src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecNone_VdwLJEw_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_VdwLJEw_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_VdwLJEw_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     /* 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             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 194                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 195
 196             /* Analytical LJ-PME */
 197             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 198             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 199             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 200             exponent         = gmx_simd_exp_d(ewcljrsq);
 201             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 202             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 203             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 204             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 205             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 206             vvdw             = _mm_msub_pd(vvdw12,one_twelfth,_mm_mul_pd(vvdw6,one_sixth));
 207             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 208             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);
 209
 210             /* Update potential sum for this i atom from the interaction with this j atom. */
 211             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 212
 213             fscal            = fvdw;
 214
 215             /* Update vectorial force */
 216             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 217             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 218             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 219
 220             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 221                                                    _mm_mul_pd(dx00,fscal),
 222                                                    _mm_mul_pd(dy00,fscal),
 223                                                    _mm_mul_pd(dz00,fscal));
 224
 225             /* Inner loop uses 50 flops */
 226         }
 227
 228         if(jidx<j_index_end)
 229         {
 230
 231             jnrA             = jjnr[jidx];
 232             j_coord_offsetA  = DIM*jnrA;
 233
 234             /* load j atom coordinates */
 235             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 236                                               &jx0,&jy0,&jz0);
 237
 238             /* Calculate displacement vector */
 239             dx00             = _mm_sub_pd(ix0,jx0);
 240             dy00             = _mm_sub_pd(iy0,jy0);
 241             dz00             = _mm_sub_pd(iz0,jz0);
 242
 243             /* Calculate squared distance and things based on it */
 244             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 245
 246             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 247
 248             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 249
 250             /* Load parameters for j particles */
 251             vdwjidx0A        = 2*vdwtype[jnrA+0];
 252
 253             /**************************
 254              * CALCULATE INTERACTIONS *
 255              **************************/
 256
 257             r00              = _mm_mul_pd(rsq00,rinv00);
 258
 259             /* Compute parameters for interactions between i and j atoms */
 260             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 261             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 262
 263             /* Analytical LJ-PME */
 264             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 265             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 266             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 267             exponent         = gmx_simd_exp_d(ewcljrsq);
 268             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 269             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 270             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 271             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 272             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 273             vvdw             = _mm_msub_pd(vvdw12,one_twelfth,_mm_mul_pd(vvdw6,one_sixth));
 274             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 275             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);
 276
 277             /* Update potential sum for this i atom from the interaction with this j atom. */
 278             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 279             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 280
 281             fscal            = fvdw;
 282
 283             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 284
 285             /* Update vectorial force */
 286             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 287             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 288             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 289
 290             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 291                                                    _mm_mul_pd(dx00,fscal),
 292                                                    _mm_mul_pd(dy00,fscal),
 293                                                    _mm_mul_pd(dz00,fscal));
 294
 295             /* Inner loop uses 50 flops */
 296         }
 297
 298         /* End of innermost loop */
 299
 300         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 301                                               f+i_coord_offset,fshift+i_shift_offset);
 302
 303         ggid                        = gid[iidx];
 304         /* Update potential energies */
 305         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 306
 307         /* Increment number of inner iterations */
 308         inneriter                  += j_index_end - j_index_start;
 309
 310         /* Outer loop uses 7 flops */
 311     }
 312
 313     /* Increment number of outer iterations */
 314     outeriter        += nri;
 315
 316     /* Update outer/inner flops */
 317
 318     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*50);
 319 }
 320 /*
 321  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEw_GeomP1P1_F_avx_128_fma_double
 322  * Electrostatics interaction: None
 323  * VdW interaction:            LJEwald
 324  * Geometry:                   Particle-Particle
 325  * Calculate force/pot:        Force
 326  */
 327 void
 328 nb_kernel_ElecNone_VdwLJEw_GeomP1P1_F_avx_128_fma_double
 329                     (t_nblist                    * gmx_restrict       nlist,
 330                      rvec                        * gmx_restrict          xx,
 331                      rvec                        * gmx_restrict          ff,
 332                      t_forcerec                  * gmx_restrict          fr,
 333                      t_mdatoms                   * gmx_restrict     mdatoms,
 334                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 335                      t_nrnb                      * gmx_restrict        nrnb)
 336 {
 337     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 338      * just 0 for non-waters.
 339      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 340      * jnr indices corresponding to data put in the four positions in the SIMD register.
 341      */
 342     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 343     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 344     int              jnrA,jnrB;
 345     int              j_coord_offsetA,j_coord_offsetB;
 346     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 347     real             rcutoff_scalar;
 348     real             *shiftvec,*fshift,*x,*f;
 349     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 350     int              vdwioffset0;
 351     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 352     int              vdwjidx0A,vdwjidx0B;
 353     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 354     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 355     int              nvdwtype;
 356     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 357     int              *vdwtype;
 358     real             *vdwparam;
 359     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 360     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 361     __m128d           c6grid_00;
 362     real             *vdwgridparam;
 363     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 364     __m128d           one_half  = _mm_set1_pd(0.5);
 365     __m128d           minus_one = _mm_set1_pd(-1.0);
 366     __m128d          dummy_mask,cutoff_mask;
 367     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 368     __m128d          one     = _mm_set1_pd(1.0);
 369     __m128d          two     = _mm_set1_pd(2.0);
 370     x                = xx[0];
 371     f                = ff[0];
 372
 373     nri              = nlist->nri;
 374     iinr             = nlist->iinr;
 375     jindex           = nlist->jindex;
 376     jjnr             = nlist->jjnr;
 377     shiftidx         = nlist->shift;
 378     gid              = nlist->gid;
 379     shiftvec         = fr->shift_vec[0];
 380     fshift           = fr->fshift[0];
 381     nvdwtype         = fr->ntype;
 382     vdwparam         = fr->nbfp;
 383     vdwtype          = mdatoms->typeA;
 384     vdwgridparam     = fr->ljpme_c6grid;
 385     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 386     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 387     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 388
 389     /* Avoid stupid compiler warnings */
 390     jnrA = jnrB = 0;
 391     j_coord_offsetA = 0;
 392     j_coord_offsetB = 0;
 393
 394     outeriter        = 0;
 395     inneriter        = 0;
 396
 397     /* Start outer loop over neighborlists */
 398     for(iidx=0; iidx<nri; iidx++)
 399     {
 400         /* Load shift vector for this list */
 401         i_shift_offset   = DIM*shiftidx[iidx];
 402
 403         /* Load limits for loop over neighbors */
 404         j_index_start    = jindex[iidx];
 405         j_index_end      = jindex[iidx+1];
 406
 407         /* Get outer coordinate index */
 408         inr              = iinr[iidx];
 409         i_coord_offset   = DIM*inr;
 410
 411         /* Load i particle coords and add shift vector */
 412         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 413
 414         fix0             = _mm_setzero_pd();
 415         fiy0             = _mm_setzero_pd();
 416         fiz0             = _mm_setzero_pd();
 417
 418         /* Load parameters for i particles */
 419         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 420
 421         /* Start inner kernel loop */
 422         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 423         {
 424
 425             /* Get j neighbor index, and coordinate index */
 426             jnrA             = jjnr[jidx];
 427             jnrB             = jjnr[jidx+1];
 428             j_coord_offsetA  = DIM*jnrA;
 429             j_coord_offsetB  = DIM*jnrB;
 430
 431             /* load j atom coordinates */
 432             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 433                                               &jx0,&jy0,&jz0);
 434
 435             /* Calculate displacement vector */
 436             dx00             = _mm_sub_pd(ix0,jx0);
 437             dy00             = _mm_sub_pd(iy0,jy0);
 438             dz00             = _mm_sub_pd(iz0,jz0);
 439
 440             /* Calculate squared distance and things based on it */
 441             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 442
 443             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 444
 445             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 446
 447             /* Load parameters for j particles */
 448             vdwjidx0A        = 2*vdwtype[jnrA+0];
 449             vdwjidx0B        = 2*vdwtype[jnrB+0];
 450
 451             /**************************
 452              * CALCULATE INTERACTIONS *
 453              **************************/
 454
 455             r00              = _mm_mul_pd(rsq00,rinv00);
 456
 457             /* Compute parameters for interactions between i and j atoms */
 458             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 459                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 460             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 461                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 462
 463             /* Analytical LJ-PME */
 464             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 465             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 466             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 467             exponent         = gmx_simd_exp_d(ewcljrsq);
 468             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 469             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 470             /* f6A = 6 * C6grid * (1 - poly) */
 471             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 472             /* f6B = C6grid * exponent * beta^6 */
 473             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 474             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 475             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 476
 477             fscal            = fvdw;
 478
 479             /* Update vectorial force */
 480             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 481             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 482             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 483
 484             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 485                                                    _mm_mul_pd(dx00,fscal),
 486                                                    _mm_mul_pd(dy00,fscal),
 487                                                    _mm_mul_pd(dz00,fscal));
 488
 489             /* Inner loop uses 47 flops */
 490         }
 491
 492         if(jidx<j_index_end)
 493         {
 494
 495             jnrA             = jjnr[jidx];
 496             j_coord_offsetA  = DIM*jnrA;
 497
 498             /* load j atom coordinates */
 499             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 500                                               &jx0,&jy0,&jz0);
 501
 502             /* Calculate displacement vector */
 503             dx00             = _mm_sub_pd(ix0,jx0);
 504             dy00             = _mm_sub_pd(iy0,jy0);
 505             dz00             = _mm_sub_pd(iz0,jz0);
 506
 507             /* Calculate squared distance and things based on it */
 508             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 509
 510             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 511
 512             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 513
 514             /* Load parameters for j particles */
 515             vdwjidx0A        = 2*vdwtype[jnrA+0];
 516
 517             /**************************
 518              * CALCULATE INTERACTIONS *
 519              **************************/
 520
 521             r00              = _mm_mul_pd(rsq00,rinv00);
 522
 523             /* Compute parameters for interactions between i and j atoms */
 524             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 525             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 526
 527             /* Analytical LJ-PME */
 528             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 529             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 530             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 531             exponent         = gmx_simd_exp_d(ewcljrsq);
 532             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 533             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 534             /* f6A = 6 * C6grid * (1 - poly) */
 535             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 536             /* f6B = C6grid * exponent * beta^6 */
 537             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 538             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 539             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 540
 541             fscal            = fvdw;
 542
 543             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 544
 545             /* Update vectorial force */
 546             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 547             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 548             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 549
 550             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 551                                                    _mm_mul_pd(dx00,fscal),
 552                                                    _mm_mul_pd(dy00,fscal),
 553                                                    _mm_mul_pd(dz00,fscal));
 554
 555             /* Inner loop uses 47 flops */
 556         }
 557
 558         /* End of innermost loop */
 559
 560         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 561                                               f+i_coord_offset,fshift+i_shift_offset);
 562
 563         /* Increment number of inner iterations */
 564         inneriter                  += j_index_end - j_index_start;
 565
 566         /* Outer loop uses 6 flops */
 567     }
 568
 569     /* Increment number of outer iterations */
 570     outeriter        += nri;
 571
 572     /* Update outer/inner flops */
 573
 574     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*47);
 575 }