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 "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_VdwLJEw_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_VdwLJEw_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     /* Avoid stupid compiler warnings */
 122     jnrA = jnrB = 0;
 123     j_coord_offsetA = 0;
 124     j_coord_offsetB = 0;
 125
 126     outeriter        = 0;
 127     inneriter        = 0;
 128
 129     /* Start outer loop over neighborlists */
 130     for(iidx=0; iidx<nri; iidx++)
 131     {
 132         /* Load shift vector for this list */
 133         i_shift_offset   = DIM*shiftidx[iidx];
 134
 135         /* Load limits for loop over neighbors */
 136         j_index_start    = jindex[iidx];
 137         j_index_end      = jindex[iidx+1];
 138
 139         /* Get outer coordinate index */
 140         inr              = iinr[iidx];
 141         i_coord_offset   = DIM*inr;
 142
 143         /* Load i particle coords and add shift vector */
 144         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 145
 146         fix0             = _mm_setzero_pd();
 147         fiy0             = _mm_setzero_pd();
 148         fiz0             = _mm_setzero_pd();
 149
 150         /* Load parameters for i particles */
 151         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 152
 153         /* Reset potential sums */
 154         vvdwsum          = _mm_setzero_pd();
 155
 156         /* Start inner kernel loop */
 157         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 158         {
 159
 160             /* Get j neighbor index, and coordinate index */
 161             jnrA             = jjnr[jidx];
 162             jnrB             = jjnr[jidx+1];
 163             j_coord_offsetA  = DIM*jnrA;
 164             j_coord_offsetB  = DIM*jnrB;
 165
 166             /* load j atom coordinates */
 167             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 168                                               &jx0,&jy0,&jz0);
 169
 170             /* Calculate displacement vector */
 171             dx00             = _mm_sub_pd(ix0,jx0);
 172             dy00             = _mm_sub_pd(iy0,jy0);
 173             dz00             = _mm_sub_pd(iz0,jz0);
 174
 175             /* Calculate squared distance and things based on it */
 176             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 177
 178             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 179
 180             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 181
 182             /* Load parameters for j particles */
 183             vdwjidx0A        = 2*vdwtype[jnrA+0];
 184             vdwjidx0B        = 2*vdwtype[jnrB+0];
 185
 186             /**************************
 187              * CALCULATE INTERACTIONS *
 188              **************************/
 189
 190             r00              = _mm_mul_pd(rsq00,rinv00);
 191
 192             /* Compute parameters for interactions between i and j atoms */
 193             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 194                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 195             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 196                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 197
 198             /* Analytical LJ-PME */
 199             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 200             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 201             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 202             exponent         = gmx_simd_exp_d(ewcljrsq);
 203             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 204             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 205             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 206             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 207             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 208             vvdw             = _mm_msub_pd(vvdw12,one_twelfth,_mm_mul_pd(vvdw6,one_sixth));
 209             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 210             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);
 211
 212             /* Update potential sum for this i atom from the interaction with this j atom. */
 213             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 214
 215             fscal            = fvdw;
 216
 217             /* Update vectorial force */
 218             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 219             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 220             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 221
 222             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 223                                                    _mm_mul_pd(dx00,fscal),
 224                                                    _mm_mul_pd(dy00,fscal),
 225                                                    _mm_mul_pd(dz00,fscal));
 226
 227             /* Inner loop uses 50 flops */
 228         }
 229
 230         if(jidx<j_index_end)
 231         {
 232
 233             jnrA             = jjnr[jidx];
 234             j_coord_offsetA  = DIM*jnrA;
 235
 236             /* load j atom coordinates */
 237             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 238                                               &jx0,&jy0,&jz0);
 239
 240             /* Calculate displacement vector */
 241             dx00             = _mm_sub_pd(ix0,jx0);
 242             dy00             = _mm_sub_pd(iy0,jy0);
 243             dz00             = _mm_sub_pd(iz0,jz0);
 244
 245             /* Calculate squared distance and things based on it */
 246             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 247
 248             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 249
 250             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 251
 252             /* Load parameters for j particles */
 253             vdwjidx0A        = 2*vdwtype[jnrA+0];
 254
 255             /**************************
 256              * CALCULATE INTERACTIONS *
 257              **************************/
 258
 259             r00              = _mm_mul_pd(rsq00,rinv00);
 260
 261             /* Compute parameters for interactions between i and j atoms */
 262             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 263             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 264
 265             /* Analytical LJ-PME */
 266             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 267             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 268             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 269             exponent         = gmx_simd_exp_d(ewcljrsq);
 270             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 271             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 272             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 273             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 274             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 275             vvdw             = _mm_msub_pd(vvdw12,one_twelfth,_mm_mul_pd(vvdw6,one_sixth));
 276             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 277             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);
 278
 279             /* Update potential sum for this i atom from the interaction with this j atom. */
 280             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 281             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 282
 283             fscal            = fvdw;
 284
 285             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 286
 287             /* Update vectorial force */
 288             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 289             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 290             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 291
 292             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 293                                                    _mm_mul_pd(dx00,fscal),
 294                                                    _mm_mul_pd(dy00,fscal),
 295                                                    _mm_mul_pd(dz00,fscal));
 296
 297             /* Inner loop uses 50 flops */
 298         }
 299
 300         /* End of innermost loop */
 301
 302         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 303                                               f+i_coord_offset,fshift+i_shift_offset);
 304
 305         ggid                        = gid[iidx];
 306         /* Update potential energies */
 307         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 308
 309         /* Increment number of inner iterations */
 310         inneriter                  += j_index_end - j_index_start;
 311
 312         /* Outer loop uses 7 flops */
 313     }
 314
 315     /* Increment number of outer iterations */
 316     outeriter        += nri;
 317
 318     /* Update outer/inner flops */
 319
 320     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*50);
 321 }
 322 /*
 323  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEw_GeomP1P1_F_avx_128_fma_double
 324  * Electrostatics interaction: None
 325  * VdW interaction:            LJEwald
 326  * Geometry:                   Particle-Particle
 327  * Calculate force/pot:        Force
 328  */
 329 void
 330 nb_kernel_ElecNone_VdwLJEw_GeomP1P1_F_avx_128_fma_double
 331                     (t_nblist                    * gmx_restrict       nlist,
 332                      rvec                        * gmx_restrict          xx,
 333                      rvec                        * gmx_restrict          ff,
 334                      t_forcerec                  * gmx_restrict          fr,
 335                      t_mdatoms                   * gmx_restrict     mdatoms,
 336                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 337                      t_nrnb                      * gmx_restrict        nrnb)
 338 {
 339     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 340      * just 0 for non-waters.
 341      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 342      * jnr indices corresponding to data put in the four positions in the SIMD register.
 343      */
 344     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 345     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 346     int              jnrA,jnrB;
 347     int              j_coord_offsetA,j_coord_offsetB;
 348     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 349     real             rcutoff_scalar;
 350     real             *shiftvec,*fshift,*x,*f;
 351     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 352     int              vdwioffset0;
 353     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 354     int              vdwjidx0A,vdwjidx0B;
 355     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 356     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 357     int              nvdwtype;
 358     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 359     int              *vdwtype;
 360     real             *vdwparam;
 361     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 362     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 363     __m128d           c6grid_00;
 364     real             *vdwgridparam;
 365     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 366     __m128d           one_half  = _mm_set1_pd(0.5);
 367     __m128d           minus_one = _mm_set1_pd(-1.0);
 368     __m128d          dummy_mask,cutoff_mask;
 369     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 370     __m128d          one     = _mm_set1_pd(1.0);
 371     __m128d          two     = _mm_set1_pd(2.0);
 372     x                = xx[0];
 373     f                = ff[0];
 374
 375     nri              = nlist->nri;
 376     iinr             = nlist->iinr;
 377     jindex           = nlist->jindex;
 378     jjnr             = nlist->jjnr;
 379     shiftidx         = nlist->shift;
 380     gid              = nlist->gid;
 381     shiftvec         = fr->shift_vec[0];
 382     fshift           = fr->fshift[0];
 383     nvdwtype         = fr->ntype;
 384     vdwparam         = fr->nbfp;
 385     vdwtype          = mdatoms->typeA;
 386     vdwgridparam     = fr->ljpme_c6grid;
 387     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 388     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 389     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 390
 391     /* Avoid stupid compiler warnings */
 392     jnrA = jnrB = 0;
 393     j_coord_offsetA = 0;
 394     j_coord_offsetB = 0;
 395
 396     outeriter        = 0;
 397     inneriter        = 0;
 398
 399     /* Start outer loop over neighborlists */
 400     for(iidx=0; iidx<nri; iidx++)
 401     {
 402         /* Load shift vector for this list */
 403         i_shift_offset   = DIM*shiftidx[iidx];
 404
 405         /* Load limits for loop over neighbors */
 406         j_index_start    = jindex[iidx];
 407         j_index_end      = jindex[iidx+1];
 408
 409         /* Get outer coordinate index */
 410         inr              = iinr[iidx];
 411         i_coord_offset   = DIM*inr;
 412
 413         /* Load i particle coords and add shift vector */
 414         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 415
 416         fix0             = _mm_setzero_pd();
 417         fiy0             = _mm_setzero_pd();
 418         fiz0             = _mm_setzero_pd();
 419
 420         /* Load parameters for i particles */
 421         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 422
 423         /* Start inner kernel loop */
 424         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 425         {
 426
 427             /* Get j neighbor index, and coordinate index */
 428             jnrA             = jjnr[jidx];
 429             jnrB             = jjnr[jidx+1];
 430             j_coord_offsetA  = DIM*jnrA;
 431             j_coord_offsetB  = DIM*jnrB;
 432
 433             /* load j atom coordinates */
 434             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 435                                               &jx0,&jy0,&jz0);
 436
 437             /* Calculate displacement vector */
 438             dx00             = _mm_sub_pd(ix0,jx0);
 439             dy00             = _mm_sub_pd(iy0,jy0);
 440             dz00             = _mm_sub_pd(iz0,jz0);
 441
 442             /* Calculate squared distance and things based on it */
 443             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 444
 445             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 446
 447             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 448
 449             /* Load parameters for j particles */
 450             vdwjidx0A        = 2*vdwtype[jnrA+0];
 451             vdwjidx0B        = 2*vdwtype[jnrB+0];
 452
 453             /**************************
 454              * CALCULATE INTERACTIONS *
 455              **************************/
 456
 457             r00              = _mm_mul_pd(rsq00,rinv00);
 458
 459             /* Compute parameters for interactions between i and j atoms */
 460             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 461                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 462             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 463                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 464
 465             /* Analytical LJ-PME */
 466             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 467             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 468             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 469             exponent         = gmx_simd_exp_d(ewcljrsq);
 470             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 471             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 472             /* f6A = 6 * C6grid * (1 - poly) */
 473             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 474             /* f6B = C6grid * exponent * beta^6 */
 475             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 476             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 477             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 478
 479             fscal            = fvdw;
 480
 481             /* Update vectorial force */
 482             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 483             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 484             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 485
 486             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 487                                                    _mm_mul_pd(dx00,fscal),
 488                                                    _mm_mul_pd(dy00,fscal),
 489                                                    _mm_mul_pd(dz00,fscal));
 490
 491             /* Inner loop uses 47 flops */
 492         }
 493
 494         if(jidx<j_index_end)
 495         {
 496
 497             jnrA             = jjnr[jidx];
 498             j_coord_offsetA  = DIM*jnrA;
 499
 500             /* load j atom coordinates */
 501             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 502                                               &jx0,&jy0,&jz0);
 503
 504             /* Calculate displacement vector */
 505             dx00             = _mm_sub_pd(ix0,jx0);
 506             dy00             = _mm_sub_pd(iy0,jy0);
 507             dz00             = _mm_sub_pd(iz0,jz0);
 508
 509             /* Calculate squared distance and things based on it */
 510             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 511
 512             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 513
 514             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 515
 516             /* Load parameters for j particles */
 517             vdwjidx0A        = 2*vdwtype[jnrA+0];
 518
 519             /**************************
 520              * CALCULATE INTERACTIONS *
 521              **************************/
 522
 523             r00              = _mm_mul_pd(rsq00,rinv00);
 524
 525             /* Compute parameters for interactions between i and j atoms */
 526             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 527             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 528
 529             /* Analytical LJ-PME */
 530             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 531             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 532             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 533             exponent         = gmx_simd_exp_d(ewcljrsq);
 534             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 535             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 536             /* f6A = 6 * C6grid * (1 - poly) */
 537             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 538             /* f6B = C6grid * exponent * beta^6 */
 539             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 540             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 541             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 542
 543             fscal            = fvdw;
 544
 545             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 546
 547             /* Update vectorial force */
 548             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 549             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 550             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 551
 552             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 553                                                    _mm_mul_pd(dx00,fscal),
 554                                                    _mm_mul_pd(dy00,fscal),
 555                                                    _mm_mul_pd(dz00,fscal));
 556
 557             /* Inner loop uses 47 flops */
 558         }
 559
 560         /* End of innermost loop */
 561
 562         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 563                                               f+i_coord_offset,fshift+i_shift_offset);
 564
 565         /* Increment number of inner iterations */
 566         inneriter                  += j_index_end - j_index_start;
 567
 568         /* Outer loop uses 6 flops */
 569     }
 570
 571     /* Increment number of outer iterations */
 572     outeriter        += nri;
 573
 574     /* Update outer/inner flops */
 575
 576     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*47);
 577 }