src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_sse2_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_VF_sse2_double
  51  * Electrostatics interaction: CubicSplineTable
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_VF_sse2_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwjidx0A,vdwjidx0B;
  82     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  83     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  84     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  85     real             *charge;
  86     int              nvdwtype;
  87     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  88     int              *vdwtype;
  89     real             *vdwparam;
  90     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  91     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  92     __m128i          vfitab;
  93     __m128i          ifour       = _mm_set1_epi32(4);
  94     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  95     real             *vftab;
  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     facel            = _mm_set1_pd(fr->ic->epsfac);
 112     charge           = mdatoms->chargeA;
 113     nvdwtype         = fr->ntype;
 114     vdwparam         = fr->nbfp;
 115     vdwtype          = mdatoms->typeA;
 116
 117     vftab            = kernel_data->table_elec->data;
 118     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 119
 120     /* Avoid stupid compiler warnings */
 121     jnrA = jnrB = 0;
 122     j_coord_offsetA = 0;
 123     j_coord_offsetB = 0;
 124
 125     outeriter        = 0;
 126     inneriter        = 0;
 127
 128     /* Start outer loop over neighborlists */
 129     for(iidx=0; iidx<nri; iidx++)
 130     {
 131         /* Load shift vector for this list */
 132         i_shift_offset   = DIM*shiftidx[iidx];
 133
 134         /* Load limits for loop over neighbors */
 135         j_index_start    = jindex[iidx];
 136         j_index_end      = jindex[iidx+1];
 137
 138         /* Get outer coordinate index */
 139         inr              = iinr[iidx];
 140         i_coord_offset   = DIM*inr;
 141
 142         /* Load i particle coords and add shift vector */
 143         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 144
 145         fix0             = _mm_setzero_pd();
 146         fiy0             = _mm_setzero_pd();
 147         fiz0             = _mm_setzero_pd();
 148
 149         /* Load parameters for i particles */
 150         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 151         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 152
 153         /* Reset potential sums */
 154         velecsum         = _mm_setzero_pd();
 155         vvdwsum          = _mm_setzero_pd();
 156
 157         /* Start inner kernel loop */
 158         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 159         {
 160
 161             /* Get j neighbor index, and coordinate index */
 162             jnrA             = jjnr[jidx];
 163             jnrB             = jjnr[jidx+1];
 164             j_coord_offsetA  = DIM*jnrA;
 165             j_coord_offsetB  = DIM*jnrB;
 166
 167             /* load j atom coordinates */
 168             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 169                                               &jx0,&jy0,&jz0);
 170
 171             /* Calculate displacement vector */
 172             dx00             = _mm_sub_pd(ix0,jx0);
 173             dy00             = _mm_sub_pd(iy0,jy0);
 174             dz00             = _mm_sub_pd(iz0,jz0);
 175
 176             /* Calculate squared distance and things based on it */
 177             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 178
 179             rinv00           = sse2_invsqrt_d(rsq00);
 180
 181             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 182
 183             /* Load parameters for j particles */
 184             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 185             vdwjidx0A        = 2*vdwtype[jnrA+0];
 186             vdwjidx0B        = 2*vdwtype[jnrB+0];
 187
 188             /**************************
 189              * CALCULATE INTERACTIONS *
 190              **************************/
 191
 192             r00              = _mm_mul_pd(rsq00,rinv00);
 193
 194             /* Compute parameters for interactions between i and j atoms */
 195             qq00             = _mm_mul_pd(iq0,jq0);
 196             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 197                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 198
 199             /* Calculate table index by multiplying r with table scale and truncate to integer */
 200             rt               = _mm_mul_pd(r00,vftabscale);
 201             vfitab           = _mm_cvttpd_epi32(rt);
 202             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 203             vfitab           = _mm_slli_epi32(vfitab,2);
 204
 205             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 206             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 207             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 208             GMX_MM_TRANSPOSE2_PD(Y,F);
 209             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 210             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 211             GMX_MM_TRANSPOSE2_PD(G,H);
 212             Heps             = _mm_mul_pd(vfeps,H);
 213             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 214             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 215             velec            = _mm_mul_pd(qq00,VV);
 216             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 217             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 218
 219             /* LENNARD-JONES DISPERSION/REPULSION */
 220
 221             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 222             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 223             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 224             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 225             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 226
 227             /* Update potential sum for this i atom from the interaction with this j atom. */
 228             velecsum         = _mm_add_pd(velecsum,velec);
 229             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 230
 231             fscal            = _mm_add_pd(felec,fvdw);
 232
 233             /* Calculate temporary vectorial force */
 234             tx               = _mm_mul_pd(fscal,dx00);
 235             ty               = _mm_mul_pd(fscal,dy00);
 236             tz               = _mm_mul_pd(fscal,dz00);
 237
 238             /* Update vectorial force */
 239             fix0             = _mm_add_pd(fix0,tx);
 240             fiy0             = _mm_add_pd(fiy0,ty);
 241             fiz0             = _mm_add_pd(fiz0,tz);
 242
 243             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 244
 245             /* Inner loop uses 56 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           = sse2_invsqrt_d(rsq00);
 267
 268             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 269
 270             /* Load parameters for j particles */
 271             jq0              = _mm_load_sd(charge+jnrA+0);
 272             vdwjidx0A        = 2*vdwtype[jnrA+0];
 273
 274             /**************************
 275              * CALCULATE INTERACTIONS *
 276              **************************/
 277
 278             r00              = _mm_mul_pd(rsq00,rinv00);
 279
 280             /* Compute parameters for interactions between i and j atoms */
 281             qq00             = _mm_mul_pd(iq0,jq0);
 282             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 283
 284             /* Calculate table index by multiplying r with table scale and truncate to integer */
 285             rt               = _mm_mul_pd(r00,vftabscale);
 286             vfitab           = _mm_cvttpd_epi32(rt);
 287             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 288             vfitab           = _mm_slli_epi32(vfitab,2);
 289
 290             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 291             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 292             F                = _mm_setzero_pd();
 293             GMX_MM_TRANSPOSE2_PD(Y,F);
 294             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 295             H                = _mm_setzero_pd();
 296             GMX_MM_TRANSPOSE2_PD(G,H);
 297             Heps             = _mm_mul_pd(vfeps,H);
 298             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 299             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 300             velec            = _mm_mul_pd(qq00,VV);
 301             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 302             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 303
 304             /* LENNARD-JONES DISPERSION/REPULSION */
 305
 306             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 307             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 308             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 309             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 310             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 311
 312             /* Update potential sum for this i atom from the interaction with this j atom. */
 313             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 314             velecsum         = _mm_add_pd(velecsum,velec);
 315             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 316             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 317
 318             fscal            = _mm_add_pd(felec,fvdw);
 319
 320             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 321
 322             /* Calculate temporary vectorial force */
 323             tx               = _mm_mul_pd(fscal,dx00);
 324             ty               = _mm_mul_pd(fscal,dy00);
 325             tz               = _mm_mul_pd(fscal,dz00);
 326
 327             /* Update vectorial force */
 328             fix0             = _mm_add_pd(fix0,tx);
 329             fiy0             = _mm_add_pd(fiy0,ty);
 330             fiz0             = _mm_add_pd(fiz0,tz);
 331
 332             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 333
 334             /* Inner loop uses 56 flops */
 335         }
 336
 337         /* End of innermost loop */
 338
 339         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 340                                               f+i_coord_offset,fshift+i_shift_offset);
 341
 342         ggid                        = gid[iidx];
 343         /* Update potential energies */
 344         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 345         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 346
 347         /* Increment number of inner iterations */
 348         inneriter                  += j_index_end - j_index_start;
 349
 350         /* Outer loop uses 9 flops */
 351     }
 352
 353     /* Increment number of outer iterations */
 354     outeriter        += nri;
 355
 356     /* Update outer/inner flops */
 357
 358     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*56);
 359 }
 360 /*
 361  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_sse2_double
 362  * Electrostatics interaction: CubicSplineTable
 363  * VdW interaction:            LennardJones
 364  * Geometry:                   Particle-Particle
 365  * Calculate force/pot:        Force
 366  */
 367 void
 368 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_sse2_double
 369                     (t_nblist                    * gmx_restrict       nlist,
 370                      rvec                        * gmx_restrict          xx,
 371                      rvec                        * gmx_restrict          ff,
 372                      struct t_forcerec           * gmx_restrict          fr,
 373                      t_mdatoms                   * gmx_restrict     mdatoms,
 374                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 375                      t_nrnb                      * gmx_restrict        nrnb)
 376 {
 377     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 378      * just 0 for non-waters.
 379      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 380      * jnr indices corresponding to data put in the four positions in the SIMD register.
 381      */
 382     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 383     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 384     int              jnrA,jnrB;
 385     int              j_coord_offsetA,j_coord_offsetB;
 386     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 387     real             rcutoff_scalar;
 388     real             *shiftvec,*fshift,*x,*f;
 389     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 390     int              vdwioffset0;
 391     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 392     int              vdwjidx0A,vdwjidx0B;
 393     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 394     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 395     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 396     real             *charge;
 397     int              nvdwtype;
 398     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 399     int              *vdwtype;
 400     real             *vdwparam;
 401     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 402     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 403     __m128i          vfitab;
 404     __m128i          ifour       = _mm_set1_epi32(4);
 405     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 406     real             *vftab;
 407     __m128d          dummy_mask,cutoff_mask;
 408     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 409     __m128d          one     = _mm_set1_pd(1.0);
 410     __m128d          two     = _mm_set1_pd(2.0);
 411     x                = xx[0];
 412     f                = ff[0];
 413
 414     nri              = nlist->nri;
 415     iinr             = nlist->iinr;
 416     jindex           = nlist->jindex;
 417     jjnr             = nlist->jjnr;
 418     shiftidx         = nlist->shift;
 419     gid              = nlist->gid;
 420     shiftvec         = fr->shift_vec[0];
 421     fshift           = fr->fshift[0];
 422     facel            = _mm_set1_pd(fr->ic->epsfac);
 423     charge           = mdatoms->chargeA;
 424     nvdwtype         = fr->ntype;
 425     vdwparam         = fr->nbfp;
 426     vdwtype          = mdatoms->typeA;
 427
 428     vftab            = kernel_data->table_elec->data;
 429     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 430
 431     /* Avoid stupid compiler warnings */
 432     jnrA = jnrB = 0;
 433     j_coord_offsetA = 0;
 434     j_coord_offsetB = 0;
 435
 436     outeriter        = 0;
 437     inneriter        = 0;
 438
 439     /* Start outer loop over neighborlists */
 440     for(iidx=0; iidx<nri; iidx++)
 441     {
 442         /* Load shift vector for this list */
 443         i_shift_offset   = DIM*shiftidx[iidx];
 444
 445         /* Load limits for loop over neighbors */
 446         j_index_start    = jindex[iidx];
 447         j_index_end      = jindex[iidx+1];
 448
 449         /* Get outer coordinate index */
 450         inr              = iinr[iidx];
 451         i_coord_offset   = DIM*inr;
 452
 453         /* Load i particle coords and add shift vector */
 454         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 455
 456         fix0             = _mm_setzero_pd();
 457         fiy0             = _mm_setzero_pd();
 458         fiz0             = _mm_setzero_pd();
 459
 460         /* Load parameters for i particles */
 461         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 462         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 463
 464         /* Start inner kernel loop */
 465         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 466         {
 467
 468             /* Get j neighbor index, and coordinate index */
 469             jnrA             = jjnr[jidx];
 470             jnrB             = jjnr[jidx+1];
 471             j_coord_offsetA  = DIM*jnrA;
 472             j_coord_offsetB  = DIM*jnrB;
 473
 474             /* load j atom coordinates */
 475             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 476                                               &jx0,&jy0,&jz0);
 477
 478             /* Calculate displacement vector */
 479             dx00             = _mm_sub_pd(ix0,jx0);
 480             dy00             = _mm_sub_pd(iy0,jy0);
 481             dz00             = _mm_sub_pd(iz0,jz0);
 482
 483             /* Calculate squared distance and things based on it */
 484             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 485
 486             rinv00           = sse2_invsqrt_d(rsq00);
 487
 488             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 489
 490             /* Load parameters for j particles */
 491             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 492             vdwjidx0A        = 2*vdwtype[jnrA+0];
 493             vdwjidx0B        = 2*vdwtype[jnrB+0];
 494
 495             /**************************
 496              * CALCULATE INTERACTIONS *
 497              **************************/
 498
 499             r00              = _mm_mul_pd(rsq00,rinv00);
 500
 501             /* Compute parameters for interactions between i and j atoms */
 502             qq00             = _mm_mul_pd(iq0,jq0);
 503             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 504                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 505
 506             /* Calculate table index by multiplying r with table scale and truncate to integer */
 507             rt               = _mm_mul_pd(r00,vftabscale);
 508             vfitab           = _mm_cvttpd_epi32(rt);
 509             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 510             vfitab           = _mm_slli_epi32(vfitab,2);
 511
 512             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 513             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 514             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 515             GMX_MM_TRANSPOSE2_PD(Y,F);
 516             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 517             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 518             GMX_MM_TRANSPOSE2_PD(G,H);
 519             Heps             = _mm_mul_pd(vfeps,H);
 520             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 521             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 522             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 523
 524             /* LENNARD-JONES DISPERSION/REPULSION */
 525
 526             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 527             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 528
 529             fscal            = _mm_add_pd(felec,fvdw);
 530
 531             /* Calculate temporary vectorial force */
 532             tx               = _mm_mul_pd(fscal,dx00);
 533             ty               = _mm_mul_pd(fscal,dy00);
 534             tz               = _mm_mul_pd(fscal,dz00);
 535
 536             /* Update vectorial force */
 537             fix0             = _mm_add_pd(fix0,tx);
 538             fiy0             = _mm_add_pd(fiy0,ty);
 539             fiz0             = _mm_add_pd(fiz0,tz);
 540
 541             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 542
 543             /* Inner loop uses 47 flops */
 544         }
 545
 546         if(jidx<j_index_end)
 547         {
 548
 549             jnrA             = jjnr[jidx];
 550             j_coord_offsetA  = DIM*jnrA;
 551
 552             /* load j atom coordinates */
 553             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 554                                               &jx0,&jy0,&jz0);
 555
 556             /* Calculate displacement vector */
 557             dx00             = _mm_sub_pd(ix0,jx0);
 558             dy00             = _mm_sub_pd(iy0,jy0);
 559             dz00             = _mm_sub_pd(iz0,jz0);
 560
 561             /* Calculate squared distance and things based on it */
 562             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 563
 564             rinv00           = sse2_invsqrt_d(rsq00);
 565
 566             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 567
 568             /* Load parameters for j particles */
 569             jq0              = _mm_load_sd(charge+jnrA+0);
 570             vdwjidx0A        = 2*vdwtype[jnrA+0];
 571
 572             /**************************
 573              * CALCULATE INTERACTIONS *
 574              **************************/
 575
 576             r00              = _mm_mul_pd(rsq00,rinv00);
 577
 578             /* Compute parameters for interactions between i and j atoms */
 579             qq00             = _mm_mul_pd(iq0,jq0);
 580             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 581
 582             /* Calculate table index by multiplying r with table scale and truncate to integer */
 583             rt               = _mm_mul_pd(r00,vftabscale);
 584             vfitab           = _mm_cvttpd_epi32(rt);
 585             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 586             vfitab           = _mm_slli_epi32(vfitab,2);
 587
 588             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 589             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 590             F                = _mm_setzero_pd();
 591             GMX_MM_TRANSPOSE2_PD(Y,F);
 592             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 593             H                = _mm_setzero_pd();
 594             GMX_MM_TRANSPOSE2_PD(G,H);
 595             Heps             = _mm_mul_pd(vfeps,H);
 596             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 597             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 598             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 599
 600             /* LENNARD-JONES DISPERSION/REPULSION */
 601
 602             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 603             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 604
 605             fscal            = _mm_add_pd(felec,fvdw);
 606
 607             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 608
 609             /* Calculate temporary vectorial force */
 610             tx               = _mm_mul_pd(fscal,dx00);
 611             ty               = _mm_mul_pd(fscal,dy00);
 612             tz               = _mm_mul_pd(fscal,dz00);
 613
 614             /* Update vectorial force */
 615             fix0             = _mm_add_pd(fix0,tx);
 616             fiy0             = _mm_add_pd(fiy0,ty);
 617             fiz0             = _mm_add_pd(fiz0,tz);
 618
 619             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 620
 621             /* Inner loop uses 47 flops */
 622         }
 623
 624         /* End of innermost loop */
 625
 626         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 627                                               f+i_coord_offset,fshift+i_shift_offset);
 628
 629         /* Increment number of inner iterations */
 630         inneriter                  += j_index_end - j_index_start;
 631
 632         /* Outer loop uses 7 flops */
 633     }
 634
 635     /* Increment number of outer iterations */
 636     outeriter        += nri;
 637
 638     /* Update outer/inner flops */
 639
 640     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*47);
 641 }