src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_sparc64_hpc_ace_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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 sparc64_hpc_ace_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "kernelutil_sparc64_hpc_ace_double.h"
  50
  51 /*
  52  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_VF_sparc64_hpc_ace_double
  53  * Electrostatics interaction: CubicSplineTable
  54  * VdW interaction:            LennardJones
  55  * Geometry:                   Particle-Particle
  56  * Calculate force/pot:        PotentialAndForce
  57  */
  58 void
  59 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_VF_sparc64_hpc_ace_double
  60                     (t_nblist                    * gmx_restrict       nlist,
  61                      rvec                        * gmx_restrict          xx,
  62                      rvec                        * gmx_restrict          ff,
  63                      t_forcerec                  * gmx_restrict          fr,
  64                      t_mdatoms                   * gmx_restrict     mdatoms,
  65                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  66                      t_nrnb                      * gmx_restrict        nrnb)
  67 {
  68     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  69      * just 0 for non-waters.
  70      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
  71      * jnr indices corresponding to data put in the four positions in the SIMD register.
  72      */
  73     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  74     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  75     int              jnrA,jnrB;
  76     int              j_coord_offsetA,j_coord_offsetB;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  81     int              vdwioffset0;
  82     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  83     int              vdwjidx0A,vdwjidx0B;
  84     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  85     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  86     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
  87     real             *charge;
  88     int              nvdwtype;
  89     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  90     int              *vdwtype;
  91     real             *vdwparam;
  92     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
  93     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
  94     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  95     real             *vftab;
  96     _fjsp_v2r8       itab_tmp;
  97     _fjsp_v2r8       dummy_mask,cutoff_mask;
  98     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
  99     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 100     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 101
 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     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115     nvdwtype         = fr->ntype;
 116     vdwparam         = fr->nbfp;
 117     vdwtype          = mdatoms->typeA;
 118
 119     vftab            = kernel_data->table_elec->data;
 120     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_elec->scale);
 121
 122     /* Avoid stupid compiler warnings */
 123     jnrA = jnrB = 0;
 124     j_coord_offsetA = 0;
 125     j_coord_offsetB = 0;
 126
 127     outeriter        = 0;
 128     inneriter        = 0;
 129
 130     /* Start outer loop over neighborlists */
 131     for(iidx=0; iidx<nri; iidx++)
 132     {
 133         /* Load shift vector for this list */
 134         i_shift_offset   = DIM*shiftidx[iidx];
 135
 136         /* Load limits for loop over neighbors */
 137         j_index_start    = jindex[iidx];
 138         j_index_end      = jindex[iidx+1];
 139
 140         /* Get outer coordinate index */
 141         inr              = iinr[iidx];
 142         i_coord_offset   = DIM*inr;
 143
 144         /* Load i particle coords and add shift vector */
 145         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 146
 147         fix0             = _fjsp_setzero_v2r8();
 148         fiy0             = _fjsp_setzero_v2r8();
 149         fiz0             = _fjsp_setzero_v2r8();
 150
 151         /* Load parameters for i particles */
 152         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 153         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 154
 155         /* Reset potential sums */
 156         velecsum         = _fjsp_setzero_v2r8();
 157         vvdwsum          = _fjsp_setzero_v2r8();
 158
 159         /* Start inner kernel loop */
 160         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 161         {
 162
 163             /* Get j neighbor index, and coordinate index */
 164             jnrA             = jjnr[jidx];
 165             jnrB             = jjnr[jidx+1];
 166             j_coord_offsetA  = DIM*jnrA;
 167             j_coord_offsetB  = DIM*jnrB;
 168
 169             /* load j atom coordinates */
 170             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 171                                               &jx0,&jy0,&jz0);
 172
 173             /* Calculate displacement vector */
 174             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 175             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 176             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 177
 178             /* Calculate squared distance and things based on it */
 179             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 180
 181             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 182
 183             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 184
 185             /* Load parameters for j particles */
 186             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 187             vdwjidx0A        = 2*vdwtype[jnrA+0];
 188             vdwjidx0B        = 2*vdwtype[jnrB+0];
 189
 190             /**************************
 191              * CALCULATE INTERACTIONS *
 192              **************************/
 193
 194             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 195
 196             /* Compute parameters for interactions between i and j atoms */
 197             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 198             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 199                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 200
 201             /* Calculate table index by multiplying r with table scale and truncate to integer */
 202             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 203             itab_tmp         = _fjsp_dtox_v2r8(rt);
 204             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 205             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 206             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 207
 208             vfconv.i[0]     *= 4;
 209             vfconv.i[1]     *= 4;
 210
 211             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 212             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 213             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 214             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 215             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] +2);
 216             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] +2);
 217             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 218             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(vfeps,H,G),F);
 219             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 220             velec            = _fjsp_mul_v2r8(qq00,VV);
 221             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twovfeps,H,G),vfeps,Fp);
 222             felec            = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,FF),_fjsp_mul_v2r8(vftabscale,rinv00)));
 223
 224             /* LENNARD-JONES DISPERSION/REPULSION */
 225
 226             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 227             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 228             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 229             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 230             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 231
 232             /* Update potential sum for this i atom from the interaction with this j atom. */
 233             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 234             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 235
 236             fscal            = _fjsp_add_v2r8(felec,fvdw);
 237
 238             /* Update vectorial force */
 239             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 240             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 241             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 242
 243             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 244
 245             /* Inner loop uses 59 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_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 256                                               &jx0,&jy0,&jz0);
 257
 258             /* Calculate displacement vector */
 259             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 260             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 261             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 262
 263             /* Calculate squared distance and things based on it */
 264             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 265
 266             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 267
 268             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 269
 270             /* Load parameters for j particles */
 271             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 272             vdwjidx0A        = 2*vdwtype[jnrA+0];
 273
 274             /**************************
 275              * CALCULATE INTERACTIONS *
 276              **************************/
 277
 278             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 279
 280             /* Compute parameters for interactions between i and j atoms */
 281             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 282             gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 283
 284             /* Calculate table index by multiplying r with table scale and truncate to integer */
 285             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 286             itab_tmp         = _fjsp_dtox_v2r8(rt);
 287             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 288             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 289             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 290
 291             vfconv.i[0]     *= 4;
 292             vfconv.i[1]     *= 4;
 293
 294             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 295             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 296             F                = _fjsp_setzero_v2r8();
 297             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 298             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] +2);
 299             H                = _fjsp_setzero_v2r8();
 300             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 301             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(vfeps,H,G),F);
 302             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 303             velec            = _fjsp_mul_v2r8(qq00,VV);
 304             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twovfeps,H,G),vfeps,Fp);
 305             felec            = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,FF),_fjsp_mul_v2r8(vftabscale,rinv00)));
 306
 307             /* LENNARD-JONES DISPERSION/REPULSION */
 308
 309             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 310             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 311             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 312             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 313             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 314
 315             /* Update potential sum for this i atom from the interaction with this j atom. */
 316             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 317             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 318             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 319             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 320
 321             fscal            = _fjsp_add_v2r8(felec,fvdw);
 322
 323             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 324
 325             /* Update vectorial force */
 326             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 327             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 328             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 329
 330             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 331
 332             /* Inner loop uses 59 flops */
 333         }
 334
 335         /* End of innermost loop */
 336
 337         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 338                                               f+i_coord_offset,fshift+i_shift_offset);
 339
 340         ggid                        = gid[iidx];
 341         /* Update potential energies */
 342         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 343         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 344
 345         /* Increment number of inner iterations */
 346         inneriter                  += j_index_end - j_index_start;
 347
 348         /* Outer loop uses 9 flops */
 349     }
 350
 351     /* Increment number of outer iterations */
 352     outeriter        += nri;
 353
 354     /* Update outer/inner flops */
 355
 356     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*59);
 357 }
 358 /*
 359  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_sparc64_hpc_ace_double
 360  * Electrostatics interaction: CubicSplineTable
 361  * VdW interaction:            LennardJones
 362  * Geometry:                   Particle-Particle
 363  * Calculate force/pot:        Force
 364  */
 365 void
 366 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_sparc64_hpc_ace_double
 367                     (t_nblist                    * gmx_restrict       nlist,
 368                      rvec                        * gmx_restrict          xx,
 369                      rvec                        * gmx_restrict          ff,
 370                      t_forcerec                  * gmx_restrict          fr,
 371                      t_mdatoms                   * gmx_restrict     mdatoms,
 372                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 373                      t_nrnb                      * gmx_restrict        nrnb)
 374 {
 375     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 376      * just 0 for non-waters.
 377      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 378      * jnr indices corresponding to data put in the four positions in the SIMD register.
 379      */
 380     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 381     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 382     int              jnrA,jnrB;
 383     int              j_coord_offsetA,j_coord_offsetB;
 384     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 385     real             rcutoff_scalar;
 386     real             *shiftvec,*fshift,*x,*f;
 387     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 388     int              vdwioffset0;
 389     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 390     int              vdwjidx0A,vdwjidx0B;
 391     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 392     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 393     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 394     real             *charge;
 395     int              nvdwtype;
 396     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 397     int              *vdwtype;
 398     real             *vdwparam;
 399     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 400     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 401     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 402     real             *vftab;
 403     _fjsp_v2r8       itab_tmp;
 404     _fjsp_v2r8       dummy_mask,cutoff_mask;
 405     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 406     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 407     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 408
 409     x                = xx[0];
 410     f                = ff[0];
 411
 412     nri              = nlist->nri;
 413     iinr             = nlist->iinr;
 414     jindex           = nlist->jindex;
 415     jjnr             = nlist->jjnr;
 416     shiftidx         = nlist->shift;
 417     gid              = nlist->gid;
 418     shiftvec         = fr->shift_vec[0];
 419     fshift           = fr->fshift[0];
 420     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 421     charge           = mdatoms->chargeA;
 422     nvdwtype         = fr->ntype;
 423     vdwparam         = fr->nbfp;
 424     vdwtype          = mdatoms->typeA;
 425
 426     vftab            = kernel_data->table_elec->data;
 427     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_elec->scale);
 428
 429     /* Avoid stupid compiler warnings */
 430     jnrA = jnrB = 0;
 431     j_coord_offsetA = 0;
 432     j_coord_offsetB = 0;
 433
 434     outeriter        = 0;
 435     inneriter        = 0;
 436
 437     /* Start outer loop over neighborlists */
 438     for(iidx=0; iidx<nri; iidx++)
 439     {
 440         /* Load shift vector for this list */
 441         i_shift_offset   = DIM*shiftidx[iidx];
 442
 443         /* Load limits for loop over neighbors */
 444         j_index_start    = jindex[iidx];
 445         j_index_end      = jindex[iidx+1];
 446
 447         /* Get outer coordinate index */
 448         inr              = iinr[iidx];
 449         i_coord_offset   = DIM*inr;
 450
 451         /* Load i particle coords and add shift vector */
 452         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 453
 454         fix0             = _fjsp_setzero_v2r8();
 455         fiy0             = _fjsp_setzero_v2r8();
 456         fiz0             = _fjsp_setzero_v2r8();
 457
 458         /* Load parameters for i particles */
 459         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 460         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 461
 462         /* Start inner kernel loop */
 463         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 464         {
 465
 466             /* Get j neighbor index, and coordinate index */
 467             jnrA             = jjnr[jidx];
 468             jnrB             = jjnr[jidx+1];
 469             j_coord_offsetA  = DIM*jnrA;
 470             j_coord_offsetB  = DIM*jnrB;
 471
 472             /* load j atom coordinates */
 473             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 474                                               &jx0,&jy0,&jz0);
 475
 476             /* Calculate displacement vector */
 477             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 478             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 479             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 480
 481             /* Calculate squared distance and things based on it */
 482             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 483
 484             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 485
 486             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 487
 488             /* Load parameters for j particles */
 489             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 490             vdwjidx0A        = 2*vdwtype[jnrA+0];
 491             vdwjidx0B        = 2*vdwtype[jnrB+0];
 492
 493             /**************************
 494              * CALCULATE INTERACTIONS *
 495              **************************/
 496
 497             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 498
 499             /* Compute parameters for interactions between i and j atoms */
 500             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 501             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 502                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 503
 504             /* Calculate table index by multiplying r with table scale and truncate to integer */
 505             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 506             itab_tmp         = _fjsp_dtox_v2r8(rt);
 507             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 508             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 509             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 510
 511             vfconv.i[0]     *= 4;
 512             vfconv.i[1]     *= 4;
 513
 514             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 515             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 516             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 517             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 518             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] +2);
 519             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] +2);
 520             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 521             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(vfeps,H,G),F);
 522             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twovfeps,H,G),vfeps,Fp);
 523             felec            = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,FF),_fjsp_mul_v2r8(vftabscale,rinv00)));
 524
 525             /* LENNARD-JONES DISPERSION/REPULSION */
 526
 527             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 528             fvdw             = _fjsp_mul_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,c6_00),_fjsp_mul_v2r8(rinvsix,rinvsq00));
 529
 530             fscal            = _fjsp_add_v2r8(felec,fvdw);
 531
 532             /* Update vectorial force */
 533             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 534             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 535             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 536
 537             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 538
 539             /* Inner loop uses 50 flops */
 540         }
 541
 542         if(jidx<j_index_end)
 543         {
 544
 545             jnrA             = jjnr[jidx];
 546             j_coord_offsetA  = DIM*jnrA;
 547
 548             /* load j atom coordinates */
 549             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 550                                               &jx0,&jy0,&jz0);
 551
 552             /* Calculate displacement vector */
 553             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 554             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 555             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 556
 557             /* Calculate squared distance and things based on it */
 558             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 559
 560             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 561
 562             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 563
 564             /* Load parameters for j particles */
 565             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 566             vdwjidx0A        = 2*vdwtype[jnrA+0];
 567
 568             /**************************
 569              * CALCULATE INTERACTIONS *
 570              **************************/
 571
 572             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 573
 574             /* Compute parameters for interactions between i and j atoms */
 575             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 576             gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 577
 578             /* Calculate table index by multiplying r with table scale and truncate to integer */
 579             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 580             itab_tmp         = _fjsp_dtox_v2r8(rt);
 581             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 582             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 583             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 584
 585             vfconv.i[0]     *= 4;
 586             vfconv.i[1]     *= 4;
 587
 588             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 589             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 590             F                = _fjsp_setzero_v2r8();
 591             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 592             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] +2);
 593             H                = _fjsp_setzero_v2r8();
 594             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 595             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(vfeps,H,G),F);
 596             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twovfeps,H,G),vfeps,Fp);
 597             felec            = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,FF),_fjsp_mul_v2r8(vftabscale,rinv00)));
 598
 599             /* LENNARD-JONES DISPERSION/REPULSION */
 600
 601             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 602             fvdw             = _fjsp_mul_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,c6_00),_fjsp_mul_v2r8(rinvsix,rinvsq00));
 603
 604             fscal            = _fjsp_add_v2r8(felec,fvdw);
 605
 606             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 607
 608             /* Update vectorial force */
 609             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 610             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 611             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 612
 613             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 614
 615             /* Inner loop uses 50 flops */
 616         }
 617
 618         /* End of innermost loop */
 619
 620         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 621                                               f+i_coord_offset,fshift+i_shift_offset);
 622
 623         /* Increment number of inner iterations */
 624         inneriter                  += j_index_end - j_index_start;
 625
 626         /* Outer loop uses 7 flops */
 627     }
 628
 629     /* Increment number of outer iterations */
 630     outeriter        += nri;
 631
 632     /* Update outer/inner flops */
 633
 634     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*50);
 635 }