src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecNone_VdwLJEwSh_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_VdwLJEwSh_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_VdwLJEwSh_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     rcutoff_scalar   = fr->rvdw;
 122     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 123     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 124
 125     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 126     rvdw             = _mm_set1_pd(fr->rvdw);
 127
 128     /* Avoid stupid compiler warnings */
 129     jnrA = jnrB = 0;
 130     j_coord_offsetA = 0;
 131     j_coord_offsetB = 0;
 132
 133     outeriter        = 0;
 134     inneriter        = 0;
 135
 136     /* Start outer loop over neighborlists */
 137     for(iidx=0; iidx<nri; iidx++)
 138     {
 139         /* Load shift vector for this list */
 140         i_shift_offset   = DIM*shiftidx[iidx];
 141
 142         /* Load limits for loop over neighbors */
 143         j_index_start    = jindex[iidx];
 144         j_index_end      = jindex[iidx+1];
 145
 146         /* Get outer coordinate index */
 147         inr              = iinr[iidx];
 148         i_coord_offset   = DIM*inr;
 149
 150         /* Load i particle coords and add shift vector */
 151         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 152
 153         fix0             = _mm_setzero_pd();
 154         fiy0             = _mm_setzero_pd();
 155         fiz0             = _mm_setzero_pd();
 156
 157         /* Load parameters for i particles */
 158         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 159
 160         /* Reset potential sums */
 161         vvdwsum          = _mm_setzero_pd();
 162
 163         /* Start inner kernel loop */
 164         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 165         {
 166
 167             /* Get j neighbor index, and coordinate index */
 168             jnrA             = jjnr[jidx];
 169             jnrB             = jjnr[jidx+1];
 170             j_coord_offsetA  = DIM*jnrA;
 171             j_coord_offsetB  = DIM*jnrB;
 172
 173             /* load j atom coordinates */
 174             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 175                                               &jx0,&jy0,&jz0);
 176
 177             /* Calculate displacement vector */
 178             dx00             = _mm_sub_pd(ix0,jx0);
 179             dy00             = _mm_sub_pd(iy0,jy0);
 180             dz00             = _mm_sub_pd(iz0,jz0);
 181
 182             /* Calculate squared distance and things based on it */
 183             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 184
 185             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 186
 187             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 188
 189             /* Load parameters for j particles */
 190             vdwjidx0A        = 2*vdwtype[jnrA+0];
 191             vdwjidx0B        = 2*vdwtype[jnrB+0];
 192
 193             /**************************
 194              * CALCULATE INTERACTIONS *
 195              **************************/
 196
 197             if (gmx_mm_any_lt(rsq00,rcutoff2))
 198             {
 199
 200             r00              = _mm_mul_pd(rsq00,rinv00);
 201
 202             /* Compute parameters for interactions between i and j atoms */
 203             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 204                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 205
 206             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 207                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 208
 209             /* Analytical LJ-PME */
 210             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 211             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 212             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 213             exponent         = gmx_simd_exp_d(ewcljrsq);
 214             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 215             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 216             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 217             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 218             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 219             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),
 220                                _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));
 221             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 222             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);
 223
 224             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 225
 226             /* Update potential sum for this i atom from the interaction with this j atom. */
 227             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 228             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 229
 230             fscal            = fvdw;
 231
 232             fscal            = _mm_and_pd(fscal,cutoff_mask);
 233
 234             /* Calculate temporary vectorial force */
 235             tx               = _mm_mul_pd(fscal,dx00);
 236             ty               = _mm_mul_pd(fscal,dy00);
 237             tz               = _mm_mul_pd(fscal,dz00);
 238
 239             /* Update vectorial force */
 240             fix0             = _mm_add_pd(fix0,tx);
 241             fiy0             = _mm_add_pd(fiy0,ty);
 242             fiz0             = _mm_add_pd(fiz0,tz);
 243
 244             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 245
 246             }
 247
 248             /* Inner loop uses 62 flops */
 249         }
 250
 251         if(jidx<j_index_end)
 252         {
 253
 254             jnrA             = jjnr[jidx];
 255             j_coord_offsetA  = DIM*jnrA;
 256
 257             /* load j atom coordinates */
 258             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 259                                               &jx0,&jy0,&jz0);
 260
 261             /* Calculate displacement vector */
 262             dx00             = _mm_sub_pd(ix0,jx0);
 263             dy00             = _mm_sub_pd(iy0,jy0);
 264             dz00             = _mm_sub_pd(iz0,jz0);
 265
 266             /* Calculate squared distance and things based on it */
 267             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 268
 269             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 270
 271             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 272
 273             /* Load parameters for j particles */
 274             vdwjidx0A        = 2*vdwtype[jnrA+0];
 275
 276             /**************************
 277              * CALCULATE INTERACTIONS *
 278              **************************/
 279
 280             if (gmx_mm_any_lt(rsq00,rcutoff2))
 281             {
 282
 283             r00              = _mm_mul_pd(rsq00,rinv00);
 284
 285             /* Compute parameters for interactions between i and j atoms */
 286             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 287
 288             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 289
 290             /* Analytical LJ-PME */
 291             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 292             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 293             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 294             exponent         = gmx_simd_exp_d(ewcljrsq);
 295             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 296             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 297             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 298             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 299             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 300             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),
 301                                _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));
 302             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 303             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);
 304
 305             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 306
 307             /* Update potential sum for this i atom from the interaction with this j atom. */
 308             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 309             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 310             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 311
 312             fscal            = fvdw;
 313
 314             fscal            = _mm_and_pd(fscal,cutoff_mask);
 315
 316             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 317
 318             /* Calculate temporary vectorial force */
 319             tx               = _mm_mul_pd(fscal,dx00);
 320             ty               = _mm_mul_pd(fscal,dy00);
 321             tz               = _mm_mul_pd(fscal,dz00);
 322
 323             /* Update vectorial force */
 324             fix0             = _mm_add_pd(fix0,tx);
 325             fiy0             = _mm_add_pd(fiy0,ty);
 326             fiz0             = _mm_add_pd(fiz0,tz);
 327
 328             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 329
 330             }
 331
 332             /* Inner loop uses 62 flops */
 333         }
 334
 335         /* End of innermost loop */
 336
 337         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 338                                               f+i_coord_offset,fshift+i_shift_offset);
 339
 340         ggid                        = gid[iidx];
 341         /* Update potential energies */
 342         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 343
 344         /* Increment number of inner iterations */
 345         inneriter                  += j_index_end - j_index_start;
 346
 347         /* Outer loop uses 7 flops */
 348     }
 349
 350     /* Increment number of outer iterations */
 351     outeriter        += nri;
 352
 353     /* Update outer/inner flops */
 354
 355     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*62);
 356 }
 357 /*
 358  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_F_sse2_double
 359  * Electrostatics interaction: None
 360  * VdW interaction:            LJEwald
 361  * Geometry:                   Particle-Particle
 362  * Calculate force/pot:        Force
 363  */
 364 void
 365 nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_F_sse2_double
 366                     (t_nblist                    * gmx_restrict       nlist,
 367                      rvec                        * gmx_restrict          xx,
 368                      rvec                        * gmx_restrict          ff,
 369                      t_forcerec                  * gmx_restrict          fr,
 370                      t_mdatoms                   * gmx_restrict     mdatoms,
 371                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 372                      t_nrnb                      * gmx_restrict        nrnb)
 373 {
 374     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 375      * just 0 for non-waters.
 376      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 377      * jnr indices corresponding to data put in the four positions in the SIMD register.
 378      */
 379     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 380     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 381     int              jnrA,jnrB;
 382     int              j_coord_offsetA,j_coord_offsetB;
 383     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 384     real             rcutoff_scalar;
 385     real             *shiftvec,*fshift,*x,*f;
 386     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 387     int              vdwioffset0;
 388     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 389     int              vdwjidx0A,vdwjidx0B;
 390     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 391     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 392     int              nvdwtype;
 393     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 394     int              *vdwtype;
 395     real             *vdwparam;
 396     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 397     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 398     __m128d           c6grid_00;
 399     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 400     real             *vdwgridparam;
 401     __m128d           one_half = _mm_set1_pd(0.5);
 402     __m128d           minus_one = _mm_set1_pd(-1.0);
 403     __m128d          dummy_mask,cutoff_mask;
 404     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 405     __m128d          one     = _mm_set1_pd(1.0);
 406     __m128d          two     = _mm_set1_pd(2.0);
 407     x                = xx[0];
 408     f                = ff[0];
 409
 410     nri              = nlist->nri;
 411     iinr             = nlist->iinr;
 412     jindex           = nlist->jindex;
 413     jjnr             = nlist->jjnr;
 414     shiftidx         = nlist->shift;
 415     gid              = nlist->gid;
 416     shiftvec         = fr->shift_vec[0];
 417     fshift           = fr->fshift[0];
 418     nvdwtype         = fr->ntype;
 419     vdwparam         = fr->nbfp;
 420     vdwtype          = mdatoms->typeA;
 421     vdwgridparam     = fr->ljpme_c6grid;
 422     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 423     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 424     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 425
 426     rcutoff_scalar   = fr->rvdw;
 427     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 428     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 429
 430     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 431     rvdw             = _mm_set1_pd(fr->rvdw);
 432
 433     /* Avoid stupid compiler warnings */
 434     jnrA = jnrB = 0;
 435     j_coord_offsetA = 0;
 436     j_coord_offsetB = 0;
 437
 438     outeriter        = 0;
 439     inneriter        = 0;
 440
 441     /* Start outer loop over neighborlists */
 442     for(iidx=0; iidx<nri; iidx++)
 443     {
 444         /* Load shift vector for this list */
 445         i_shift_offset   = DIM*shiftidx[iidx];
 446
 447         /* Load limits for loop over neighbors */
 448         j_index_start    = jindex[iidx];
 449         j_index_end      = jindex[iidx+1];
 450
 451         /* Get outer coordinate index */
 452         inr              = iinr[iidx];
 453         i_coord_offset   = DIM*inr;
 454
 455         /* Load i particle coords and add shift vector */
 456         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 457
 458         fix0             = _mm_setzero_pd();
 459         fiy0             = _mm_setzero_pd();
 460         fiz0             = _mm_setzero_pd();
 461
 462         /* Load parameters for i particles */
 463         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 464
 465         /* Start inner kernel loop */
 466         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 467         {
 468
 469             /* Get j neighbor index, and coordinate index */
 470             jnrA             = jjnr[jidx];
 471             jnrB             = jjnr[jidx+1];
 472             j_coord_offsetA  = DIM*jnrA;
 473             j_coord_offsetB  = DIM*jnrB;
 474
 475             /* load j atom coordinates */
 476             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 477                                               &jx0,&jy0,&jz0);
 478
 479             /* Calculate displacement vector */
 480             dx00             = _mm_sub_pd(ix0,jx0);
 481             dy00             = _mm_sub_pd(iy0,jy0);
 482             dz00             = _mm_sub_pd(iz0,jz0);
 483
 484             /* Calculate squared distance and things based on it */
 485             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 486
 487             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 488
 489             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 490
 491             /* Load parameters for j particles */
 492             vdwjidx0A        = 2*vdwtype[jnrA+0];
 493             vdwjidx0B        = 2*vdwtype[jnrB+0];
 494
 495             /**************************
 496              * CALCULATE INTERACTIONS *
 497              **************************/
 498
 499             if (gmx_mm_any_lt(rsq00,rcutoff2))
 500             {
 501
 502             r00              = _mm_mul_pd(rsq00,rinv00);
 503
 504             /* Compute parameters for interactions between i and j atoms */
 505             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 506                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 507
 508             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 509                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 510
 511             /* Analytical LJ-PME */
 512             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 513             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 514             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 515             exponent         = gmx_simd_exp_d(ewcljrsq);
 516             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 517             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 518             /* f6A = 6 * C6grid * (1 - poly) */
 519             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 520             /* f6B = C6grid * exponent * beta^6 */
 521             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 522             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 523             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);
 524
 525             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 526
 527             fscal            = fvdw;
 528
 529             fscal            = _mm_and_pd(fscal,cutoff_mask);
 530
 531             /* Calculate temporary vectorial force */
 532             tx               = _mm_mul_pd(fscal,dx00);
 533             ty               = _mm_mul_pd(fscal,dy00);
 534             tz               = _mm_mul_pd(fscal,dz00);
 535
 536             /* Update vectorial force */
 537             fix0             = _mm_add_pd(fix0,tx);
 538             fiy0             = _mm_add_pd(fiy0,ty);
 539             fiz0             = _mm_add_pd(fiz0,tz);
 540
 541             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 542
 543             }
 544
 545             /* Inner loop uses 49 flops */
 546         }
 547
 548         if(jidx<j_index_end)
 549         {
 550
 551             jnrA             = jjnr[jidx];
 552             j_coord_offsetA  = DIM*jnrA;
 553
 554             /* load j atom coordinates */
 555             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 556                                               &jx0,&jy0,&jz0);
 557
 558             /* Calculate displacement vector */
 559             dx00             = _mm_sub_pd(ix0,jx0);
 560             dy00             = _mm_sub_pd(iy0,jy0);
 561             dz00             = _mm_sub_pd(iz0,jz0);
 562
 563             /* Calculate squared distance and things based on it */
 564             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 565
 566             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 567
 568             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 569
 570             /* Load parameters for j particles */
 571             vdwjidx0A        = 2*vdwtype[jnrA+0];
 572
 573             /**************************
 574              * CALCULATE INTERACTIONS *
 575              **************************/
 576
 577             if (gmx_mm_any_lt(rsq00,rcutoff2))
 578             {
 579
 580             r00              = _mm_mul_pd(rsq00,rinv00);
 581
 582             /* Compute parameters for interactions between i and j atoms */
 583             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 584
 585             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 586
 587             /* Analytical LJ-PME */
 588             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 589             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 590             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 591             exponent         = gmx_simd_exp_d(ewcljrsq);
 592             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 593             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 594             /* f6A = 6 * C6grid * (1 - poly) */
 595             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 596             /* f6B = C6grid * exponent * beta^6 */
 597             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 598             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 599             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);
 600
 601             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 602
 603             fscal            = fvdw;
 604
 605             fscal            = _mm_and_pd(fscal,cutoff_mask);
 606
 607             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 608
 609             /* Calculate temporary vectorial force */
 610             tx               = _mm_mul_pd(fscal,dx00);
 611             ty               = _mm_mul_pd(fscal,dy00);
 612             tz               = _mm_mul_pd(fscal,dz00);
 613
 614             /* Update vectorial force */
 615             fix0             = _mm_add_pd(fix0,tx);
 616             fiy0             = _mm_add_pd(fiy0,ty);
 617             fiz0             = _mm_add_pd(fiz0,tz);
 618
 619             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 620
 621             }
 622
 623             /* Inner loop uses 49 flops */
 624         }
 625
 626         /* End of innermost loop */
 627
 628         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 629                                               f+i_coord_offset,fshift+i_shift_offset);
 630
 631         /* Increment number of inner iterations */
 632         inneriter                  += j_index_end - j_index_start;
 633
 634         /* Outer loop uses 6 flops */
 635     }
 636
 637     /* Increment number of outer iterations */
 638     outeriter        += nri;
 639
 640     /* Update outer/inner flops */
 641
 642     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*49);
 643 }