src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_avx_128_fma_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 avx_128_fma_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_avx_128_fma_double.h"
  48 #include "kernelutil_x86_avx_128_fma_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_VF_avx_128_fma_double
  52  * Electrostatics interaction: None
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_VF_avx_128_fma_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     real             *vdwgridparam;
  93     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
  94     __m128d           one_half  = _mm_set1_pd(0.5);
  95     __m128d           minus_one = _mm_set1_pd(-1.0);
  96     __m128d          dummy_mask,cutoff_mask;
  97     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  98     __m128d          one     = _mm_set1_pd(1.0);
  99     __m128d          two     = _mm_set1_pd(2.0);
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     nvdwtype         = fr->ntype;
 112     vdwparam         = fr->nbfp;
 113     vdwtype          = mdatoms->typeA;
 114     vdwgridparam     = fr->ljpme_c6grid;
 115     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 116     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 117     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 118
 119     rcutoff_scalar   = fr->rvdw;
 120     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 121     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 122
 123     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 124     rvdw             = _mm_set1_pd(fr->rvdw);
 125
 126     /* Avoid stupid compiler warnings */
 127     jnrA = jnrB = 0;
 128     j_coord_offsetA = 0;
 129     j_coord_offsetB = 0;
 130
 131     outeriter        = 0;
 132     inneriter        = 0;
 133
 134     /* Start outer loop over neighborlists */
 135     for(iidx=0; iidx<nri; iidx++)
 136     {
 137         /* Load shift vector for this list */
 138         i_shift_offset   = DIM*shiftidx[iidx];
 139
 140         /* Load limits for loop over neighbors */
 141         j_index_start    = jindex[iidx];
 142         j_index_end      = jindex[iidx+1];
 143
 144         /* Get outer coordinate index */
 145         inr              = iinr[iidx];
 146         i_coord_offset   = DIM*inr;
 147
 148         /* Load i particle coords and add shift vector */
 149         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 150
 151         fix0             = _mm_setzero_pd();
 152         fiy0             = _mm_setzero_pd();
 153         fiz0             = _mm_setzero_pd();
 154
 155         /* Load parameters for i particles */
 156         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 157
 158         /* Reset potential sums */
 159         vvdwsum          = _mm_setzero_pd();
 160
 161         /* Start inner kernel loop */
 162         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 163         {
 164
 165             /* Get j neighbor index, and coordinate index */
 166             jnrA             = jjnr[jidx];
 167             jnrB             = jjnr[jidx+1];
 168             j_coord_offsetA  = DIM*jnrA;
 169             j_coord_offsetB  = DIM*jnrB;
 170
 171             /* load j atom coordinates */
 172             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 173                                               &jx0,&jy0,&jz0);
 174
 175             /* Calculate displacement vector */
 176             dx00             = _mm_sub_pd(ix0,jx0);
 177             dy00             = _mm_sub_pd(iy0,jy0);
 178             dz00             = _mm_sub_pd(iz0,jz0);
 179
 180             /* Calculate squared distance and things based on it */
 181             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 182
 183             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 184
 185             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 186
 187             /* Load parameters for j particles */
 188             vdwjidx0A        = 2*vdwtype[jnrA+0];
 189             vdwjidx0B        = 2*vdwtype[jnrB+0];
 190
 191             /**************************
 192              * CALCULATE INTERACTIONS *
 193              **************************/
 194
 195             if (gmx_mm_any_lt(rsq00,rcutoff2))
 196             {
 197
 198             r00              = _mm_mul_pd(rsq00,rinv00);
 199
 200             /* Compute parameters for interactions between i and j atoms */
 201             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 202                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 203             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 204                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 205
 206             /* Analytical LJ-PME */
 207             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 208             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 209             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 210             exponent         = gmx_simd_exp_d(ewcljrsq);
 211             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 212             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 213             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 214             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 215             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 216             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 217                                _mm_mul_pd(_mm_sub_pd(vvdw6,_mm_macc_pd(c6grid_00,sh_lj_ewald,_mm_mul_pd(c6_00,sh_vdw_invrcut6))),one_sixth));
 218             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 219             fvdw             = _mm_mul_pd(_mm_add_pd(vvdw12,_mm_msub_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6),vvdw6)),rinvsq00);
 220
 221             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 222
 223             /* Update potential sum for this i atom from the interaction with this j atom. */
 224             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 225             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 226
 227             fscal            = fvdw;
 228
 229             fscal            = _mm_and_pd(fscal,cutoff_mask);
 230
 231             /* Update vectorial force */
 232             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 233             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 234             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 235
 236             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 237                                                    _mm_mul_pd(dx00,fscal),
 238                                                    _mm_mul_pd(dy00,fscal),
 239                                                    _mm_mul_pd(dz00,fscal));
 240
 241             }
 242
 243             /* Inner loop uses 58 flops */
 244         }
 245
 246         if(jidx<j_index_end)
 247         {
 248
 249             jnrA             = jjnr[jidx];
 250             j_coord_offsetA  = DIM*jnrA;
 251
 252             /* load j atom coordinates */
 253             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 254                                               &jx0,&jy0,&jz0);
 255
 256             /* Calculate displacement vector */
 257             dx00             = _mm_sub_pd(ix0,jx0);
 258             dy00             = _mm_sub_pd(iy0,jy0);
 259             dz00             = _mm_sub_pd(iz0,jz0);
 260
 261             /* Calculate squared distance and things based on it */
 262             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 263
 264             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 265
 266             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 267
 268             /* Load parameters for j particles */
 269             vdwjidx0A        = 2*vdwtype[jnrA+0];
 270
 271             /**************************
 272              * CALCULATE INTERACTIONS *
 273              **************************/
 274
 275             if (gmx_mm_any_lt(rsq00,rcutoff2))
 276             {
 277
 278             r00              = _mm_mul_pd(rsq00,rinv00);
 279
 280             /* Compute parameters for interactions between i and j atoms */
 281             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 282             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 283
 284             /* Analytical LJ-PME */
 285             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 286             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 287             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 288             exponent         = gmx_simd_exp_d(ewcljrsq);
 289             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 290             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 291             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 292             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 293             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 294             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 295                                _mm_mul_pd(_mm_sub_pd(vvdw6,_mm_macc_pd(c6grid_00,sh_lj_ewald,_mm_mul_pd(c6_00,sh_vdw_invrcut6))),one_sixth));
 296             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 297             fvdw             = _mm_mul_pd(_mm_add_pd(vvdw12,_mm_msub_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6),vvdw6)),rinvsq00);
 298
 299             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 300
 301             /* Update potential sum for this i atom from the interaction with this j atom. */
 302             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 303             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 304             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 305
 306             fscal            = fvdw;
 307
 308             fscal            = _mm_and_pd(fscal,cutoff_mask);
 309
 310             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 311
 312             /* Update vectorial force */
 313             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 314             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 315             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 316
 317             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 318                                                    _mm_mul_pd(dx00,fscal),
 319                                                    _mm_mul_pd(dy00,fscal),
 320                                                    _mm_mul_pd(dz00,fscal));
 321
 322             }
 323
 324             /* Inner loop uses 58 flops */
 325         }
 326
 327         /* End of innermost loop */
 328
 329         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 330                                               f+i_coord_offset,fshift+i_shift_offset);
 331
 332         ggid                        = gid[iidx];
 333         /* Update potential energies */
 334         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 335
 336         /* Increment number of inner iterations */
 337         inneriter                  += j_index_end - j_index_start;
 338
 339         /* Outer loop uses 7 flops */
 340     }
 341
 342     /* Increment number of outer iterations */
 343     outeriter        += nri;
 344
 345     /* Update outer/inner flops */
 346
 347     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*58);
 348 }
 349 /*
 350  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_F_avx_128_fma_double
 351  * Electrostatics interaction: None
 352  * VdW interaction:            LJEwald
 353  * Geometry:                   Particle-Particle
 354  * Calculate force/pot:        Force
 355  */
 356 void
 357 nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_F_avx_128_fma_double
 358                     (t_nblist                    * gmx_restrict       nlist,
 359                      rvec                        * gmx_restrict          xx,
 360                      rvec                        * gmx_restrict          ff,
 361                      t_forcerec                  * gmx_restrict          fr,
 362                      t_mdatoms                   * gmx_restrict     mdatoms,
 363                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 364                      t_nrnb                      * gmx_restrict        nrnb)
 365 {
 366     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 367      * just 0 for non-waters.
 368      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 369      * jnr indices corresponding to data put in the four positions in the SIMD register.
 370      */
 371     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 372     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 373     int              jnrA,jnrB;
 374     int              j_coord_offsetA,j_coord_offsetB;
 375     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 376     real             rcutoff_scalar;
 377     real             *shiftvec,*fshift,*x,*f;
 378     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 379     int              vdwioffset0;
 380     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 381     int              vdwjidx0A,vdwjidx0B;
 382     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 383     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 384     int              nvdwtype;
 385     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 386     int              *vdwtype;
 387     real             *vdwparam;
 388     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 389     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 390     __m128d           c6grid_00;
 391     real             *vdwgridparam;
 392     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 393     __m128d           one_half  = _mm_set1_pd(0.5);
 394     __m128d           minus_one = _mm_set1_pd(-1.0);
 395     __m128d          dummy_mask,cutoff_mask;
 396     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 397     __m128d          one     = _mm_set1_pd(1.0);
 398     __m128d          two     = _mm_set1_pd(2.0);
 399     x                = xx[0];
 400     f                = ff[0];
 401
 402     nri              = nlist->nri;
 403     iinr             = nlist->iinr;
 404     jindex           = nlist->jindex;
 405     jjnr             = nlist->jjnr;
 406     shiftidx         = nlist->shift;
 407     gid              = nlist->gid;
 408     shiftvec         = fr->shift_vec[0];
 409     fshift           = fr->fshift[0];
 410     nvdwtype         = fr->ntype;
 411     vdwparam         = fr->nbfp;
 412     vdwtype          = mdatoms->typeA;
 413     vdwgridparam     = fr->ljpme_c6grid;
 414     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 415     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 416     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 417
 418     rcutoff_scalar   = fr->rvdw;
 419     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 420     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 421
 422     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 423     rvdw             = _mm_set1_pd(fr->rvdw);
 424
 425     /* Avoid stupid compiler warnings */
 426     jnrA = jnrB = 0;
 427     j_coord_offsetA = 0;
 428     j_coord_offsetB = 0;
 429
 430     outeriter        = 0;
 431     inneriter        = 0;
 432
 433     /* Start outer loop over neighborlists */
 434     for(iidx=0; iidx<nri; iidx++)
 435     {
 436         /* Load shift vector for this list */
 437         i_shift_offset   = DIM*shiftidx[iidx];
 438
 439         /* Load limits for loop over neighbors */
 440         j_index_start    = jindex[iidx];
 441         j_index_end      = jindex[iidx+1];
 442
 443         /* Get outer coordinate index */
 444         inr              = iinr[iidx];
 445         i_coord_offset   = DIM*inr;
 446
 447         /* Load i particle coords and add shift vector */
 448         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 449
 450         fix0             = _mm_setzero_pd();
 451         fiy0             = _mm_setzero_pd();
 452         fiz0             = _mm_setzero_pd();
 453
 454         /* Load parameters for i particles */
 455         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 456
 457         /* Start inner kernel loop */
 458         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 459         {
 460
 461             /* Get j neighbor index, and coordinate index */
 462             jnrA             = jjnr[jidx];
 463             jnrB             = jjnr[jidx+1];
 464             j_coord_offsetA  = DIM*jnrA;
 465             j_coord_offsetB  = DIM*jnrB;
 466
 467             /* load j atom coordinates */
 468             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 469                                               &jx0,&jy0,&jz0);
 470
 471             /* Calculate displacement vector */
 472             dx00             = _mm_sub_pd(ix0,jx0);
 473             dy00             = _mm_sub_pd(iy0,jy0);
 474             dz00             = _mm_sub_pd(iz0,jz0);
 475
 476             /* Calculate squared distance and things based on it */
 477             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 478
 479             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 480
 481             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 482
 483             /* Load parameters for j particles */
 484             vdwjidx0A        = 2*vdwtype[jnrA+0];
 485             vdwjidx0B        = 2*vdwtype[jnrB+0];
 486
 487             /**************************
 488              * CALCULATE INTERACTIONS *
 489              **************************/
 490
 491             if (gmx_mm_any_lt(rsq00,rcutoff2))
 492             {
 493
 494             r00              = _mm_mul_pd(rsq00,rinv00);
 495
 496             /* Compute parameters for interactions between i and j atoms */
 497             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 498                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 499             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 500                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 501
 502             /* Analytical LJ-PME */
 503             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 504             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 505             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 506             exponent         = gmx_simd_exp_d(ewcljrsq);
 507             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 508             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 509             /* f6A = 6 * C6grid * (1 - poly) */
 510             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 511             /* f6B = C6grid * exponent * beta^6 */
 512             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 513             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 514             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 515
 516             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 517
 518             fscal            = fvdw;
 519
 520             fscal            = _mm_and_pd(fscal,cutoff_mask);
 521
 522             /* Update vectorial force */
 523             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 524             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 525             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 526
 527             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 528                                                    _mm_mul_pd(dx00,fscal),
 529                                                    _mm_mul_pd(dy00,fscal),
 530                                                    _mm_mul_pd(dz00,fscal));
 531
 532             }
 533
 534             /* Inner loop uses 50 flops */
 535         }
 536
 537         if(jidx<j_index_end)
 538         {
 539
 540             jnrA             = jjnr[jidx];
 541             j_coord_offsetA  = DIM*jnrA;
 542
 543             /* load j atom coordinates */
 544             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 545                                               &jx0,&jy0,&jz0);
 546
 547             /* Calculate displacement vector */
 548             dx00             = _mm_sub_pd(ix0,jx0);
 549             dy00             = _mm_sub_pd(iy0,jy0);
 550             dz00             = _mm_sub_pd(iz0,jz0);
 551
 552             /* Calculate squared distance and things based on it */
 553             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 554
 555             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 556
 557             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 558
 559             /* Load parameters for j particles */
 560             vdwjidx0A        = 2*vdwtype[jnrA+0];
 561
 562             /**************************
 563              * CALCULATE INTERACTIONS *
 564              **************************/
 565
 566             if (gmx_mm_any_lt(rsq00,rcutoff2))
 567             {
 568
 569             r00              = _mm_mul_pd(rsq00,rinv00);
 570
 571             /* Compute parameters for interactions between i and j atoms */
 572             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 573             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 574
 575             /* Analytical LJ-PME */
 576             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 577             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 578             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 579             exponent         = gmx_simd_exp_d(ewcljrsq);
 580             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 581             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 582             /* f6A = 6 * C6grid * (1 - poly) */
 583             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 584             /* f6B = C6grid * exponent * beta^6 */
 585             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 586             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 587             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 588
 589             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 590
 591             fscal            = fvdw;
 592
 593             fscal            = _mm_and_pd(fscal,cutoff_mask);
 594
 595             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 596
 597             /* Update vectorial force */
 598             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 599             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 600             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 601
 602             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 603                                                    _mm_mul_pd(dx00,fscal),
 604                                                    _mm_mul_pd(dy00,fscal),
 605                                                    _mm_mul_pd(dz00,fscal));
 606
 607             }
 608
 609             /* Inner loop uses 50 flops */
 610         }
 611
 612         /* End of innermost loop */
 613
 614         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 615                                               f+i_coord_offset,fshift+i_shift_offset);
 616
 617         /* Increment number of inner iterations */
 618         inneriter                  += j_index_end - j_index_start;
 619
 620         /* Outer loop uses 6 flops */
 621     }
 622
 623     /* Increment number of outer iterations */
 624     outeriter        += nri;
 625
 626     /* Update outer/inner flops */
 627
 628     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*50);
 629 }