<j_index_end)
 251         {
 252
 253             jnrA             = jjnr[jidx];
 254             j_coord_offsetA  = DIM*jnrA;
 255
 256             /* load j atom coordinates */
 257             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 258                                               &jx0,&jy0,&jz0);
 259
 260             /* Calculate displacement vector */
 261             dx00             = _mm_sub_pd(ix0,jx0);
 262             dy00             = _mm_sub_pd(iy0,jy0);
 263             dz00             = _mm_sub_pd(iz0,jz0);
 264
 265             /* Calculate squared distance and things based on it */
 266             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 267
 268             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 269
 270             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 271
 272             /* Load parameters for j particles */
 273             vdwjidx0A        = 2*vdwtype[jnrA+0];
 274
 275             /**************************
 276              * CALCULATE INTERACTIONS *
 277              **************************/
 278
 279             if (gmx_mm_any_lt(rsq00,rcutoff2))
 280             {
 281
 282             r00              = _mm_mul_pd(rsq00,rinv00);
 283
 284             /* Compute parameters for interactions between i and j atoms */
 285             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 286
 287             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 288
 289             /* Analytical LJ-PME */
 290             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 291             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 292             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 293             exponent         = gmx_simd_exp_d(ewcljrsq);
 294             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 295             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 296             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 297             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 298             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 299             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),
 300                                _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_add_pd(_mm_mul_pd(c6_00,sh_vdw_invrcut6),_mm_mul_pd(c6grid_00,sh_lj_ewald))),one_sixth));
 301             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 302             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,_mm_sub_pd(vvdw6,_mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6)))),rinvsq00);
 303
 304             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 305
 306             /* Update potential sum for this i atom from the interaction with this j atom. */
 307             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 308             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 309             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 310
 311             fscal            = fvdw;
 312
 313             fscal            = _mm_and_pd(fscal,cutoff_mask);
 314
 315             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 316
 317             /* Calculate temporary vectorial force */
 318             tx               = _mm_mul_pd(fscal,dx00);
 319             ty               = _mm_mul_pd(fscal,dy00);
 320             tz               = _mm_mul_pd(fscal,dz00);
 321
 322             /* Update vectorial force */
 323             fix0             = _mm_add_pd(fix0,tx);
 324             fiy0             = _mm_add_pd(fiy0,ty);
 325             fiz0             = _mm_add_pd(fiz0,tz);
 326
 327             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 328
 329             }
 330
 331             /* Inner loop uses 61 flops */
 332         }
 333
 334         /* End of innermost loop */
 335
 336         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 337                                               f+i_coord_offset,fshift+i_shift_offset);
 338
 339         ggid                        = gid[iidx];
 340         /* Update potential energies */
 341         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 342
 343         /* Increment number of inner iterations */
 344         inneriter                  += j_index_end - j_index_start;
 345
 346         /* Outer loop uses 7 flops */
 347     }
 348
 349     /* Increment number of outer iterations */
 350     outeriter        += nri;
 351
 352     /* Update outer/inner flops */
 353
 354     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*61);
 355 }
 356 /*
 357  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_F_sse4_1_double
 358  * Electrostatics interaction: None
 359  * VdW interaction:            LJEwald
 360  * Geometry:                   Particle-Particle
 361  * Calculate force/pot:        Force
 362  */
 363 void
 364 nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_F_sse4_1_double
 365                     (t_nblist                    * gmx_restrict       nlist,
 366                      rvec                        * gmx_restrict          xx,
 367                      rvec                        * gmx_restrict          ff,
 368                      t_forcerec                  * gmx_restrict          fr,
 369                      t_mdatoms                   * gmx_restrict     mdatoms,
 370                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 371                      t_nrnb                      * gmx_restrict        nrnb)
 372 {
 373     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 374      * just 0 for non-waters.
 375      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 376      * jnr indices corresponding to data put in the four positions in the SIMD register.
 377      */
 378     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 379     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 380     int              jnrA,jnrB;
 381     int              j_coord_offsetA,j_coord_offsetB;
 382     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 383     real             rcutoff_scalar;
 384     real             *shiftvec,*fshift,*x,*f;
 385     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 386     int              vdwioffset0;
 387     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 388     int              vdwjidx0A,vdwjidx0B;
 389     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 390     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 391     int              nvdwtype;
 392     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 393     int              *vdwtype;
 394     real             *vdwparam;
 395     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 396     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 397     __m128d           c6grid_00;
 398     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 399     real             *vdwgridparam;
 400     __m128d           one_half = _mm_set1_pd(0.5);
 401     __m128d           minus_one = _mm_set1_pd(-1.0);
 402     __m128d          dummy_mask,cutoff_mask;
 403     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 404     __m128d          one     = _mm_set1_pd(1.0);
 405     __m128d          two     = _mm_set1_pd(2.0);
 406     x                = xx[0];
 407     f                = ff[0];
 408
 409     nri              = nlist->nri;
 410     iinr             = nlist->iinr;
 411     jindex           = nlist->jindex;
 412     jjnr             = nlist->jjnr;
 413     shiftidx         = nlist->shift;
 414     gid              = nlist->gid;
 415     shiftvec         = fr->shift_vec[0];
 416     fshift           = fr->fshift[0];
 417     nvdwtype         = fr->ntype;
 418     vdwparam         = fr->nbfp;
 419     vdwtype          = mdatoms->typeA;
 420     vdwgridparam     = fr->ljpme_c6grid;
 421     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 422     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 423     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 424
 425     rcutoff_scalar   = fr->rvdw;
 426     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 427     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 428
 429     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 430     rvdw             = _mm_set1_pd(fr->rvdw);
 431
 432     /* Avoid stupid compiler warnings */
 433     jnrA = jnrB = 0;
 434     j_coord_offsetA = 0;
 435     j_coord_offsetB = 0;
 436
 437     outeriter        = 0;
 438     inneriter        = 0;
 439
 440     /* Start outer loop over neighborlists */
 441     for(iidx=0; iidx<nri; iidx++)
 442     {
 443         /* Load shift vector for this list */
 444         i_shift_offset   = DIM*shiftidx[iidx];
 445
 446         /* Load limits for loop over neighbors */
 447         j_index_start    = jindex[iidx];
 448         j_index_end      = jindex[iidx+1];
 449
 450         /* Get outer coordinate index */
 451         inr              = iinr[iidx];
 452         i_coord_offset   = DIM*inr;
 453
 454         /* Load i particle coords and add shift vector */
 455         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 456
 457         fix0             = _mm_setzero_pd();
 458         fiy0             = _mm_setzero_pd();
 459         fiz0             = _mm_setzero_pd();
 460
 461         /* Load parameters for i particles */
 462         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 463
 464         /* Start inner kernel loop */
 465         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 466         {
 467
 468             /* Get j neighbor index, and coordinate index */
 469             jnrA             = jjnr[jidx];
 470             jnrB             = jjnr[jidx+1];
 471             j_coord_offsetA  = DIM*jnrA;
 472             j_coord_offsetB  = DIM*jnrB;
 473
 474             /* load j atom coordinates */
 475             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 476                                               &jx0,&jy0,&jz0);
 477
 478             /* Calculate displacement vector */
 479             dx00             = _mm_sub_pd(ix0,jx0);
 480             dy00             = _mm_sub_pd(iy0,jy0);
 481             dz00             = _mm_sub_pd(iz0,jz0);
 482
 483             /* Calculate squared distance and things based on it */
 484             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 485
 486             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 487
 488             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 489
 490             /* Load parameters for j particles */
 491             vdwjidx0A        = 2*vdwtype[jnrA+0];
 492             vdwjidx0B        = 2*vdwtype[jnrB+0];
 493
 494             /**************************
 495              * CALCULATE INTERACTIONS *
 496              **************************/
 497
 498             if (gmx_mm_any_lt(rsq00,rcutoff2))
 499             {
 500
 501             r00              = _mm_mul_pd(rsq00,rinv00);
 502
 503             /* Compute parameters for interactions between i and j atoms */
 504             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 505                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 506             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 507                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 508
 509             /* Analytical LJ-PME */
 510             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 511             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 512             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 513             exponent         = gmx_simd_exp_d(ewcljrsq);
 514             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 515             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 516             /* f6A = 6 * C6grid * (1 - poly) */
 517             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 518             /* f6B = C6grid * exponent * beta^6 */
 519             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 520             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 521             fvdw              = _mm_mul_pd(_mm_add_pd(_mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),_mm_sub_pd(c6_00,f6A)),rinvsix),f6B),rinvsq00);
 522
 523             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 524
 525             fscal            = fvdw;
 526
 527             fscal            = _mm_and_pd(fscal,cutoff_mask);
 528
 529             /* Calculate temporary vectorial force */
 530             tx               = _mm_mul_pd(fscal,dx00);
 531             ty               = _mm_mul_pd(fscal,dy00);
 532             tz               = _mm_mul_pd(fscal,dz00);
 533
 534             /* Update vectorial force */
 535             fix0             = _mm_add_pd(fix0,tx);
 536             fiy0             = _mm_add_pd(fiy0,ty);
 537             fiz0             = _mm_add_pd(fiz0,tz);
 538
 539             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 540
 541             }
 542
 543             /* Inner loop uses 49 flops */
 544         }
 545
 546         if(jidx<j_index_end)
 547         {
 548
 549             jnrA             = jjnr[jidx];
 550             j_coord_offsetA  = DIM*jnrA;
 551
 552             /* load j atom coordinates */
 553             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 554                                               &jx0,&jy0,&jz0);
 555
 556             /* Calculate displacement vector */
 557             dx00             = _mm_sub_pd(ix0,jx0);
 558             dy00             = _mm_sub_pd(iy0,jy0);
 559             dz00             = _mm_sub_pd(iz0,jz0);
 560
 561             /* Calculate squared distance and things based on it */
 562             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 563
 564             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 565
 566             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 567
 568             /* Load parameters for j particles */
 569             vdwjidx0A        = 2*vdwtype[jnrA+0];
 570
 571             /**************************
 572              * CALCULATE INTERACTIONS *
 573              **************************/
 574
 575             if (gmx_mm_any_lt(rsq00,rcutoff2))
 576             {
 577
 578             r00              = _mm_mul_pd(rsq00,rinv00);
 579
 580             /* Compute parameters for interactions between i and j atoms */
 581             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 582
 583             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 584
 585             /* Analytical LJ-PME */
 586             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 587             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 588             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 589             exponent         = gmx_simd_exp_d(ewcljrsq);
 590             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 591             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 592             /* f6A = 6 * C6grid * (1 - poly) */
 593             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 594             /* f6B = C6grid * exponent * beta^6 */
 595             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 596             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 597             fvdw              = _mm_mul_pd(_mm_add_pd(_mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),_mm_sub_pd(c6_00,f6A)),rinvsix),f6B),rinvsq00);
 598
 599             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 600
 601             fscal            = fvdw;
 602
 603             fscal            = _mm_and_pd(fscal,cutoff_mask);
 604
 605             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 606
 607             /* Calculate temporary vectorial force */
 608             tx               = _mm_mul_pd(fscal,dx00);
 609             ty               = _mm_mul_pd(fscal,dy00);
 610             tz               = _mm_mul_pd(fscal,dz00);
 611
 612             /* Update vectorial force */
 613             fix0             = _mm_add_pd(fix0,tx);
 614             fiy0             = _mm_add_pd(fiy0,ty);
 615             fiz0             = _mm_add_pd(fiz0,tz);
 616
 617             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 618
 619             }
 620
 621             /* Inner loop uses 49 flops */
 622         }
 623
 624         /* End of innermost loop */
 625
 626         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 627                                               f+i_coord_offset,fshift+i_shift_offset);
 628
 629         /* Increment number of inner iterations */
 630         inneriter                  += j_index_end - j_index_start;
 631
 632         /* Outer loop uses 6 flops */
 633     }
 634
 635     /* Increment number of outer iterations */
 636     outeriter        += nri;
 637
 638     /* Update outer/inner flops */
 639
 640     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*49);
 641 }