src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecNone_VdwCSTab_GeomP1P1_sparc64_hpc_ace_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 sparc64_hpc_ace_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 "kernelutil_sparc64_hpc_ace_double.h"
  50
  51 /*
  52  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwCSTab_GeomP1P1_VF_sparc64_hpc_ace_double
  53  * Electrostatics interaction: None
  54  * VdW interaction:            CubicSplineTable
  55  * Geometry:                   Particle-Particle
  56  * Calculate force/pot:        PotentialAndForce
  57  */
  58 void
  59 nb_kernel_ElecNone_VdwCSTab_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     int              nvdwtype;
  87     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  88     int              *vdwtype;
  89     real             *vdwparam;
  90     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
  91     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
  92     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  93     real             *vftab;
  94     _fjsp_v2r8       itab_tmp;
  95     _fjsp_v2r8       dummy_mask,cutoff_mask;
  96     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
  97     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
  98     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
  99
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     nvdwtype         = fr->ntype;
 112     vdwparam         = fr->nbfp;
 113     vdwtype          = mdatoms->typeA;
 114
 115     vftab            = kernel_data->table_vdw->data;
 116     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_vdw->scale);
 117
 118     /* Avoid stupid compiler warnings */
 119     jnrA = jnrB = 0;
 120     j_coord_offsetA = 0;
 121     j_coord_offsetB = 0;
 122
 123     outeriter        = 0;
 124     inneriter        = 0;
 125
 126     /* Start outer loop over neighborlists */
 127     for(iidx=0; iidx<nri; iidx++)
 128     {
 129         /* Load shift vector for this list */
 130         i_shift_offset   = DIM*shiftidx[iidx];
 131
 132         /* Load limits for loop over neighbors */
 133         j_index_start    = jindex[iidx];
 134         j_index_end      = jindex[iidx+1];
 135
 136         /* Get outer coordinate index */
 137         inr              = iinr[iidx];
 138         i_coord_offset   = DIM*inr;
 139
 140         /* Load i particle coords and add shift vector */
 141         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 142
 143         fix0             = _fjsp_setzero_v2r8();
 144         fiy0             = _fjsp_setzero_v2r8();
 145         fiz0             = _fjsp_setzero_v2r8();
 146
 147         /* Load parameters for i particles */
 148         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 149
 150         /* Reset potential sums */
 151         vvdwsum          = _fjsp_setzero_v2r8();
 152
 153         /* Start inner kernel loop */
 154         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 155         {
 156
 157             /* Get j neighbor index, and coordinate index */
 158             jnrA             = jjnr[jidx];
 159             jnrB             = jjnr[jidx+1];
 160             j_coord_offsetA  = DIM*jnrA;
 161             j_coord_offsetB  = DIM*jnrB;
 162
 163             /* load j atom coordinates */
 164             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 165                                               &jx0,&jy0,&jz0);
 166
 167             /* Calculate displacement vector */
 168             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 169             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 170             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 171
 172             /* Calculate squared distance and things based on it */
 173             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 174
 175             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 176
 177             /* Load parameters for j particles */
 178             vdwjidx0A        = 2*vdwtype[jnrA+0];
 179             vdwjidx0B        = 2*vdwtype[jnrB+0];
 180
 181             /**************************
 182              * CALCULATE INTERACTIONS *
 183              **************************/
 184
 185             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 186
 187             /* Compute parameters for interactions between i and j atoms */
 188             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 189                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 190
 191             /* Calculate table index by multiplying r with table scale and truncate to integer */
 192             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 193             itab_tmp         = _fjsp_dtox_v2r8(rt);
 194             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 195             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 196             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 197
 198             vfconv.i[0]     *= 8;
 199             vfconv.i[1]     *= 8;
 200
 201             /* CUBIC SPLINE TABLE DISPERSION */
 202             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 203             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 204             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 205             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 206             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 2 );
 207             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 208             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 209             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 210             vvdw6            = _fjsp_mul_v2r8(c6_00,VV);
 211             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 212             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 213
 214             /* CUBIC SPLINE TABLE REPULSION */
 215             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 216             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 4 );
 217             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 218             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 219             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 6 );
 220             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 221             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 222             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 223             vvdw12           = _fjsp_mul_v2r8(c12_00,VV);
 224             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 225             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 226             vvdw             = _fjsp_add_v2r8(vvdw12,vvdw6);
 227             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 228
 229             /* Update potential sum for this i atom from the interaction with this j atom. */
 230             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 231
 232             fscal            = fvdw;
 233
 234             /* Update vectorial force */
 235             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 236             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 237             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 238
 239             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 240
 241             /* Inner loop uses 59 flops */
 242         }
 243
 244         if(jidx<j_index_end)
 245         {
 246
 247             jnrA             = jjnr[jidx];
 248             j_coord_offsetA  = DIM*jnrA;
 249
 250             /* load j atom coordinates */
 251             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 252                                               &jx0,&jy0,&jz0);
 253
 254             /* Calculate displacement vector */
 255             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 256             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 257             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 258
 259             /* Calculate squared distance and things based on it */
 260             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 261
 262             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 263
 264             /* Load parameters for j particles */
 265             vdwjidx0A        = 2*vdwtype[jnrA+0];
 266
 267             /**************************
 268              * CALCULATE INTERACTIONS *
 269              **************************/
 270
 271             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 272
 273             /* Compute parameters for interactions between i and j atoms */
 274             gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 275
 276             /* Calculate table index by multiplying r with table scale and truncate to integer */
 277             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 278             itab_tmp         = _fjsp_dtox_v2r8(rt);
 279             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 280             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 281             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 282
 283             vfconv.i[0]     *= 8;
 284             vfconv.i[1]     *= 8;
 285
 286             /* CUBIC SPLINE TABLE DISPERSION */
 287             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 288             F                = _fjsp_setzero_v2r8();
 289             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 290             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 291             H                = _fjsp_setzero_v2r8();
 292             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 293             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 294             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 295             vvdw6            = _fjsp_mul_v2r8(c6_00,VV);
 296             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 297             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 298
 299             /* CUBIC SPLINE TABLE REPULSION */
 300             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 301             F                = _fjsp_setzero_v2r8();
 302             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 303             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 304             H                = _fjsp_setzero_v2r8();
 305             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 306             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 307             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 308             vvdw12           = _fjsp_mul_v2r8(c12_00,VV);
 309             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 310             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 311             vvdw             = _fjsp_add_v2r8(vvdw12,vvdw6);
 312             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 313
 314             /* Update potential sum for this i atom from the interaction with this j atom. */
 315             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 316             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 317
 318             fscal            = fvdw;
 319
 320             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 321
 322             /* Update vectorial force */
 323             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 324             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 325             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 326
 327             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 328
 329             /* Inner loop uses 59 flops */
 330         }
 331
 332         /* End of innermost loop */
 333
 334         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 335                                               f+i_coord_offset,fshift+i_shift_offset);
 336
 337         ggid                        = gid[iidx];
 338         /* Update potential energies */
 339         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 340
 341         /* Increment number of inner iterations */
 342         inneriter                  += j_index_end - j_index_start;
 343
 344         /* Outer loop uses 7 flops */
 345     }
 346
 347     /* Increment number of outer iterations */
 348     outeriter        += nri;
 349
 350     /* Update outer/inner flops */
 351
 352     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*59);
 353 }
 354 /*
 355  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwCSTab_GeomP1P1_F_sparc64_hpc_ace_double
 356  * Electrostatics interaction: None
 357  * VdW interaction:            CubicSplineTable
 358  * Geometry:                   Particle-Particle
 359  * Calculate force/pot:        Force
 360  */
 361 void
 362 nb_kernel_ElecNone_VdwCSTab_GeomP1P1_F_sparc64_hpc_ace_double
 363                     (t_nblist                    * gmx_restrict       nlist,
 364                      rvec                        * gmx_restrict          xx,
 365                      rvec                        * gmx_restrict          ff,
 366                      t_forcerec                  * gmx_restrict          fr,
 367                      t_mdatoms                   * gmx_restrict     mdatoms,
 368                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 369                      t_nrnb                      * gmx_restrict        nrnb)
 370 {
 371     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 372      * just 0 for non-waters.
 373      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 374      * jnr indices corresponding to data put in the four positions in the SIMD register.
 375      */
 376     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 377     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 378     int              jnrA,jnrB;
 379     int              j_coord_offsetA,j_coord_offsetB;
 380     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 381     real             rcutoff_scalar;
 382     real             *shiftvec,*fshift,*x,*f;
 383     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 384     int              vdwioffset0;
 385     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 386     int              vdwjidx0A,vdwjidx0B;
 387     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 388     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 389     int              nvdwtype;
 390     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 391     int              *vdwtype;
 392     real             *vdwparam;
 393     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 394     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 395     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 396     real             *vftab;
 397     _fjsp_v2r8       itab_tmp;
 398     _fjsp_v2r8       dummy_mask,cutoff_mask;
 399     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 400     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 401     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 402
 403     x                = xx[0];
 404     f                = ff[0];
 405
 406     nri              = nlist->nri;
 407     iinr             = nlist->iinr;
 408     jindex           = nlist->jindex;
 409     jjnr             = nlist->jjnr;
 410     shiftidx         = nlist->shift;
 411     gid              = nlist->gid;
 412     shiftvec         = fr->shift_vec[0];
 413     fshift           = fr->fshift[0];
 414     nvdwtype         = fr->ntype;
 415     vdwparam         = fr->nbfp;
 416     vdwtype          = mdatoms->typeA;
 417
 418     vftab            = kernel_data->table_vdw->data;
 419     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_vdw->scale);
 420
 421     /* Avoid stupid compiler warnings */
 422     jnrA = jnrB = 0;
 423     j_coord_offsetA = 0;
 424     j_coord_offsetB = 0;
 425
 426     outeriter        = 0;
 427     inneriter        = 0;
 428
 429     /* Start outer loop over neighborlists */
 430     for(iidx=0; iidx<nri; iidx++)
 431     {
 432         /* Load shift vector for this list */
 433         i_shift_offset   = DIM*shiftidx[iidx];
 434
 435         /* Load limits for loop over neighbors */
 436         j_index_start    = jindex[iidx];
 437         j_index_end      = jindex[iidx+1];
 438
 439         /* Get outer coordinate index */
 440         inr              = iinr[iidx];
 441         i_coord_offset   = DIM*inr;
 442
 443         /* Load i particle coords and add shift vector */
 444         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 445
 446         fix0             = _fjsp_setzero_v2r8();
 447         fiy0             = _fjsp_setzero_v2r8();
 448         fiz0             = _fjsp_setzero_v2r8();
 449
 450         /* Load parameters for i particles */
 451         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 452
 453         /* Start inner kernel loop */
 454         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 455         {
 456
 457             /* Get j neighbor index, and coordinate index */
 458             jnrA             = jjnr[jidx];
 459             jnrB             = jjnr[jidx+1];
 460             j_coord_offsetA  = DIM*jnrA;
 461             j_coord_offsetB  = DIM*jnrB;
 462
 463             /* load j atom coordinates */
 464             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 465                                               &jx0,&jy0,&jz0);
 466
 467             /* Calculate displacement vector */
 468             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 469             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 470             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 471
 472             /* Calculate squared distance and things based on it */
 473             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 474
 475             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 476
 477             /* Load parameters for j particles */
 478             vdwjidx0A        = 2*vdwtype[jnrA+0];
 479             vdwjidx0B        = 2*vdwtype[jnrB+0];
 480
 481             /**************************
 482              * CALCULATE INTERACTIONS *
 483              **************************/
 484
 485             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 486
 487             /* Compute parameters for interactions between i and j atoms */
 488             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 489                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 490
 491             /* Calculate table index by multiplying r with table scale and truncate to integer */
 492             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 493             itab_tmp         = _fjsp_dtox_v2r8(rt);
 494             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 495             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 496             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 497
 498             vfconv.i[0]     *= 8;
 499             vfconv.i[1]     *= 8;
 500
 501             /* CUBIC SPLINE TABLE DISPERSION */
 502             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 503             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 504             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 505             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 506             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 2 );
 507             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 508             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 509             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 510             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 511
 512             /* CUBIC SPLINE TABLE REPULSION */
 513             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 514             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 4 );
 515             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 516             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 517             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 6 );
 518             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 519             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 520             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 521             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 522             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 523
 524             fscal            = fvdw;
 525
 526             /* Update vectorial force */
 527             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 528             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 529             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 530
 531             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 532
 533             /* Inner loop uses 51 flops */
 534         }
 535
 536         if(jidx<j_index_end)
 537         {
 538
 539             jnrA             = jjnr[jidx];
 540             j_coord_offsetA  = DIM*jnrA;
 541
 542             /* load j atom coordinates */
 543             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 544                                               &jx0,&jy0,&jz0);
 545
 546             /* Calculate displacement vector */
 547             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 548             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 549             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 550
 551             /* Calculate squared distance and things based on it */
 552             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 553
 554             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 555
 556             /* Load parameters for j particles */
 557             vdwjidx0A        = 2*vdwtype[jnrA+0];
 558
 559             /**************************
 560              * CALCULATE INTERACTIONS *
 561              **************************/
 562
 563             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 564
 565             /* Compute parameters for interactions between i and j atoms */
 566             gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 567
 568             /* Calculate table index by multiplying r with table scale and truncate to integer */
 569             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 570             itab_tmp         = _fjsp_dtox_v2r8(rt);
 571             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 572             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 573             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 574
 575             vfconv.i[0]     *= 8;
 576             vfconv.i[1]     *= 8;
 577
 578             /* CUBIC SPLINE TABLE DISPERSION */
 579             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 580             F                = _fjsp_setzero_v2r8();
 581             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 582             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 583             H                = _fjsp_setzero_v2r8();
 584             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 585             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 586             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 587             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 588
 589             /* CUBIC SPLINE TABLE REPULSION */
 590             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 591             F                = _fjsp_setzero_v2r8();
 592             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 593             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 594             H                = _fjsp_setzero_v2r8();
 595             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 596             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 597             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 598             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 599             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 600
 601             fscal            = fvdw;
 602
 603             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 604
 605             /* Update vectorial force */
 606             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 607             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 608             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 609
 610             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 611
 612             /* Inner loop uses 51 flops */
 613         }
 614
 615         /* End of innermost loop */
 616
 617         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 618                                               f+i_coord_offset,fshift+i_shift_offset);
 619
 620         /* Increment number of inner iterations */
 621         inneriter                  += j_index_end - j_index_start;
 622
 623         /* Outer loop uses 6 flops */
 624     }
 625
 626     /* Increment number of outer iterations */
 627     outeriter        += nri;
 628
 629     /* Update outer/inner flops */
 630
 631     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*51);
 632 }