src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecNone_VdwLJEw_GeomP1P1_sse2_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 sse2_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_sse2_double.h"
  50 #include "kernelutil_x86_sse2_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEw_GeomP1P1_VF_sse2_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_VdwLJEw_GeomP1P1_VF_sse2_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     /* Avoid stupid compiler warnings */
 122     jnrA = jnrB = 0;
 123     j_coord_offsetA = 0;
 124     j_coord_offsetB = 0;
 125
 126     outeriter        = 0;
 127     inneriter        = 0;
 128
 129     /* Start outer loop over neighborlists */
 130     for(iidx=0; iidx<nri; iidx++)
 131     {
 132         /* Load shift vector for this list */
 133         i_shift_offset   = DIM*shiftidx[iidx];
 134
 135         /* Load limits for loop over neighbors */
 136         j_index_start    = jindex[iidx];
 137         j_index_end      = jindex[iidx+1];
 138
 139         /* Get outer coordinate index */
 140         inr              = iinr[iidx];
 141         i_coord_offset   = DIM*inr;
 142
 143         /* Load i particle coords and add shift vector */
 144         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 145
 146         fix0             = _mm_setzero_pd();
 147         fiy0             = _mm_setzero_pd();
 148         fiz0             = _mm_setzero_pd();
 149
 150         /* Load parameters for i particles */
 151         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 152
 153         /* Reset potential sums */
 154         vvdwsum          = _mm_setzero_pd();
 155
 156         /* Start inner kernel loop */
 157         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 158         {
 159
 160             /* Get j neighbor index, and coordinate index */
 161             jnrA             = jjnr[jidx];
 162             jnrB             = jjnr[jidx+1];
 163             j_coord_offsetA  = DIM*jnrA;
 164             j_coord_offsetB  = DIM*jnrB;
 165
 166             /* load j atom coordinates */
 167             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 168                                               &jx0,&jy0,&jz0);
 169
 170             /* Calculate displacement vector */
 171             dx00             = _mm_sub_pd(ix0,jx0);
 172             dy00             = _mm_sub_pd(iy0,jy0);
 173             dz00             = _mm_sub_pd(iz0,jz0);
 174
 175             /* Calculate squared distance and things based on it */
 176             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 177
 178             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 179
 180             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 181
 182             /* Load parameters for j particles */
 183             vdwjidx0A        = 2*vdwtype[jnrA+0];
 184             vdwjidx0B        = 2*vdwtype[jnrB+0];
 185
 186             /**************************
 187              * CALCULATE INTERACTIONS *
 188              **************************/
 189
 190             r00              = _mm_mul_pd(rsq00,rinv00);
 191
 192             /* Compute parameters for interactions between i and j atoms */
 193             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 194                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 195
 196             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 197                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 198
 199             /* Analytical LJ-PME */
 200             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 201             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 202             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 203             exponent         = gmx_simd_exp_d(ewcljrsq);
 204             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 205             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 206             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 207             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 208             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 209             vvdw             = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
 210             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 211             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);
 212
 213             /* Update potential sum for this i atom from the interaction with this j atom. */
 214             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 215
 216             fscal            = fvdw;
 217
 218             /* Calculate temporary vectorial force */
 219             tx               = _mm_mul_pd(fscal,dx00);
 220             ty               = _mm_mul_pd(fscal,dy00);
 221             tz               = _mm_mul_pd(fscal,dz00);
 222
 223             /* Update vectorial force */
 224             fix0             = _mm_add_pd(fix0,tx);
 225             fiy0             = _mm_add_pd(fiy0,ty);
 226             fiz0             = _mm_add_pd(fiz0,tz);
 227
 228             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 229
 230             /* Inner loop uses 51 flops */
 231         }
 232
 233         if(jidx<j_index_end)
 234         {
 235
 236             jnrA             = jjnr[jidx];
 237             j_coord_offsetA  = DIM*jnrA;
 238
 239             /* load j atom coordinates */
 240             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 241                                               &jx0,&jy0,&jz0);
 242
 243             /* Calculate displacement vector */
 244             dx00             = _mm_sub_pd(ix0,jx0);
 245             dy00             = _mm_sub_pd(iy0,jy0);
 246             dz00             = _mm_sub_pd(iz0,jz0);
 247
 248             /* Calculate squared distance and things based on it */
 249             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 250
 251             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 252
 253             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 254
 255             /* Load parameters for j particles */
 256             vdwjidx0A        = 2*vdwtype[jnrA+0];
 257
 258             /**************************
 259              * CALCULATE INTERACTIONS *
 260              **************************/
 261
 262             r00              = _mm_mul_pd(rsq00,rinv00);
 263
 264             /* Compute parameters for interactions between i and j atoms */
 265             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 266
 267             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 268
 269             /* Analytical LJ-PME */
 270             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 271             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 272             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 273             exponent         = gmx_simd_exp_d(ewcljrsq);
 274             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 275             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 276             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 277             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 278             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 279             vvdw             = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
 280             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 281             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);
 282
 283             /* Update potential sum for this i atom from the interaction with this j atom. */
 284             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 285             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 286
 287             fscal            = fvdw;
 288
 289             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 290
 291             /* Calculate temporary vectorial force */
 292             tx               = _mm_mul_pd(fscal,dx00);
 293             ty               = _mm_mul_pd(fscal,dy00);
 294             tz               = _mm_mul_pd(fscal,dz00);
 295
 296             /* Update vectorial force */
 297             fix0             = _mm_add_pd(fix0,tx);
 298             fiy0             = _mm_add_pd(fiy0,ty);
 299             fiz0             = _mm_add_pd(fiz0,tz);
 300
 301             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 302
 303             /* Inner loop uses 51 flops */
 304         }
 305
 306         /* End of innermost loop */
 307
 308         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 309                                               f+i_coord_offset,fshift+i_shift_offset);
 310
 311         ggid                        = gid[iidx];
 312         /* Update potential energies */
 313         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 314
 315         /* Increment number of inner iterations */
 316         inneriter                  += j_index_end - j_index_start;
 317
 318         /* Outer loop uses 7 flops */
 319     }
 320
 321     /* Increment number of outer iterations */
 322     outeriter        += nri;
 323
 324     /* Update outer/inner flops */
 325
 326     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*51);
 327 }
 328 /*
 329  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEw_GeomP1P1_F_sse2_double
 330  * Electrostatics interaction: None
 331  * VdW interaction:            LJEwald
 332  * Geometry:                   Particle-Particle
 333  * Calculate force/pot:        Force
 334  */
 335 void
 336 nb_kernel_ElecNone_VdwLJEw_GeomP1P1_F_sse2_double
 337                     (t_nblist                    * gmx_restrict       nlist,
 338                      rvec                        * gmx_restrict          xx,
 339                      rvec                        * gmx_restrict          ff,
 340                      t_forcerec                  * gmx_restrict          fr,
 341                      t_mdatoms                   * gmx_restrict     mdatoms,
 342                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 343                      t_nrnb                      * gmx_restrict        nrnb)
 344 {
 345     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 346      * just 0 for non-waters.
 347      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 348      * jnr indices corresponding to data put in the four positions in the SIMD register.
 349      */
 350     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 351     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 352     int              jnrA,jnrB;
 353     int              j_coord_offsetA,j_coord_offsetB;
 354     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 355     real             rcutoff_scalar;
 356     real             *shiftvec,*fshift,*x,*f;
 357     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 358     int              vdwioffset0;
 359     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 360     int              vdwjidx0A,vdwjidx0B;
 361     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 362     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 363     int              nvdwtype;
 364     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 365     int              *vdwtype;
 366     real             *vdwparam;
 367     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 368     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 369     __m128d           c6grid_00;
 370     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 371     real             *vdwgridparam;
 372     __m128d           one_half = _mm_set1_pd(0.5);
 373     __m128d           minus_one = _mm_set1_pd(-1.0);
 374     __m128d          dummy_mask,cutoff_mask;
 375     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 376     __m128d          one     = _mm_set1_pd(1.0);
 377     __m128d          two     = _mm_set1_pd(2.0);
 378     x                = xx[0];
 379     f                = ff[0];
 380
 381     nri              = nlist->nri;
 382     iinr             = nlist->iinr;
 383     jindex           = nlist->jindex;
 384     jjnr             = nlist->jjnr;
 385     shiftidx         = nlist->shift;
 386     gid              = nlist->gid;
 387     shiftvec         = fr->shift_vec[0];
 388     fshift           = fr->fshift[0];
 389     nvdwtype         = fr->ntype;
 390     vdwparam         = fr->nbfp;
 391     vdwtype          = mdatoms->typeA;
 392     vdwgridparam     = fr->ljpme_c6grid;
 393     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 394     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 395     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 396
 397     /* Avoid stupid compiler warnings */
 398     jnrA = jnrB = 0;
 399     j_coord_offsetA = 0;
 400     j_coord_offsetB = 0;
 401
 402     outeriter        = 0;
 403     inneriter        = 0;
 404
 405     /* Start outer loop over neighborlists */
 406     for(iidx=0; iidx<nri; iidx++)
 407     {
 408         /* Load shift vector for this list */
 409         i_shift_offset   = DIM*shiftidx[iidx];
 410
 411         /* Load limits for loop over neighbors */
 412         j_index_start    = jindex[iidx];
 413         j_index_end      = jindex[iidx+1];
 414
 415         /* Get outer coordinate index */
 416         inr              = iinr[iidx];
 417         i_coord_offset   = DIM*inr;
 418
 419         /* Load i particle coords and add shift vector */
 420         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 421
 422         fix0             = _mm_setzero_pd();
 423         fiy0             = _mm_setzero_pd();
 424         fiz0             = _mm_setzero_pd();
 425
 426         /* Load parameters for i particles */
 427         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 428
 429         /* Start inner kernel loop */
 430         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 431         {
 432
 433             /* Get j neighbor index, and coordinate index */
 434             jnrA             = jjnr[jidx];
 435             jnrB             = jjnr[jidx+1];
 436             j_coord_offsetA  = DIM*jnrA;
 437             j_coord_offsetB  = DIM*jnrB;
 438
 439             /* load j atom coordinates */
 440             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 441                                               &jx0,&jy0,&jz0);
 442
 443             /* Calculate displacement vector */
 444             dx00             = _mm_sub_pd(ix0,jx0);
 445             dy00             = _mm_sub_pd(iy0,jy0);
 446             dz00             = _mm_sub_pd(iz0,jz0);
 447
 448             /* Calculate squared distance and things based on it */
 449             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 450
 451             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 452
 453             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 454
 455             /* Load parameters for j particles */
 456             vdwjidx0A        = 2*vdwtype[jnrA+0];
 457             vdwjidx0B        = 2*vdwtype[jnrB+0];
 458
 459             /**************************
 460              * CALCULATE INTERACTIONS *
 461              **************************/
 462
 463             r00              = _mm_mul_pd(rsq00,rinv00);
 464
 465             /* Compute parameters for interactions between i and j atoms */
 466             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 467                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 468
 469             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 470                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 471
 472             /* Analytical LJ-PME */
 473             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 474             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 475             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 476             exponent         = gmx_simd_exp_d(ewcljrsq);
 477             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 478             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 479             /* f6A = 6 * C6grid * (1 - poly) */
 480             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 481             /* f6B = C6grid * exponent * beta^6 */
 482             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 483             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 484             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);
 485
 486             fscal            = fvdw;
 487
 488             /* Calculate temporary vectorial force */
 489             tx               = _mm_mul_pd(fscal,dx00);
 490             ty               = _mm_mul_pd(fscal,dy00);
 491             tz               = _mm_mul_pd(fscal,dz00);
 492
 493             /* Update vectorial force */
 494             fix0             = _mm_add_pd(fix0,tx);
 495             fiy0             = _mm_add_pd(fiy0,ty);
 496             fiz0             = _mm_add_pd(fiz0,tz);
 497
 498             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 499
 500             /* Inner loop uses 46 flops */
 501         }
 502
 503         if(jidx<j_index_end)
 504         {
 505
 506             jnrA             = jjnr[jidx];
 507             j_coord_offsetA  = DIM*jnrA;
 508
 509             /* load j atom coordinates */
 510             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 511                                               &jx0,&jy0,&jz0);
 512
 513             /* Calculate displacement vector */
 514             dx00             = _mm_sub_pd(ix0,jx0);
 515             dy00             = _mm_sub_pd(iy0,jy0);
 516             dz00             = _mm_sub_pd(iz0,jz0);
 517
 518             /* Calculate squared distance and things based on it */
 519             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 520
 521             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 522
 523             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 524
 525             /* Load parameters for j particles */
 526             vdwjidx0A        = 2*vdwtype[jnrA+0];
 527
 528             /**************************
 529              * CALCULATE INTERACTIONS *
 530              **************************/
 531
 532             r00              = _mm_mul_pd(rsq00,rinv00);
 533
 534             /* Compute parameters for interactions between i and j atoms */
 535             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 536
 537             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 538
 539             /* Analytical LJ-PME */
 540             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 541             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 542             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 543             exponent         = gmx_simd_exp_d(ewcljrsq);
 544             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 545             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 546             /* f6A = 6 * C6grid * (1 - poly) */
 547             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 548             /* f6B = C6grid * exponent * beta^6 */
 549             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 550             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 551             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);
 552
 553             fscal            = fvdw;
 554
 555             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 556
 557             /* Calculate temporary vectorial force */
 558             tx               = _mm_mul_pd(fscal,dx00);
 559             ty               = _mm_mul_pd(fscal,dy00);
 560             tz               = _mm_mul_pd(fscal,dz00);
 561
 562             /* Update vectorial force */
 563             fix0             = _mm_add_pd(fix0,tx);
 564             fiy0             = _mm_add_pd(fiy0,ty);
 565             fiz0             = _mm_add_pd(fiz0,tz);
 566
 567             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 568
 569             /* Inner loop uses 46 flops */
 570         }
 571
 572         /* End of innermost loop */
 573
 574         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 575                                               f+i_coord_offset,fshift+i_shift_offset);
 576
 577         /* Increment number of inner iterations */
 578         inneriter                  += j_index_end - j_index_start;
 579
 580         /* Outer loop uses 6 flops */
 581     }
 582
 583     /* Increment number of outer iterations */
 584     outeriter        += nri;
 585
 586     /* Update outer/inner flops */
 587
 588     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*46);
 589 }