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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_VF_avx_128_fma_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_avx_128_fma_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     real             *vdwgridparam;
  95     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
  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             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 206                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 207
 208             /* Analytical LJ-PME */
 209             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 210             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 211             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 212             exponent         = gmx_simd_exp_d(ewcljrsq);
 213             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 214             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 215             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 216             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 217             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 218             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 219                                _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));
 220             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 221             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);
 222
 223             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 224
 225             /* Update potential sum for this i atom from the interaction with this j atom. */
 226             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 227             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 228
 229             fscal            = fvdw;
 230
 231             fscal            = _mm_and_pd(fscal,cutoff_mask);
 232
 233             /* Update vectorial force */
 234             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 235             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 236             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 237
 238             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 239                                                    _mm_mul_pd(dx00,fscal),
 240                                                    _mm_mul_pd(dy00,fscal),
 241                                                    _mm_mul_pd(dz00,fscal));
 242
 243             }
 244
 245             /* Inner loop uses 58 flops */
 246         }
 247
 248         if(jidx<j_index_end)
 249         {
 250
 251             jnrA             = jjnr[jidx];
 252             j_coord_offsetA  = DIM*jnrA;
 253
 254             /* load j atom coordinates */
 255             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 256                                               &jx0,&jy0,&jz0);
 257
 258             /* Calculate displacement vector */
 259             dx00             = _mm_sub_pd(ix0,jx0);
 260             dy00             = _mm_sub_pd(iy0,jy0);
 261             dz00             = _mm_sub_pd(iz0,jz0);
 262
 263             /* Calculate squared distance and things based on it */
 264             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 265
 266             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 267
 268             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 269
 270             /* Load parameters for j particles */
 271             vdwjidx0A        = 2*vdwtype[jnrA+0];
 272
 273             /**************************
 274              * CALCULATE INTERACTIONS *
 275              **************************/
 276
 277             if (gmx_mm_any_lt(rsq00,rcutoff2))
 278             {
 279
 280             r00              = _mm_mul_pd(rsq00,rinv00);
 281
 282             /* Compute parameters for interactions between i and j atoms */
 283             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 284             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 285
 286             /* Analytical LJ-PME */
 287             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 288             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 289             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 290             exponent         = gmx_simd_exp_d(ewcljrsq);
 291             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 292             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 293             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 294             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 295             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 296             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 297                                _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));
 298             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 299             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);
 300
 301             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 302
 303             /* Update potential sum for this i atom from the interaction with this j atom. */
 304             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 305             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 306             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 307
 308             fscal            = fvdw;
 309
 310             fscal            = _mm_and_pd(fscal,cutoff_mask);
 311
 312             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 313
 314             /* Update vectorial force */
 315             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 316             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 317             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 318
 319             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 320                                                    _mm_mul_pd(dx00,fscal),
 321                                                    _mm_mul_pd(dy00,fscal),
 322                                                    _mm_mul_pd(dz00,fscal));
 323
 324             }
 325
 326             /* Inner loop uses 58 flops */
 327         }
 328
 329         /* End of innermost loop */
 330
 331         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 332                                               f+i_coord_offset,fshift+i_shift_offset);
 333
 334         ggid                        = gid[iidx];
 335         /* Update potential energies */
 336         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 337
 338         /* Increment number of inner iterations */
 339         inneriter                  += j_index_end - j_index_start;
 340
 341         /* Outer loop uses 7 flops */
 342     }
 343
 344     /* Increment number of outer iterations */
 345     outeriter        += nri;
 346
 347     /* Update outer/inner flops */
 348
 349     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*58);
 350 }
 351 /*
 352  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_F_avx_128_fma_double
 353  * Electrostatics interaction: None
 354  * VdW interaction:            LJEwald
 355  * Geometry:                   Particle-Particle
 356  * Calculate force/pot:        Force
 357  */
 358 void
 359 nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_F_avx_128_fma_double
 360                     (t_nblist                    * gmx_restrict       nlist,
 361                      rvec                        * gmx_restrict          xx,
 362                      rvec                        * gmx_restrict          ff,
 363                      t_forcerec                  * gmx_restrict          fr,
 364                      t_mdatoms                   * gmx_restrict     mdatoms,
 365                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 366                      t_nrnb                      * gmx_restrict        nrnb)
 367 {
 368     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 369      * just 0 for non-waters.
 370      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 371      * jnr indices corresponding to data put in the four positions in the SIMD register.
 372      */
 373     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 374     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 375     int              jnrA,jnrB;
 376     int              j_coord_offsetA,j_coord_offsetB;
 377     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 378     real             rcutoff_scalar;
 379     real             *shiftvec,*fshift,*x,*f;
 380     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 381     int              vdwioffset0;
 382     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 383     int              vdwjidx0A,vdwjidx0B;
 384     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 385     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 386     int              nvdwtype;
 387     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 388     int              *vdwtype;
 389     real             *vdwparam;
 390     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 391     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 392     __m128d           c6grid_00;
 393     real             *vdwgridparam;
 394     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 395     __m128d           one_half  = _mm_set1_pd(0.5);
 396     __m128d           minus_one = _mm_set1_pd(-1.0);
 397     __m128d          dummy_mask,cutoff_mask;
 398     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 399     __m128d          one     = _mm_set1_pd(1.0);
 400     __m128d          two     = _mm_set1_pd(2.0);
 401     x                = xx[0];
 402     f                = ff[0];
 403
 404     nri              = nlist->nri;
 405     iinr             = nlist->iinr;
 406     jindex           = nlist->jindex;
 407     jjnr             = nlist->jjnr;
 408     shiftidx         = nlist->shift;
 409     gid              = nlist->gid;
 410     shiftvec         = fr->shift_vec[0];
 411     fshift           = fr->fshift[0];
 412     nvdwtype         = fr->ntype;
 413     vdwparam         = fr->nbfp;
 414     vdwtype          = mdatoms->typeA;
 415     vdwgridparam     = fr->ljpme_c6grid;
 416     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 417     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 418     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 419
 420     rcutoff_scalar   = fr->rvdw;
 421     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 422     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 423
 424     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 425     rvdw             = _mm_set1_pd(fr->rvdw);
 426
 427     /* Avoid stupid compiler warnings */
 428     jnrA = jnrB = 0;
 429     j_coord_offsetA = 0;
 430     j_coord_offsetB = 0;
 431
 432     outeriter        = 0;
 433     inneriter        = 0;
 434
 435     /* Start outer loop over neighborlists */
 436     for(iidx=0; iidx<nri; iidx++)
 437     {
 438         /* Load shift vector for this list */
 439         i_shift_offset   = DIM*shiftidx[iidx];
 440
 441         /* Load limits for loop over neighbors */
 442         j_index_start    = jindex[iidx];
 443         j_index_end      = jindex[iidx+1];
 444
 445         /* Get outer coordinate index */
 446         inr              = iinr[iidx];
 447         i_coord_offset   = DIM*inr;
 448
 449         /* Load i particle coords and add shift vector */
 450         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 451
 452         fix0             = _mm_setzero_pd();
 453         fiy0             = _mm_setzero_pd();
 454         fiz0             = _mm_setzero_pd();
 455
 456         /* Load parameters for i particles */
 457         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 458
 459         /* Start inner kernel loop */
 460         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 461         {
 462
 463             /* Get j neighbor index, and coordinate index */
 464             jnrA             = jjnr[jidx];
 465             jnrB             = jjnr[jidx+1];
 466             j_coord_offsetA  = DIM*jnrA;
 467             j_coord_offsetB  = DIM*jnrB;
 468
 469             /* load j atom coordinates */
 470             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 471                                               &jx0,&jy0,&jz0);
 472
 473             /* Calculate displacement vector */
 474             dx00             = _mm_sub_pd(ix0,jx0);
 475             dy00             = _mm_sub_pd(iy0,jy0);
 476             dz00             = _mm_sub_pd(iz0,jz0);
 477
 478             /* Calculate squared distance and things based on it */
 479             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 480
 481             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 482
 483             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 484
 485             /* Load parameters for j particles */
 486             vdwjidx0A        = 2*vdwtype[jnrA+0];
 487             vdwjidx0B        = 2*vdwtype[jnrB+0];
 488
 489             /**************************
 490              * CALCULATE INTERACTIONS *
 491              **************************/
 492
 493             if (gmx_mm_any_lt(rsq00,rcutoff2))
 494             {
 495
 496             r00              = _mm_mul_pd(rsq00,rinv00);
 497
 498             /* Compute parameters for interactions between i and j atoms */
 499             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 500                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 501             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 502                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 503
 504             /* Analytical LJ-PME */
 505             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 506             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 507             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 508             exponent         = gmx_simd_exp_d(ewcljrsq);
 509             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 510             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 511             /* f6A = 6 * C6grid * (1 - poly) */
 512             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 513             /* f6B = C6grid * exponent * beta^6 */
 514             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 515             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 516             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 517
 518             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 519
 520             fscal            = fvdw;
 521
 522             fscal            = _mm_and_pd(fscal,cutoff_mask);
 523
 524             /* Update vectorial force */
 525             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 526             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 527             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 528
 529             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 530                                                    _mm_mul_pd(dx00,fscal),
 531                                                    _mm_mul_pd(dy00,fscal),
 532                                                    _mm_mul_pd(dz00,fscal));
 533
 534             }
 535
 536             /* Inner loop uses 50 flops */
 537         }
 538
 539         if(jidx<j_index_end)
 540         {
 541
 542             jnrA             = jjnr[jidx];
 543             j_coord_offsetA  = DIM*jnrA;
 544
 545             /* load j atom coordinates */
 546             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 547                                               &jx0,&jy0,&jz0);
 548
 549             /* Calculate displacement vector */
 550             dx00             = _mm_sub_pd(ix0,jx0);
 551             dy00             = _mm_sub_pd(iy0,jy0);
 552             dz00             = _mm_sub_pd(iz0,jz0);
 553
 554             /* Calculate squared distance and things based on it */
 555             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 556
 557             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 558
 559             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 560
 561             /* Load parameters for j particles */
 562             vdwjidx0A        = 2*vdwtype[jnrA+0];
 563
 564             /**************************
 565              * CALCULATE INTERACTIONS *
 566              **************************/
 567
 568             if (gmx_mm_any_lt(rsq00,rcutoff2))
 569             {
 570
 571             r00              = _mm_mul_pd(rsq00,rinv00);
 572
 573             /* Compute parameters for interactions between i and j atoms */
 574             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 575             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 576
 577             /* Analytical LJ-PME */
 578             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 579             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 580             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 581             exponent         = gmx_simd_exp_d(ewcljrsq);
 582             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 583             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 584             /* f6A = 6 * C6grid * (1 - poly) */
 585             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 586             /* f6B = C6grid * exponent * beta^6 */
 587             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 588             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 589             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 590
 591             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 592
 593             fscal            = fvdw;
 594
 595             fscal            = _mm_and_pd(fscal,cutoff_mask);
 596
 597             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 598
 599             /* Update vectorial force */
 600             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 601             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 602             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 603
 604             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 605                                                    _mm_mul_pd(dx00,fscal),
 606                                                    _mm_mul_pd(dy00,fscal),
 607                                                    _mm_mul_pd(dz00,fscal));
 608
 609             }
 610
 611             /* Inner loop uses 50 flops */
 612         }
 613
 614         /* End of innermost loop */
 615
 616         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 617                                               f+i_coord_offset,fshift+i_shift_offset);
 618
 619         /* Increment number of inner iterations */
 620         inneriter                  += j_index_end - j_index_start;
 621
 622         /* Outer loop uses 6 flops */
 623     }
 624
 625     /* Increment number of outer iterations */
 626     outeriter        += nri;
 627
 628     /* Update outer/inner flops */
 629
 630     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*50);
 631 }