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 #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 "gromacs/legacyheaders/vec.h"
  47 #include "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_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 275                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 276
 277             /* Calculate table index by multiplying r with table scale and truncate to integer */
 278             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 279             itab_tmp         = _fjsp_dtox_v2r8(rt);
 280             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 281             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 282             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 283
 284             vfconv.i[0]     *= 8;
 285             vfconv.i[1]     *= 8;
 286
 287             /* CUBIC SPLINE TABLE DISPERSION */
 288             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 289             F                = _fjsp_setzero_v2r8();
 290             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 291             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 292             H                = _fjsp_setzero_v2r8();
 293             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 294             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 295             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 296             vvdw6            = _fjsp_mul_v2r8(c6_00,VV);
 297             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 298             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 299
 300             /* CUBIC SPLINE TABLE REPULSION */
 301             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 302             F                = _fjsp_setzero_v2r8();
 303             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 304             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 305             H                = _fjsp_setzero_v2r8();
 306             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 307             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 308             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 309             vvdw12           = _fjsp_mul_v2r8(c12_00,VV);
 310             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 311             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 312             vvdw             = _fjsp_add_v2r8(vvdw12,vvdw6);
 313             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 314
 315             /* Update potential sum for this i atom from the interaction with this j atom. */
 316             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 317             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 318
 319             fscal            = fvdw;
 320
 321             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 322
 323             /* Update vectorial force */
 324             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 325             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 326             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 327
 328             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 329
 330             /* Inner loop uses 59 flops */
 331         }
 332
 333         /* End of innermost loop */
 334
 335         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 336                                               f+i_coord_offset,fshift+i_shift_offset);
 337
 338         ggid                        = gid[iidx];
 339         /* Update potential energies */
 340         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 341
 342         /* Increment number of inner iterations */
 343         inneriter                  += j_index_end - j_index_start;
 344
 345         /* Outer loop uses 7 flops */
 346     }
 347
 348     /* Increment number of outer iterations */
 349     outeriter        += nri;
 350
 351     /* Update outer/inner flops */
 352
 353     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*59);
 354 }
 355 /*
 356  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwCSTab_GeomP1P1_F_sparc64_hpc_ace_double
 357  * Electrostatics interaction: None
 358  * VdW interaction:            CubicSplineTable
 359  * Geometry:                   Particle-Particle
 360  * Calculate force/pot:        Force
 361  */
 362 void
 363 nb_kernel_ElecNone_VdwCSTab_GeomP1P1_F_sparc64_hpc_ace_double
 364                     (t_nblist                    * gmx_restrict       nlist,
 365                      rvec                        * gmx_restrict          xx,
 366                      rvec                        * gmx_restrict          ff,
 367                      t_forcerec                  * gmx_restrict          fr,
 368                      t_mdatoms                   * gmx_restrict     mdatoms,
 369                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 370                      t_nrnb                      * gmx_restrict        nrnb)
 371 {
 372     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 373      * just 0 for non-waters.
 374      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 375      * jnr indices corresponding to data put in the four positions in the SIMD register.
 376      */
 377     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 378     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 379     int              jnrA,jnrB;
 380     int              j_coord_offsetA,j_coord_offsetB;
 381     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 382     real             rcutoff_scalar;
 383     real             *shiftvec,*fshift,*x,*f;
 384     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 385     int              vdwioffset0;
 386     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 387     int              vdwjidx0A,vdwjidx0B;
 388     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 389     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 390     int              nvdwtype;
 391     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 392     int              *vdwtype;
 393     real             *vdwparam;
 394     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 395     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 396     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 397     real             *vftab;
 398     _fjsp_v2r8       itab_tmp;
 399     _fjsp_v2r8       dummy_mask,cutoff_mask;
 400     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 401     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 402     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 403
 404     x                = xx[0];
 405     f                = ff[0];
 406
 407     nri              = nlist->nri;
 408     iinr             = nlist->iinr;
 409     jindex           = nlist->jindex;
 410     jjnr             = nlist->jjnr;
 411     shiftidx         = nlist->shift;
 412     gid              = nlist->gid;
 413     shiftvec         = fr->shift_vec[0];
 414     fshift           = fr->fshift[0];
 415     nvdwtype         = fr->ntype;
 416     vdwparam         = fr->nbfp;
 417     vdwtype          = mdatoms->typeA;
 418
 419     vftab            = kernel_data->table_vdw->data;
 420     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_vdw->scale);
 421
 422     /* Avoid stupid compiler warnings */
 423     jnrA = jnrB = 0;
 424     j_coord_offsetA = 0;
 425     j_coord_offsetB = 0;
 426
 427     outeriter        = 0;
 428     inneriter        = 0;
 429
 430     /* Start outer loop over neighborlists */
 431     for(iidx=0; iidx<nri; iidx++)
 432     {
 433         /* Load shift vector for this list */
 434         i_shift_offset   = DIM*shiftidx[iidx];
 435
 436         /* Load limits for loop over neighbors */
 437         j_index_start    = jindex[iidx];
 438         j_index_end      = jindex[iidx+1];
 439
 440         /* Get outer coordinate index */
 441         inr              = iinr[iidx];
 442         i_coord_offset   = DIM*inr;
 443
 444         /* Load i particle coords and add shift vector */
 445         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 446
 447         fix0             = _fjsp_setzero_v2r8();
 448         fiy0             = _fjsp_setzero_v2r8();
 449         fiz0             = _fjsp_setzero_v2r8();
 450
 451         /* Load parameters for i particles */
 452         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 453
 454         /* Start inner kernel loop */
 455         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 456         {
 457
 458             /* Get j neighbor index, and coordinate index */
 459             jnrA             = jjnr[jidx];
 460             jnrB             = jjnr[jidx+1];
 461             j_coord_offsetA  = DIM*jnrA;
 462             j_coord_offsetB  = DIM*jnrB;
 463
 464             /* load j atom coordinates */
 465             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 466                                               &jx0,&jy0,&jz0);
 467
 468             /* Calculate displacement vector */
 469             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 470             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 471             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 472
 473             /* Calculate squared distance and things based on it */
 474             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 475
 476             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 477
 478             /* Load parameters for j particles */
 479             vdwjidx0A        = 2*vdwtype[jnrA+0];
 480             vdwjidx0B        = 2*vdwtype[jnrB+0];
 481
 482             /**************************
 483              * CALCULATE INTERACTIONS *
 484              **************************/
 485
 486             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 487
 488             /* Compute parameters for interactions between i and j atoms */
 489             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 490                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 491
 492             /* Calculate table index by multiplying r with table scale and truncate to integer */
 493             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 494             itab_tmp         = _fjsp_dtox_v2r8(rt);
 495             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 496             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 497             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 498
 499             vfconv.i[0]     *= 8;
 500             vfconv.i[1]     *= 8;
 501
 502             /* CUBIC SPLINE TABLE DISPERSION */
 503             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 504             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 505             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 506             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 507             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 2 );
 508             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 509             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 510             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 511             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 512
 513             /* CUBIC SPLINE TABLE REPULSION */
 514             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 515             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 4 );
 516             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 517             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 518             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 6 );
 519             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 520             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 521             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 522             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 523             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 524
 525             fscal            = fvdw;
 526
 527             /* Update vectorial force */
 528             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 529             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 530             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 531
 532             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 533
 534             /* Inner loop uses 51 flops */
 535         }
 536
 537         if(jidx<j_index_end)
 538         {
 539
 540             jnrA             = jjnr[jidx];
 541             j_coord_offsetA  = DIM*jnrA;
 542
 543             /* load j atom coordinates */
 544             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 545                                               &jx0,&jy0,&jz0);
 546
 547             /* Calculate displacement vector */
 548             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 549             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 550             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 551
 552             /* Calculate squared distance and things based on it */
 553             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 554
 555             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 556
 557             /* Load parameters for j particles */
 558             vdwjidx0A        = 2*vdwtype[jnrA+0];
 559
 560             /**************************
 561              * CALCULATE INTERACTIONS *
 562              **************************/
 563
 564             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 565
 566             /* Compute parameters for interactions between i and j atoms */
 567             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 568                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 569
 570             /* Calculate table index by multiplying r with table scale and truncate to integer */
 571             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 572             itab_tmp         = _fjsp_dtox_v2r8(rt);
 573             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 574             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 575             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 576
 577             vfconv.i[0]     *= 8;
 578             vfconv.i[1]     *= 8;
 579
 580             /* CUBIC SPLINE TABLE DISPERSION */
 581             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 582             F                = _fjsp_setzero_v2r8();
 583             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 584             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 585             H                = _fjsp_setzero_v2r8();
 586             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 587             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 588             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 589             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 590
 591             /* CUBIC SPLINE TABLE REPULSION */
 592             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 593             F                = _fjsp_setzero_v2r8();
 594             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 595             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 596             H                = _fjsp_setzero_v2r8();
 597             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 598             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 599             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 600             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 601             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 602
 603             fscal            = fvdw;
 604
 605             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 606
 607             /* Update vectorial force */
 608             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 609             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 610             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 611
 612             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 613
 614             /* Inner loop uses 51 flops */
 615         }
 616
 617         /* End of innermost loop */
 618
 619         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 620                                               f+i_coord_offset,fshift+i_shift_offset);
 621
 622         /* Increment number of inner iterations */
 623         inneriter                  += j_index_end - j_index_start;
 624
 625         /* Outer loop uses 6 flops */
 626     }
 627
 628     /* Increment number of outer iterations */
 629     outeriter        += nri;
 630
 631     /* Update outer/inner flops */
 632
 633     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*51);
 634 }