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 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_double.h"
  48 #include "kernelutil_x86_sse4_1_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_VF_sse4_1_double
  52  * Electrostatics interaction: None
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_VF_sse4_1_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     int              nvdwtype;
  86     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  90     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  91     __m128d           c6grid_00;
  92     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
  93     real             *vdwgridparam;
  94     __m128d           one_half = _mm_set1_pd(0.5);
  95     __m128d           minus_one = _mm_set1_pd(-1.0);
  96     __m128d          dummy_mask,cutoff_mask;
  97     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  98     __m128d          one     = _mm_set1_pd(1.0);
  99     __m128d          two     = _mm_set1_pd(2.0);
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     nvdwtype         = fr->ntype;
 112     vdwparam         = fr->nbfp;
 113     vdwtype          = mdatoms->typeA;
 114     vdwgridparam     = fr->ljpme_c6grid;
 115     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 116     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 117     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 118
 119     rcutoff_scalar   = fr->rvdw;
 120     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 121     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 122
 123     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 124     rvdw             = _mm_set1_pd(fr->rvdw);
 125
 126     /* Avoid stupid compiler warnings */
 127     jnrA = jnrB = 0;
 128     j_coord_offsetA = 0;
 129     j_coord_offsetB = 0;
 130
 131     outeriter        = 0;
 132     inneriter        = 0;
 133
 134     /* Start outer loop over neighborlists */
 135     for(iidx=0; iidx<nri; iidx++)
 136     {
 137         /* Load shift vector for this list */
 138         i_shift_offset   = DIM*shiftidx[iidx];
 139
 140         /* Load limits for loop over neighbors */
 141         j_index_start    = jindex[iidx];
 142         j_index_end      = jindex[iidx+1];
 143
 144         /* Get outer coordinate index */
 145         inr              = iinr[iidx];
 146         i_coord_offset   = DIM*inr;
 147
 148         /* Load i particle coords and add shift vector */
 149         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 150
 151         fix0             = _mm_setzero_pd();
 152         fiy0             = _mm_setzero_pd();
 153         fiz0             = _mm_setzero_pd();
 154
 155         /* Load parameters for i particles */
 156         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 157
 158         /* Reset potential sums */
 159         vvdwsum          = _mm_setzero_pd();
 160
 161         /* Start inner kernel loop */
 162         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 163         {
 164
 165             /* Get j neighbor index, and coordinate index */
 166             jnrA             = jjnr[jidx];
 167             jnrB             = jjnr[jidx+1];
 168             j_coord_offsetA  = DIM*jnrA;
 169             j_coord_offsetB  = DIM*jnrB;
 170
 171             /* load j atom coordinates */
 172             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 173                                               &jx0,&jy0,&jz0);
 174
 175             /* Calculate displacement vector */
 176             dx00             = _mm_sub_pd(ix0,jx0);
 177             dy00             = _mm_sub_pd(iy0,jy0);
 178             dz00             = _mm_sub_pd(iz0,jz0);
 179
 180             /* Calculate squared distance and things based on it */
 181             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 182
 183             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 184
 185             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 186
 187             /* Load parameters for j particles */
 188             vdwjidx0A        = 2*vdwtype[jnrA+0];
 189             vdwjidx0B        = 2*vdwtype[jnrB+0];
 190
 191             /**************************
 192              * CALCULATE INTERACTIONS *
 193              **************************/
 194
 195             if (gmx_mm_any_lt(rsq00,rcutoff2))
 196             {
 197
 198             r00              = _mm_mul_pd(rsq00,rinv00);
 199
 200             /* Compute parameters for interactions between i and j atoms */
 201             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 202                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 203             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 204                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 205
 206             /* Analytical LJ-PME */
 207             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 208             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 209             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 210             exponent         = gmx_simd_exp_d(ewcljrsq);
 211             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 212             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 213             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 214             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 215             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 216             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),
 217                                _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));
 218             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 219             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);
 220
 221             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 222
 223             /* Update potential sum for this i atom from the interaction with this j atom. */
 224             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 225             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 226
 227             fscal            = fvdw;
 228
 229             fscal            = _mm_and_pd(fscal,cutoff_mask);
 230
 231             /* Calculate temporary vectorial force */
 232             tx               = _mm_mul_pd(fscal,dx00);
 233             ty               = _mm_mul_pd(fscal,dy00);
 234             tz               = _mm_mul_pd(fscal,dz00);
 235
 236             /* Update vectorial force */
 237             fix0             = _mm_add_pd(fix0,tx);
 238             fiy0             = _mm_add_pd(fiy0,ty);
 239             fiz0             = _mm_add_pd(fiz0,tz);
 240
 241             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 242
 243             }
 244
 245             /* Inner loop uses 61 flops */
 246         }
 247
 248         if(jidx<j_index_end)
 249         {
 250
 251             jnrA             = jjnr[jidx];
 252             j_coord_offsetA  = DIM*jnrA;
 253
 254             /* load j atom coordinates */
 255             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 256                                               &jx0,&jy0,&jz0);
 257
 258             /* Calculate displacement vector */
 259             dx00             = _mm_sub_pd(ix0,jx0);
 260             dy00             = _mm_sub_pd(iy0,jy0);
 261             dz00             = _mm_sub_pd(iz0,jz0);
 262
 263             /* Calculate squared distance and things based on it */
 264             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 265
 266             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 267
 268             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 269
 270             /* Load parameters for j particles */
 271             vdwjidx0A        = 2*vdwtype[jnrA+0];
 272
 273             /**************************
 274              * CALCULATE INTERACTIONS *
 275              **************************/
 276
 277             if (gmx_mm_any_lt(rsq00,rcutoff2))
 278             {
 279
 280             r00              = _mm_mul_pd(rsq00,rinv00);
 281
 282             /* Compute parameters for interactions between i and j atoms */
 283             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 284
 285             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 286
 287             /* Analytical LJ-PME */
 288             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 289             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 290             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 291             exponent         = gmx_simd_exp_d(ewcljrsq);
 292             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 293             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 294             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 295             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 296             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 297             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),
 298                                _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));
 299             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 300             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);
 301
 302             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 303
 304             /* Update potential sum for this i atom from the interaction with this j atom. */
 305             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 306             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 307             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 308
 309             fscal            = fvdw;
 310
 311             fscal            = _mm_and_pd(fscal,cutoff_mask);
 312
 313             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 314
 315             /* Calculate temporary vectorial force */
 316             tx               = _mm_mul_pd(fscal,dx00);
 317             ty               = _mm_mul_pd(fscal,dy00);
 318             tz               = _mm_mul_pd(fscal,dz00);
 319
 320             /* Update vectorial force */
 321             fix0             = _mm_add_pd(fix0,tx);
 322             fiy0             = _mm_add_pd(fiy0,ty);
 323             fiz0             = _mm_add_pd(fiz0,tz);
 324
 325             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 326
 327             }
 328
 329             /* Inner loop uses 61 flops */
 330         }
 331
 332         /* End of innermost loop */
 333
 334         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 335                                               f+i_coord_offset,fshift+i_shift_offset);
 336
 337         ggid                        = gid[iidx];
 338         /* Update potential energies */
 339         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 340
 341         /* Increment number of inner iterations */
 342         inneriter                  += j_index_end - j_index_start;
 343
 344         /* Outer loop uses 7 flops */
 345     }
 346
 347     /* Increment number of outer iterations */
 348     outeriter        += nri;
 349
 350     /* Update outer/inner flops */
 351
 352     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*61);
 353 }
 354 /*
 355  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_F_sse4_1_double
 356  * Electrostatics interaction: None
 357  * VdW interaction:            LJEwald
 358  * Geometry:                   Particle-Particle
 359  * Calculate force/pot:        Force
 360  */
 361 void
 362 nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_F_sse4_1_double
 363                     (t_nblist                    * gmx_restrict       nlist,
 364                      rvec                        * gmx_restrict          xx,
 365                      rvec                        * gmx_restrict          ff,
 366                      t_forcerec                  * gmx_restrict          fr,
 367                      t_mdatoms                   * gmx_restrict     mdatoms,
 368                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 369                      t_nrnb                      * gmx_restrict        nrnb)
 370 {
 371     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 372      * just 0 for non-waters.
 373      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 374      * jnr indices corresponding to data put in the four positions in the SIMD register.
 375      */
 376     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 377     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 378     int              jnrA,jnrB;
 379     int              j_coord_offsetA,j_coord_offsetB;
 380     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 381     real             rcutoff_scalar;
 382     real             *shiftvec,*fshift,*x,*f;
 383     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 384     int              vdwioffset0;
 385     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 386     int              vdwjidx0A,vdwjidx0B;
 387     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 388     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 389     int              nvdwtype;
 390     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 391     int              *vdwtype;
 392     real             *vdwparam;
 393     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 394     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 395     __m128d           c6grid_00;
 396     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 397     real             *vdwgridparam;
 398     __m128d           one_half = _mm_set1_pd(0.5);
 399     __m128d           minus_one = _mm_set1_pd(-1.0);
 400     __m128d          dummy_mask,cutoff_mask;
 401     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 402     __m128d          one     = _mm_set1_pd(1.0);
 403     __m128d          two     = _mm_set1_pd(2.0);
 404     x                = xx[0];
 405     f                = ff[0];
 406
 407     nri              = nlist->nri;
 408     iinr             = nlist->iinr;
 409     jindex           = nlist->jindex;
 410     jjnr             = nlist->jjnr;
 411     shiftidx         = nlist->shift;
 412     gid              = nlist->gid;
 413     shiftvec         = fr->shift_vec[0];
 414     fshift           = fr->fshift[0];
 415     nvdwtype         = fr->ntype;
 416     vdwparam         = fr->nbfp;
 417     vdwtype          = mdatoms->typeA;
 418     vdwgridparam     = fr->ljpme_c6grid;
 419     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 420     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 421     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 422
 423     rcutoff_scalar   = fr->rvdw;
 424     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 425     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 426
 427     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 428     rvdw             = _mm_set1_pd(fr->rvdw);
 429
 430     /* Avoid stupid compiler warnings */
 431     jnrA = jnrB = 0;
 432     j_coord_offsetA = 0;
 433     j_coord_offsetB = 0;
 434
 435     outeriter        = 0;
 436     inneriter        = 0;
 437
 438     /* Start outer loop over neighborlists */
 439     for(iidx=0; iidx<nri; iidx++)
 440     {
 441         /* Load shift vector for this list */
 442         i_shift_offset   = DIM*shiftidx[iidx];
 443
 444         /* Load limits for loop over neighbors */
 445         j_index_start    = jindex[iidx];
 446         j_index_end      = jindex[iidx+1];
 447
 448         /* Get outer coordinate index */
 449         inr              = iinr[iidx];
 450         i_coord_offset   = DIM*inr;
 451
 452         /* Load i particle coords and add shift vector */
 453         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 454
 455         fix0             = _mm_setzero_pd();
 456         fiy0             = _mm_setzero_pd();
 457         fiz0             = _mm_setzero_pd();
 458
 459         /* Load parameters for i particles */
 460         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 461
 462         /* Start inner kernel loop */
 463         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 464         {
 465
 466             /* Get j neighbor index, and coordinate index */
 467             jnrA             = jjnr[jidx];
 468             jnrB             = jjnr[jidx+1];
 469             j_coord_offsetA  = DIM*jnrA;
 470             j_coord_offsetB  = DIM*jnrB;
 471
 472             /* load j atom coordinates */
 473             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 474                                               &jx0,&jy0,&jz0);
 475
 476             /* Calculate displacement vector */
 477             dx00             = _mm_sub_pd(ix0,jx0);
 478             dy00             = _mm_sub_pd(iy0,jy0);
 479             dz00             = _mm_sub_pd(iz0,jz0);
 480
 481             /* Calculate squared distance and things based on it */
 482             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 483
 484             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 485
 486             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 487
 488             /* Load parameters for j particles */
 489             vdwjidx0A        = 2*vdwtype[jnrA+0];
 490             vdwjidx0B        = 2*vdwtype[jnrB+0];
 491
 492             /**************************
 493              * CALCULATE INTERACTIONS *
 494              **************************/
 495
 496             if (gmx_mm_any_lt(rsq00,rcutoff2))
 497             {
 498
 499             r00              = _mm_mul_pd(rsq00,rinv00);
 500
 501             /* Compute parameters for interactions between i and j atoms */
 502             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 503                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 504             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 505                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 506
 507             /* Analytical LJ-PME */
 508             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 509             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 510             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 511             exponent         = gmx_simd_exp_d(ewcljrsq);
 512             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 513             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 514             /* f6A = 6 * C6grid * (1 - poly) */
 515             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 516             /* f6B = C6grid * exponent * beta^6 */
 517             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 518             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 519             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);
 520
 521             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 522
 523             fscal            = fvdw;
 524
 525             fscal            = _mm_and_pd(fscal,cutoff_mask);
 526
 527             /* Calculate temporary vectorial force */
 528             tx               = _mm_mul_pd(fscal,dx00);
 529             ty               = _mm_mul_pd(fscal,dy00);
 530             tz               = _mm_mul_pd(fscal,dz00);
 531
 532             /* Update vectorial force */
 533             fix0             = _mm_add_pd(fix0,tx);
 534             fiy0             = _mm_add_pd(fiy0,ty);
 535             fiz0             = _mm_add_pd(fiz0,tz);
 536
 537             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 538
 539             }
 540
 541             /* Inner loop uses 49 flops */
 542         }
 543
 544         if(jidx<j_index_end)
 545         {
 546
 547             jnrA             = jjnr[jidx];
 548             j_coord_offsetA  = DIM*jnrA;
 549
 550             /* load j atom coordinates */
 551             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 552                                               &jx0,&jy0,&jz0);
 553
 554             /* Calculate displacement vector */
 555             dx00             = _mm_sub_pd(ix0,jx0);
 556             dy00             = _mm_sub_pd(iy0,jy0);
 557             dz00             = _mm_sub_pd(iz0,jz0);
 558
 559             /* Calculate squared distance and things based on it */
 560             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 561
 562             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 563
 564             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 565
 566             /* Load parameters for j particles */
 567             vdwjidx0A        = 2*vdwtype[jnrA+0];
 568
 569             /**************************
 570              * CALCULATE INTERACTIONS *
 571              **************************/
 572
 573             if (gmx_mm_any_lt(rsq00,rcutoff2))
 574             {
 575
 576             r00              = _mm_mul_pd(rsq00,rinv00);
 577
 578             /* Compute parameters for interactions between i and j atoms */
 579             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 580
 581             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 582
 583             /* Analytical LJ-PME */
 584             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 585             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 586             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 587             exponent         = gmx_simd_exp_d(ewcljrsq);
 588             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 589             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 590             /* f6A = 6 * C6grid * (1 - poly) */
 591             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 592             /* f6B = C6grid * exponent * beta^6 */
 593             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 594             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 595             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);
 596
 597             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 598
 599             fscal            = fvdw;
 600
 601             fscal            = _mm_and_pd(fscal,cutoff_mask);
 602
 603             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 604
 605             /* Calculate temporary vectorial force */
 606             tx               = _mm_mul_pd(fscal,dx00);
 607             ty               = _mm_mul_pd(fscal,dy00);
 608             tz               = _mm_mul_pd(fscal,dz00);
 609
 610             /* Update vectorial force */
 611             fix0             = _mm_add_pd(fix0,tx);
 612             fiy0             = _mm_add_pd(fiy0,ty);
 613             fiz0             = _mm_add_pd(fiz0,tz);
 614
 615             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 616
 617             }
 618
 619             /* Inner loop uses 49 flops */
 620         }
 621
 622         /* End of innermost loop */
 623
 624         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 625                                               f+i_coord_offset,fshift+i_shift_offset);
 626
 627         /* Increment number of inner iterations */
 628         inneriter                  += j_index_end - j_index_start;
 629
 630         /* Outer loop uses 6 flops */
 631     }
 632
 633     /* Increment number of outer iterations */
 634     outeriter        += nri;
 635
 636     /* Update outer/inner flops */
 637
 638     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*49);
 639 }