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 #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_sse2_double.h"
  48 #include "kernelutil_x86_sse2_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEw_GeomP1P1_VF_sse2_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_VdwLJEw_GeomP1P1_VF_sse2_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     /* Avoid stupid compiler warnings */
 120     jnrA = jnrB = 0;
 121     j_coord_offsetA = 0;
 122     j_coord_offsetB = 0;
 123
 124     outeriter        = 0;
 125     inneriter        = 0;
 126
 127     /* Start outer loop over neighborlists */
 128     for(iidx=0; iidx<nri; iidx++)
 129     {
 130         /* Load shift vector for this list */
 131         i_shift_offset   = DIM*shiftidx[iidx];
 132
 133         /* Load limits for loop over neighbors */
 134         j_index_start    = jindex[iidx];
 135         j_index_end      = jindex[iidx+1];
 136
 137         /* Get outer coordinate index */
 138         inr              = iinr[iidx];
 139         i_coord_offset   = DIM*inr;
 140
 141         /* Load i particle coords and add shift vector */
 142         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 143
 144         fix0             = _mm_setzero_pd();
 145         fiy0             = _mm_setzero_pd();
 146         fiz0             = _mm_setzero_pd();
 147
 148         /* Load parameters for i particles */
 149         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 150
 151         /* Reset potential sums */
 152         vvdwsum          = _mm_setzero_pd();
 153
 154         /* Start inner kernel loop */
 155         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 156         {
 157
 158             /* Get j neighbor index, and coordinate index */
 159             jnrA             = jjnr[jidx];
 160             jnrB             = jjnr[jidx+1];
 161             j_coord_offsetA  = DIM*jnrA;
 162             j_coord_offsetB  = DIM*jnrB;
 163
 164             /* load j atom coordinates */
 165             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 166                                               &jx0,&jy0,&jz0);
 167
 168             /* Calculate displacement vector */
 169             dx00             = _mm_sub_pd(ix0,jx0);
 170             dy00             = _mm_sub_pd(iy0,jy0);
 171             dz00             = _mm_sub_pd(iz0,jz0);
 172
 173             /* Calculate squared distance and things based on it */
 174             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 175
 176             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 177
 178             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 179
 180             /* Load parameters for j particles */
 181             vdwjidx0A        = 2*vdwtype[jnrA+0];
 182             vdwjidx0B        = 2*vdwtype[jnrB+0];
 183
 184             /**************************
 185              * CALCULATE INTERACTIONS *
 186              **************************/
 187
 188             r00              = _mm_mul_pd(rsq00,rinv00);
 189
 190             /* Compute parameters for interactions between i and j atoms */
 191             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 192                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 193
 194             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 195                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 196
 197             /* Analytical LJ-PME */
 198             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 199             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 200             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 201             exponent         = gmx_simd_exp_d(ewcljrsq);
 202             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 203             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 204             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 205             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 206             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 207             vvdw             = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
 208             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 209             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);
 210
 211             /* Update potential sum for this i atom from the interaction with this j atom. */
 212             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 213
 214             fscal            = fvdw;
 215
 216             /* Calculate temporary vectorial force */
 217             tx               = _mm_mul_pd(fscal,dx00);
 218             ty               = _mm_mul_pd(fscal,dy00);
 219             tz               = _mm_mul_pd(fscal,dz00);
 220
 221             /* Update vectorial force */
 222             fix0             = _mm_add_pd(fix0,tx);
 223             fiy0             = _mm_add_pd(fiy0,ty);
 224             fiz0             = _mm_add_pd(fiz0,tz);
 225
 226             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 227
 228             /* Inner loop uses 51 flops */
 229         }
 230
 231         if(jidx<j_index_end)
 232         {
 233
 234             jnrA             = jjnr[jidx];
 235             j_coord_offsetA  = DIM*jnrA;
 236
 237             /* load j atom coordinates */
 238             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 239                                               &jx0,&jy0,&jz0);
 240
 241             /* Calculate displacement vector */
 242             dx00             = _mm_sub_pd(ix0,jx0);
 243             dy00             = _mm_sub_pd(iy0,jy0);
 244             dz00             = _mm_sub_pd(iz0,jz0);
 245
 246             /* Calculate squared distance and things based on it */
 247             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 248
 249             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 250
 251             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 252
 253             /* Load parameters for j particles */
 254             vdwjidx0A        = 2*vdwtype[jnrA+0];
 255
 256             /**************************
 257              * CALCULATE INTERACTIONS *
 258              **************************/
 259
 260             r00              = _mm_mul_pd(rsq00,rinv00);
 261
 262             /* Compute parameters for interactions between i and j atoms */
 263             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 264
 265             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 266
 267             /* Analytical LJ-PME */
 268             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 269             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 270             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 271             exponent         = gmx_simd_exp_d(ewcljrsq);
 272             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 273             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 274             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 275             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 276             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 277             vvdw             = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
 278             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 279             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);
 280
 281             /* Update potential sum for this i atom from the interaction with this j atom. */
 282             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 283             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 284
 285             fscal            = fvdw;
 286
 287             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 288
 289             /* Calculate temporary vectorial force */
 290             tx               = _mm_mul_pd(fscal,dx00);
 291             ty               = _mm_mul_pd(fscal,dy00);
 292             tz               = _mm_mul_pd(fscal,dz00);
 293
 294             /* Update vectorial force */
 295             fix0             = _mm_add_pd(fix0,tx);
 296             fiy0             = _mm_add_pd(fiy0,ty);
 297             fiz0             = _mm_add_pd(fiz0,tz);
 298
 299             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 300
 301             /* Inner loop uses 51 flops */
 302         }
 303
 304         /* End of innermost loop */
 305
 306         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 307                                               f+i_coord_offset,fshift+i_shift_offset);
 308
 309         ggid                        = gid[iidx];
 310         /* Update potential energies */
 311         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 312
 313         /* Increment number of inner iterations */
 314         inneriter                  += j_index_end - j_index_start;
 315
 316         /* Outer loop uses 7 flops */
 317     }
 318
 319     /* Increment number of outer iterations */
 320     outeriter        += nri;
 321
 322     /* Update outer/inner flops */
 323
 324     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*51);
 325 }
 326 /*
 327  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEw_GeomP1P1_F_sse2_double
 328  * Electrostatics interaction: None
 329  * VdW interaction:            LJEwald
 330  * Geometry:                   Particle-Particle
 331  * Calculate force/pot:        Force
 332  */
 333 void
 334 nb_kernel_ElecNone_VdwLJEw_GeomP1P1_F_sse2_double
 335                     (t_nblist                    * gmx_restrict       nlist,
 336                      rvec                        * gmx_restrict          xx,
 337                      rvec                        * gmx_restrict          ff,
 338                      t_forcerec                  * gmx_restrict          fr,
 339                      t_mdatoms                   * gmx_restrict     mdatoms,
 340                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 341                      t_nrnb                      * gmx_restrict        nrnb)
 342 {
 343     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 344      * just 0 for non-waters.
 345      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 346      * jnr indices corresponding to data put in the four positions in the SIMD register.
 347      */
 348     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 349     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 350     int              jnrA,jnrB;
 351     int              j_coord_offsetA,j_coord_offsetB;
 352     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 353     real             rcutoff_scalar;
 354     real             *shiftvec,*fshift,*x,*f;
 355     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 356     int              vdwioffset0;
 357     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 358     int              vdwjidx0A,vdwjidx0B;
 359     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 360     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 361     int              nvdwtype;
 362     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 363     int              *vdwtype;
 364     real             *vdwparam;
 365     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 366     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 367     __m128d           c6grid_00;
 368     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 369     real             *vdwgridparam;
 370     __m128d           one_half = _mm_set1_pd(0.5);
 371     __m128d           minus_one = _mm_set1_pd(-1.0);
 372     __m128d          dummy_mask,cutoff_mask;
 373     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 374     __m128d          one     = _mm_set1_pd(1.0);
 375     __m128d          two     = _mm_set1_pd(2.0);
 376     x                = xx[0];
 377     f                = ff[0];
 378
 379     nri              = nlist->nri;
 380     iinr             = nlist->iinr;
 381     jindex           = nlist->jindex;
 382     jjnr             = nlist->jjnr;
 383     shiftidx         = nlist->shift;
 384     gid              = nlist->gid;
 385     shiftvec         = fr->shift_vec[0];
 386     fshift           = fr->fshift[0];
 387     nvdwtype         = fr->ntype;
 388     vdwparam         = fr->nbfp;
 389     vdwtype          = mdatoms->typeA;
 390     vdwgridparam     = fr->ljpme_c6grid;
 391     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 392     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 393     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 394
 395     /* Avoid stupid compiler warnings */
 396     jnrA = jnrB = 0;
 397     j_coord_offsetA = 0;
 398     j_coord_offsetB = 0;
 399
 400     outeriter        = 0;
 401     inneriter        = 0;
 402
 403     /* Start outer loop over neighborlists */
 404     for(iidx=0; iidx<nri; iidx++)
 405     {
 406         /* Load shift vector for this list */
 407         i_shift_offset   = DIM*shiftidx[iidx];
 408
 409         /* Load limits for loop over neighbors */
 410         j_index_start    = jindex[iidx];
 411         j_index_end      = jindex[iidx+1];
 412
 413         /* Get outer coordinate index */
 414         inr              = iinr[iidx];
 415         i_coord_offset   = DIM*inr;
 416
 417         /* Load i particle coords and add shift vector */
 418         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 419
 420         fix0             = _mm_setzero_pd();
 421         fiy0             = _mm_setzero_pd();
 422         fiz0             = _mm_setzero_pd();
 423
 424         /* Load parameters for i particles */
 425         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 426
 427         /* Start inner kernel loop */
 428         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 429         {
 430
 431             /* Get j neighbor index, and coordinate index */
 432             jnrA             = jjnr[jidx];
 433             jnrB             = jjnr[jidx+1];
 434             j_coord_offsetA  = DIM*jnrA;
 435             j_coord_offsetB  = DIM*jnrB;
 436
 437             /* load j atom coordinates */
 438             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 439                                               &jx0,&jy0,&jz0);
 440
 441             /* Calculate displacement vector */
 442             dx00             = _mm_sub_pd(ix0,jx0);
 443             dy00             = _mm_sub_pd(iy0,jy0);
 444             dz00             = _mm_sub_pd(iz0,jz0);
 445
 446             /* Calculate squared distance and things based on it */
 447             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 448
 449             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 450
 451             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 452
 453             /* Load parameters for j particles */
 454             vdwjidx0A        = 2*vdwtype[jnrA+0];
 455             vdwjidx0B        = 2*vdwtype[jnrB+0];
 456
 457             /**************************
 458              * CALCULATE INTERACTIONS *
 459              **************************/
 460
 461             r00              = _mm_mul_pd(rsq00,rinv00);
 462
 463             /* Compute parameters for interactions between i and j atoms */
 464             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 465                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 466
 467             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 468                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 469
 470             /* Analytical LJ-PME */
 471             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 472             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 473             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 474             exponent         = gmx_simd_exp_d(ewcljrsq);
 475             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 476             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 477             /* f6A = 6 * C6grid * (1 - poly) */
 478             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 479             /* f6B = C6grid * exponent * beta^6 */
 480             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 481             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 482             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);
 483
 484             fscal            = fvdw;
 485
 486             /* Calculate temporary vectorial force */
 487             tx               = _mm_mul_pd(fscal,dx00);
 488             ty               = _mm_mul_pd(fscal,dy00);
 489             tz               = _mm_mul_pd(fscal,dz00);
 490
 491             /* Update vectorial force */
 492             fix0             = _mm_add_pd(fix0,tx);
 493             fiy0             = _mm_add_pd(fiy0,ty);
 494             fiz0             = _mm_add_pd(fiz0,tz);
 495
 496             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 497
 498             /* Inner loop uses 46 flops */
 499         }
 500
 501         if(jidx<j_index_end)
 502         {
 503
 504             jnrA             = jjnr[jidx];
 505             j_coord_offsetA  = DIM*jnrA;
 506
 507             /* load j atom coordinates */
 508             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 509                                               &jx0,&jy0,&jz0);
 510
 511             /* Calculate displacement vector */
 512             dx00             = _mm_sub_pd(ix0,jx0);
 513             dy00             = _mm_sub_pd(iy0,jy0);
 514             dz00             = _mm_sub_pd(iz0,jz0);
 515
 516             /* Calculate squared distance and things based on it */
 517             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 518
 519             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 520
 521             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 522
 523             /* Load parameters for j particles */
 524             vdwjidx0A        = 2*vdwtype[jnrA+0];
 525
 526             /**************************
 527              * CALCULATE INTERACTIONS *
 528              **************************/
 529
 530             r00              = _mm_mul_pd(rsq00,rinv00);
 531
 532             /* Compute parameters for interactions between i and j atoms */
 533             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 534
 535             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 536
 537             /* Analytical LJ-PME */
 538             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 539             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 540             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 541             exponent         = gmx_simd_exp_d(ewcljrsq);
 542             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 543             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 544             /* f6A = 6 * C6grid * (1 - poly) */
 545             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 546             /* f6B = C6grid * exponent * beta^6 */
 547             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 548             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 549             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);
 550
 551             fscal            = fvdw;
 552
 553             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 554
 555             /* Calculate temporary vectorial force */
 556             tx               = _mm_mul_pd(fscal,dx00);
 557             ty               = _mm_mul_pd(fscal,dy00);
 558             tz               = _mm_mul_pd(fscal,dz00);
 559
 560             /* Update vectorial force */
 561             fix0             = _mm_add_pd(fix0,tx);
 562             fiy0             = _mm_add_pd(fiy0,ty);
 563             fiz0             = _mm_add_pd(fiz0,tz);
 564
 565             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 566
 567             /* Inner loop uses 46 flops */
 568         }
 569
 570         /* End of innermost loop */
 571
 572         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 573                                               f+i_coord_offset,fshift+i_shift_offset);
 574
 575         /* Increment number of inner iterations */
 576         inneriter                  += j_index_end - j_index_start;
 577
 578         /* Outer loop uses 6 flops */
 579     }
 580
 581     /* Increment number of outer iterations */
 582     outeriter        += nri;
 583
 584     /* Update outer/inner flops */
 585
 586     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*46);
 587 }