src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecNone_VdwCSTab_GeomP1P1_sparc64_hpc_ace_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_sparc64_hpc_ace_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwCSTab_GeomP1P1_VF_sparc64_hpc_ace_double
  51  * Electrostatics interaction: None
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecNone_VdwCSTab_GeomP1P1_VF_sparc64_hpc_ace_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 double precision SIMD, 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     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwjidx0A,vdwjidx0B;
  82     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  83     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  84     int              nvdwtype;
  85     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  86     int              *vdwtype;
  87     real             *vdwparam;
  88     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
  89     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
  90     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  91     real             *vftab;
  92     _fjsp_v2r8       itab_tmp;
  93     _fjsp_v2r8       dummy_mask,cutoff_mask;
  94     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
  95     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
  96     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
  97
  98     x                = xx[0];
  99     f                = ff[0];
 100
 101     nri              = nlist->nri;
 102     iinr             = nlist->iinr;
 103     jindex           = nlist->jindex;
 104     jjnr             = nlist->jjnr;
 105     shiftidx         = nlist->shift;
 106     gid              = nlist->gid;
 107     shiftvec         = fr->shift_vec[0];
 108     fshift           = fr->fshift[0];
 109     nvdwtype         = fr->ntype;
 110     vdwparam         = fr->nbfp;
 111     vdwtype          = mdatoms->typeA;
 112
 113     vftab            = kernel_data->table_vdw->data;
 114     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_vdw->scale);
 115
 116     /* Avoid stupid compiler warnings */
 117     jnrA = jnrB = 0;
 118     j_coord_offsetA = 0;
 119     j_coord_offsetB = 0;
 120
 121     outeriter        = 0;
 122     inneriter        = 0;
 123
 124     /* Start outer loop over neighborlists */
 125     for(iidx=0; iidx<nri; iidx++)
 126     {
 127         /* Load shift vector for this list */
 128         i_shift_offset   = DIM*shiftidx[iidx];
 129
 130         /* Load limits for loop over neighbors */
 131         j_index_start    = jindex[iidx];
 132         j_index_end      = jindex[iidx+1];
 133
 134         /* Get outer coordinate index */
 135         inr              = iinr[iidx];
 136         i_coord_offset   = DIM*inr;
 137
 138         /* Load i particle coords and add shift vector */
 139         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 140
 141         fix0             = _fjsp_setzero_v2r8();
 142         fiy0             = _fjsp_setzero_v2r8();
 143         fiz0             = _fjsp_setzero_v2r8();
 144
 145         /* Load parameters for i particles */
 146         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 147
 148         /* Reset potential sums */
 149         vvdwsum          = _fjsp_setzero_v2r8();
 150
 151         /* Start inner kernel loop */
 152         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 153         {
 154
 155             /* Get j neighbor index, and coordinate index */
 156             jnrA             = jjnr[jidx];
 157             jnrB             = jjnr[jidx+1];
 158             j_coord_offsetA  = DIM*jnrA;
 159             j_coord_offsetB  = DIM*jnrB;
 160
 161             /* load j atom coordinates */
 162             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 163                                               &jx0,&jy0,&jz0);
 164
 165             /* Calculate displacement vector */
 166             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 167             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 168             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 169
 170             /* Calculate squared distance and things based on it */
 171             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 172
 173             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 174
 175             /* Load parameters for j particles */
 176             vdwjidx0A        = 2*vdwtype[jnrA+0];
 177             vdwjidx0B        = 2*vdwtype[jnrB+0];
 178
 179             /**************************
 180              * CALCULATE INTERACTIONS *
 181              **************************/
 182
 183             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 184
 185             /* Compute parameters for interactions between i and j atoms */
 186             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 187                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 188
 189             /* Calculate table index by multiplying r with table scale and truncate to integer */
 190             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 191             itab_tmp         = _fjsp_dtox_v2r8(rt);
 192             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 193             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 194             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 195
 196             vfconv.i[0]     *= 8;
 197             vfconv.i[1]     *= 8;
 198
 199             /* CUBIC SPLINE TABLE DISPERSION */
 200             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 201             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 202             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 203             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 204             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 2 );
 205             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 206             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 207             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 208             vvdw6            = _fjsp_mul_v2r8(c6_00,VV);
 209             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 210             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 211
 212             /* CUBIC SPLINE TABLE REPULSION */
 213             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 214             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 4 );
 215             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 216             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 217             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 6 );
 218             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 219             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 220             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 221             vvdw12           = _fjsp_mul_v2r8(c12_00,VV);
 222             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 223             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 224             vvdw             = _fjsp_add_v2r8(vvdw12,vvdw6);
 225             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 226
 227             /* Update potential sum for this i atom from the interaction with this j atom. */
 228             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 229
 230             fscal            = fvdw;
 231
 232             /* Update vectorial force */
 233             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 234             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 235             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 236
 237             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 238
 239             /* Inner loop uses 59 flops */
 240         }
 241
 242         if(jidx<j_index_end)
 243         {
 244
 245             jnrA             = jjnr[jidx];
 246             j_coord_offsetA  = DIM*jnrA;
 247
 248             /* load j atom coordinates */
 249             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 250                                               &jx0,&jy0,&jz0);
 251
 252             /* Calculate displacement vector */
 253             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 254             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 255             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 256
 257             /* Calculate squared distance and things based on it */
 258             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 259
 260             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 261
 262             /* Load parameters for j particles */
 263             vdwjidx0A        = 2*vdwtype[jnrA+0];
 264
 265             /**************************
 266              * CALCULATE INTERACTIONS *
 267              **************************/
 268
 269             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 270
 271             /* Compute parameters for interactions between i and j atoms */
 272             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 273                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 274
 275             /* Calculate table index by multiplying r with table scale and truncate to integer */
 276             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 277             itab_tmp         = _fjsp_dtox_v2r8(rt);
 278             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 279             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 280             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 281
 282             vfconv.i[0]     *= 8;
 283             vfconv.i[1]     *= 8;
 284
 285             /* CUBIC SPLINE TABLE DISPERSION */
 286             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 287             F                = _fjsp_setzero_v2r8();
 288             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 289             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 290             H                = _fjsp_setzero_v2r8();
 291             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 292             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 293             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 294             vvdw6            = _fjsp_mul_v2r8(c6_00,VV);
 295             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 296             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 297
 298             /* CUBIC SPLINE TABLE REPULSION */
 299             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 300             F                = _fjsp_setzero_v2r8();
 301             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 302             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 303             H                = _fjsp_setzero_v2r8();
 304             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 305             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 306             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 307             vvdw12           = _fjsp_mul_v2r8(c12_00,VV);
 308             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 309             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 310             vvdw             = _fjsp_add_v2r8(vvdw12,vvdw6);
 311             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 312
 313             /* Update potential sum for this i atom from the interaction with this j atom. */
 314             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 315             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 316
 317             fscal            = fvdw;
 318
 319             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 320
 321             /* Update vectorial force */
 322             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 323             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 324             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 325
 326             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 327
 328             /* Inner loop uses 59 flops */
 329         }
 330
 331         /* End of innermost loop */
 332
 333         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 334                                               f+i_coord_offset,fshift+i_shift_offset);
 335
 336         ggid                        = gid[iidx];
 337         /* Update potential energies */
 338         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 339
 340         /* Increment number of inner iterations */
 341         inneriter                  += j_index_end - j_index_start;
 342
 343         /* Outer loop uses 7 flops */
 344     }
 345
 346     /* Increment number of outer iterations */
 347     outeriter        += nri;
 348
 349     /* Update outer/inner flops */
 350
 351     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*59);
 352 }
 353 /*
 354  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwCSTab_GeomP1P1_F_sparc64_hpc_ace_double
 355  * Electrostatics interaction: None
 356  * VdW interaction:            CubicSplineTable
 357  * Geometry:                   Particle-Particle
 358  * Calculate force/pot:        Force
 359  */
 360 void
 361 nb_kernel_ElecNone_VdwCSTab_GeomP1P1_F_sparc64_hpc_ace_double
 362                     (t_nblist                    * gmx_restrict       nlist,
 363                      rvec                        * gmx_restrict          xx,
 364                      rvec                        * gmx_restrict          ff,
 365                      struct t_forcerec           * gmx_restrict          fr,
 366                      t_mdatoms                   * gmx_restrict     mdatoms,
 367                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 368                      t_nrnb                      * gmx_restrict        nrnb)
 369 {
 370     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 371      * just 0 for non-waters.
 372      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 373      * jnr indices corresponding to data put in the four positions in the SIMD register.
 374      */
 375     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 376     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 377     int              jnrA,jnrB;
 378     int              j_coord_offsetA,j_coord_offsetB;
 379     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 380     real             rcutoff_scalar;
 381     real             *shiftvec,*fshift,*x,*f;
 382     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 383     int              vdwioffset0;
 384     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 385     int              vdwjidx0A,vdwjidx0B;
 386     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 387     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 388     int              nvdwtype;
 389     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 390     int              *vdwtype;
 391     real             *vdwparam;
 392     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 393     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 394     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 395     real             *vftab;
 396     _fjsp_v2r8       itab_tmp;
 397     _fjsp_v2r8       dummy_mask,cutoff_mask;
 398     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 399     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 400     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 401
 402     x                = xx[0];
 403     f                = ff[0];
 404
 405     nri              = nlist->nri;
 406     iinr             = nlist->iinr;
 407     jindex           = nlist->jindex;
 408     jjnr             = nlist->jjnr;
 409     shiftidx         = nlist->shift;
 410     gid              = nlist->gid;
 411     shiftvec         = fr->shift_vec[0];
 412     fshift           = fr->fshift[0];
 413     nvdwtype         = fr->ntype;
 414     vdwparam         = fr->nbfp;
 415     vdwtype          = mdatoms->typeA;
 416
 417     vftab            = kernel_data->table_vdw->data;
 418     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_vdw->scale);
 419
 420     /* Avoid stupid compiler warnings */
 421     jnrA = jnrB = 0;
 422     j_coord_offsetA = 0;
 423     j_coord_offsetB = 0;
 424
 425     outeriter        = 0;
 426     inneriter        = 0;
 427
 428     /* Start outer loop over neighborlists */
 429     for(iidx=0; iidx<nri; iidx++)
 430     {
 431         /* Load shift vector for this list */
 432         i_shift_offset   = DIM*shiftidx[iidx];
 433
 434         /* Load limits for loop over neighbors */
 435         j_index_start    = jindex[iidx];
 436         j_index_end      = jindex[iidx+1];
 437
 438         /* Get outer coordinate index */
 439         inr              = iinr[iidx];
 440         i_coord_offset   = DIM*inr;
 441
 442         /* Load i particle coords and add shift vector */
 443         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 444
 445         fix0             = _fjsp_setzero_v2r8();
 446         fiy0             = _fjsp_setzero_v2r8();
 447         fiz0             = _fjsp_setzero_v2r8();
 448
 449         /* Load parameters for i particles */
 450         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 451
 452         /* Start inner kernel loop */
 453         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 454         {
 455
 456             /* Get j neighbor index, and coordinate index */
 457             jnrA             = jjnr[jidx];
 458             jnrB             = jjnr[jidx+1];
 459             j_coord_offsetA  = DIM*jnrA;
 460             j_coord_offsetB  = DIM*jnrB;
 461
 462             /* load j atom coordinates */
 463             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 464                                               &jx0,&jy0,&jz0);
 465
 466             /* Calculate displacement vector */
 467             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 468             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 469             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 470
 471             /* Calculate squared distance and things based on it */
 472             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 473
 474             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 475
 476             /* Load parameters for j particles */
 477             vdwjidx0A        = 2*vdwtype[jnrA+0];
 478             vdwjidx0B        = 2*vdwtype[jnrB+0];
 479
 480             /**************************
 481              * CALCULATE INTERACTIONS *
 482              **************************/
 483
 484             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 485
 486             /* Compute parameters for interactions between i and j atoms */
 487             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 488                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 489
 490             /* Calculate table index by multiplying r with table scale and truncate to integer */
 491             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 492             itab_tmp         = _fjsp_dtox_v2r8(rt);
 493             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 494             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 495             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 496
 497             vfconv.i[0]     *= 8;
 498             vfconv.i[1]     *= 8;
 499
 500             /* CUBIC SPLINE TABLE DISPERSION */
 501             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 502             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 503             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 504             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 505             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 2 );
 506             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 507             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 508             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 509             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 510
 511             /* CUBIC SPLINE TABLE REPULSION */
 512             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 513             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 4 );
 514             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 515             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 516             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 6 );
 517             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 518             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 519             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 520             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 521             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 522
 523             fscal            = fvdw;
 524
 525             /* Update vectorial force */
 526             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 527             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 528             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 529
 530             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 531
 532             /* Inner loop uses 51 flops */
 533         }
 534
 535         if(jidx<j_index_end)
 536         {
 537
 538             jnrA             = jjnr[jidx];
 539             j_coord_offsetA  = DIM*jnrA;
 540
 541             /* load j atom coordinates */
 542             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 543                                               &jx0,&jy0,&jz0);
 544
 545             /* Calculate displacement vector */
 546             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 547             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 548             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 549
 550             /* Calculate squared distance and things based on it */
 551             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 552
 553             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 554
 555             /* Load parameters for j particles */
 556             vdwjidx0A        = 2*vdwtype[jnrA+0];
 557
 558             /**************************
 559              * CALCULATE INTERACTIONS *
 560              **************************/
 561
 562             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 563
 564             /* Compute parameters for interactions between i and j atoms */
 565             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 566                                          vdwparam+vdwioffset0+vdwjidx0B,&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 }