src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecNone_VdwLJEw_GeomP1P1_sse4_1_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse4_1_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEw_GeomP1P1_VF_sse4_1_double
  51  * Electrostatics interaction: None
  52  * VdW interaction:            LJEwald
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecNone_VdwLJEw_GeomP1P1_VF_sse4_1_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwjidx0A,vdwjidx0B;
  82     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  83     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  84     int              nvdwtype;
  85     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  86     int              *vdwtype;
  87     real             *vdwparam;
  88     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  89     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  90     __m128d           c6grid_00;
  91     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
  92     real             *vdwgridparam;
  93     __m128d           one_half = _mm_set1_pd(0.5);
  94     __m128d           minus_one = _mm_set1_pd(-1.0);
  95     __m128d          dummy_mask,cutoff_mask;
  96     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  97     __m128d          one     = _mm_set1_pd(1.0);
  98     __m128d          two     = _mm_set1_pd(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     nvdwtype         = fr->ntype;
 111     vdwparam         = fr->nbfp;
 112     vdwtype          = mdatoms->typeA;
 113     vdwgridparam     = fr->ljpme_c6grid;
 114     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 115     ewclj            = _mm_set1_pd(fr->ic->ewaldcoeff_lj);
 116     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 117
 118     /* Avoid stupid compiler warnings */
 119     jnrA = jnrB = 0;
 120     j_coord_offsetA = 0;
 121     j_coord_offsetB = 0;
 122
 123     outeriter        = 0;
 124     inneriter        = 0;
 125
 126     /* Start outer loop over neighborlists */
 127     for(iidx=0; iidx<nri; iidx++)
 128     {
 129         /* Load shift vector for this list */
 130         i_shift_offset   = DIM*shiftidx[iidx];
 131
 132         /* Load limits for loop over neighbors */
 133         j_index_start    = jindex[iidx];
 134         j_index_end      = jindex[iidx+1];
 135
 136         /* Get outer coordinate index */
 137         inr              = iinr[iidx];
 138         i_coord_offset   = DIM*inr;
 139
 140         /* Load i particle coords and add shift vector */
 141         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 142
 143         fix0             = _mm_setzero_pd();
 144         fiy0             = _mm_setzero_pd();
 145         fiz0             = _mm_setzero_pd();
 146
 147         /* Load parameters for i particles */
 148         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 149
 150         /* Reset potential sums */
 151         vvdwsum          = _mm_setzero_pd();
 152
 153         /* Start inner kernel loop */
 154         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 155         {
 156
 157             /* Get j neighbor index, and coordinate index */
 158             jnrA             = jjnr[jidx];
 159             jnrB             = jjnr[jidx+1];
 160             j_coord_offsetA  = DIM*jnrA;
 161             j_coord_offsetB  = DIM*jnrB;
 162
 163             /* load j atom coordinates */
 164             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 165                                               &jx0,&jy0,&jz0);
 166
 167             /* Calculate displacement vector */
 168             dx00             = _mm_sub_pd(ix0,jx0);
 169             dy00             = _mm_sub_pd(iy0,jy0);
 170             dz00             = _mm_sub_pd(iz0,jz0);
 171
 172             /* Calculate squared distance and things based on it */
 173             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 174
 175             rinv00           = sse41_invsqrt_d(rsq00);
 176
 177             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 178
 179             /* Load parameters for j particles */
 180             vdwjidx0A        = 2*vdwtype[jnrA+0];
 181             vdwjidx0B        = 2*vdwtype[jnrB+0];
 182
 183             /**************************
 184              * CALCULATE INTERACTIONS *
 185              **************************/
 186
 187             r00              = _mm_mul_pd(rsq00,rinv00);
 188
 189             /* Compute parameters for interactions between i and j atoms */
 190             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 191                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 192             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 193                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 194
 195             /* Analytical LJ-PME */
 196             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 197             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 198             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 199             exponent         = sse41_exp_d(ewcljrsq);
 200             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 201             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 202             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 203             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 204             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 205             vvdw             = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
 206             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 207             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);
 208
 209             /* Update potential sum for this i atom from the interaction with this j atom. */
 210             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 211
 212             fscal            = fvdw;
 213
 214             /* Calculate temporary vectorial force */
 215             tx               = _mm_mul_pd(fscal,dx00);
 216             ty               = _mm_mul_pd(fscal,dy00);
 217             tz               = _mm_mul_pd(fscal,dz00);
 218
 219             /* Update vectorial force */
 220             fix0             = _mm_add_pd(fix0,tx);
 221             fiy0             = _mm_add_pd(fiy0,ty);
 222             fiz0             = _mm_add_pd(fiz0,tz);
 223
 224             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 225
 226             /* Inner loop uses 51 flops */
 227         }
 228
 229         if(jidx<j_index_end)
 230         {
 231
 232             jnrA             = jjnr[jidx];
 233             j_coord_offsetA  = DIM*jnrA;
 234
 235             /* load j atom coordinates */
 236             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 237                                               &jx0,&jy0,&jz0);
 238
 239             /* Calculate displacement vector */
 240             dx00             = _mm_sub_pd(ix0,jx0);
 241             dy00             = _mm_sub_pd(iy0,jy0);
 242             dz00             = _mm_sub_pd(iz0,jz0);
 243
 244             /* Calculate squared distance and things based on it */
 245             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 246
 247             rinv00           = sse41_invsqrt_d(rsq00);
 248
 249             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 250
 251             /* Load parameters for j particles */
 252             vdwjidx0A        = 2*vdwtype[jnrA+0];
 253
 254             /**************************
 255              * CALCULATE INTERACTIONS *
 256              **************************/
 257
 258             r00              = _mm_mul_pd(rsq00,rinv00);
 259
 260             /* Compute parameters for interactions between i and j atoms */
 261             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 262
 263             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 264
 265             /* Analytical LJ-PME */
 266             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 267             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 268             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 269             exponent         = sse41_exp_d(ewcljrsq);
 270             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 271             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 272             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 273             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 274             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 275             vvdw             = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
 276             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 277             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);
 278
 279             /* Update potential sum for this i atom from the interaction with this j atom. */
 280             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 281             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 282
 283             fscal            = fvdw;
 284
 285             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 286
 287             /* Calculate temporary vectorial force */
 288             tx               = _mm_mul_pd(fscal,dx00);
 289             ty               = _mm_mul_pd(fscal,dy00);
 290             tz               = _mm_mul_pd(fscal,dz00);
 291
 292             /* Update vectorial force */
 293             fix0             = _mm_add_pd(fix0,tx);
 294             fiy0             = _mm_add_pd(fiy0,ty);
 295             fiz0             = _mm_add_pd(fiz0,tz);
 296
 297             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 298
 299             /* Inner loop uses 51 flops */
 300         }
 301
 302         /* End of innermost loop */
 303
 304         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 305                                               f+i_coord_offset,fshift+i_shift_offset);
 306
 307         ggid                        = gid[iidx];
 308         /* Update potential energies */
 309         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 310
 311         /* Increment number of inner iterations */
 312         inneriter                  += j_index_end - j_index_start;
 313
 314         /* Outer loop uses 7 flops */
 315     }
 316
 317     /* Increment number of outer iterations */
 318     outeriter        += nri;
 319
 320     /* Update outer/inner flops */
 321
 322     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*51);
 323 }
 324 /*
 325  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEw_GeomP1P1_F_sse4_1_double
 326  * Electrostatics interaction: None
 327  * VdW interaction:            LJEwald
 328  * Geometry:                   Particle-Particle
 329  * Calculate force/pot:        Force
 330  */
 331 void
 332 nb_kernel_ElecNone_VdwLJEw_GeomP1P1_F_sse4_1_double
 333                     (t_nblist                    * gmx_restrict       nlist,
 334                      rvec                        * gmx_restrict          xx,
 335                      rvec                        * gmx_restrict          ff,
 336                      struct t_forcerec           * gmx_restrict          fr,
 337                      t_mdatoms                   * gmx_restrict     mdatoms,
 338                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 339                      t_nrnb                      * gmx_restrict        nrnb)
 340 {
 341     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 342      * just 0 for non-waters.
 343      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 344      * jnr indices corresponding to data put in the four positions in the SIMD register.
 345      */
 346     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 347     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 348     int              jnrA,jnrB;
 349     int              j_coord_offsetA,j_coord_offsetB;
 350     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 351     real             rcutoff_scalar;
 352     real             *shiftvec,*fshift,*x,*f;
 353     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 354     int              vdwioffset0;
 355     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 356     int              vdwjidx0A,vdwjidx0B;
 357     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 358     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 359     int              nvdwtype;
 360     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 361     int              *vdwtype;
 362     real             *vdwparam;
 363     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 364     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 365     __m128d           c6grid_00;
 366     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 367     real             *vdwgridparam;
 368     __m128d           one_half = _mm_set1_pd(0.5);
 369     __m128d           minus_one = _mm_set1_pd(-1.0);
 370     __m128d          dummy_mask,cutoff_mask;
 371     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 372     __m128d          one     = _mm_set1_pd(1.0);
 373     __m128d          two     = _mm_set1_pd(2.0);
 374     x                = xx[0];
 375     f                = ff[0];
 376
 377     nri              = nlist->nri;
 378     iinr             = nlist->iinr;
 379     jindex           = nlist->jindex;
 380     jjnr             = nlist->jjnr;
 381     shiftidx         = nlist->shift;
 382     gid              = nlist->gid;
 383     shiftvec         = fr->shift_vec[0];
 384     fshift           = fr->fshift[0];
 385     nvdwtype         = fr->ntype;
 386     vdwparam         = fr->nbfp;
 387     vdwtype          = mdatoms->typeA;
 388     vdwgridparam     = fr->ljpme_c6grid;
 389     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 390     ewclj            = _mm_set1_pd(fr->ic->ewaldcoeff_lj);
 391     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 392
 393     /* Avoid stupid compiler warnings */
 394     jnrA = jnrB = 0;
 395     j_coord_offsetA = 0;
 396     j_coord_offsetB = 0;
 397
 398     outeriter        = 0;
 399     inneriter        = 0;
 400
 401     /* Start outer loop over neighborlists */
 402     for(iidx=0; iidx<nri; iidx++)
 403     {
 404         /* Load shift vector for this list */
 405         i_shift_offset   = DIM*shiftidx[iidx];
 406
 407         /* Load limits for loop over neighbors */
 408         j_index_start    = jindex[iidx];
 409         j_index_end      = jindex[iidx+1];
 410
 411         /* Get outer coordinate index */
 412         inr              = iinr[iidx];
 413         i_coord_offset   = DIM*inr;
 414
 415         /* Load i particle coords and add shift vector */
 416         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 417
 418         fix0             = _mm_setzero_pd();
 419         fiy0             = _mm_setzero_pd();
 420         fiz0             = _mm_setzero_pd();
 421
 422         /* Load parameters for i particles */
 423         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 424
 425         /* Start inner kernel loop */
 426         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 427         {
 428
 429             /* Get j neighbor index, and coordinate index */
 430             jnrA             = jjnr[jidx];
 431             jnrB             = jjnr[jidx+1];
 432             j_coord_offsetA  = DIM*jnrA;
 433             j_coord_offsetB  = DIM*jnrB;
 434
 435             /* load j atom coordinates */
 436             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 437                                               &jx0,&jy0,&jz0);
 438
 439             /* Calculate displacement vector */
 440             dx00             = _mm_sub_pd(ix0,jx0);
 441             dy00             = _mm_sub_pd(iy0,jy0);
 442             dz00             = _mm_sub_pd(iz0,jz0);
 443
 444             /* Calculate squared distance and things based on it */
 445             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 446
 447             rinv00           = sse41_invsqrt_d(rsq00);
 448
 449             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 450
 451             /* Load parameters for j particles */
 452             vdwjidx0A        = 2*vdwtype[jnrA+0];
 453             vdwjidx0B        = 2*vdwtype[jnrB+0];
 454
 455             /**************************
 456              * CALCULATE INTERACTIONS *
 457              **************************/
 458
 459             r00              = _mm_mul_pd(rsq00,rinv00);
 460
 461             /* Compute parameters for interactions between i and j atoms */
 462             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 463                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 464             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 465                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 466
 467             /* Analytical LJ-PME */
 468             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 469             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 470             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 471             exponent         = sse41_exp_d(ewcljrsq);
 472             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 473             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 474             /* f6A = 6 * C6grid * (1 - poly) */
 475             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 476             /* f6B = C6grid * exponent * beta^6 */
 477             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 478             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 479             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);
 480
 481             fscal            = fvdw;
 482
 483             /* Calculate temporary vectorial force */
 484             tx               = _mm_mul_pd(fscal,dx00);
 485             ty               = _mm_mul_pd(fscal,dy00);
 486             tz               = _mm_mul_pd(fscal,dz00);
 487
 488             /* Update vectorial force */
 489             fix0             = _mm_add_pd(fix0,tx);
 490             fiy0             = _mm_add_pd(fiy0,ty);
 491             fiz0             = _mm_add_pd(fiz0,tz);
 492
 493             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 494
 495             /* Inner loop uses 46 flops */
 496         }
 497
 498         if(jidx<j_index_end)
 499         {
 500
 501             jnrA             = jjnr[jidx];
 502             j_coord_offsetA  = DIM*jnrA;
 503
 504             /* load j atom coordinates */
 505             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 506                                               &jx0,&jy0,&jz0);
 507
 508             /* Calculate displacement vector */
 509             dx00             = _mm_sub_pd(ix0,jx0);
 510             dy00             = _mm_sub_pd(iy0,jy0);
 511             dz00             = _mm_sub_pd(iz0,jz0);
 512
 513             /* Calculate squared distance and things based on it */
 514             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 515
 516             rinv00           = sse41_invsqrt_d(rsq00);
 517
 518             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 519
 520             /* Load parameters for j particles */
 521             vdwjidx0A        = 2*vdwtype[jnrA+0];
 522
 523             /**************************
 524              * CALCULATE INTERACTIONS *
 525              **************************/
 526
 527             r00              = _mm_mul_pd(rsq00,rinv00);
 528
 529             /* Compute parameters for interactions between i and j atoms */
 530             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 531
 532             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 533
 534             /* Analytical LJ-PME */
 535             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 536             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 537             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 538             exponent         = sse41_exp_d(ewcljrsq);
 539             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 540             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 541             /* f6A = 6 * C6grid * (1 - poly) */
 542             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 543             /* f6B = C6grid * exponent * beta^6 */
 544             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 545             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 546             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);
 547
 548             fscal            = fvdw;
 549
 550             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 551
 552             /* Calculate temporary vectorial force */
 553             tx               = _mm_mul_pd(fscal,dx00);
 554             ty               = _mm_mul_pd(fscal,dy00);
 555             tz               = _mm_mul_pd(fscal,dz00);
 556
 557             /* Update vectorial force */
 558             fix0             = _mm_add_pd(fix0,tx);
 559             fiy0             = _mm_add_pd(fiy0,ty);
 560             fiz0             = _mm_add_pd(fiz0,tz);
 561
 562             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 563
 564             /* Inner loop uses 46 flops */
 565         }
 566
 567         /* End of innermost loop */
 568
 569         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 570                                               f+i_coord_offset,fshift+i_shift_offset);
 571
 572         /* Increment number of inner iterations */
 573         inneriter                  += j_index_end - j_index_start;
 574
 575         /* Outer loop uses 6 flops */
 576     }
 577
 578     /* Increment number of outer iterations */
 579     outeriter        += nri;
 580
 581     /* Update outer/inner flops */
 582
 583     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*46);
 584 }