src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecGB_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, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sparc64_hpc_ace_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "kernelutil_sparc64_hpc_ace_double.h"
  50
  51 /*
  52  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwCSTab_GeomP1P1_VF_sparc64_hpc_ace_double
  53  * Electrostatics interaction: GeneralizedBorn
  54  * VdW interaction:            CubicSplineTable
  55  * Geometry:                   Particle-Particle
  56  * Calculate force/pot:        PotentialAndForce
  57  */
  58 void
  59 nb_kernel_ElecGB_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     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
  87     real             *charge;
  88     _fjsp_v2r8       vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,twogbeps,dvdatmp;
  89     _fjsp_v2r8       minushalf = gmx_fjsp_set1_v2r8(-0.5);
  90     real             *invsqrta,*dvda,*gbtab;
  91     int              nvdwtype;
  92     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  93     int              *vdwtype;
  94     real             *vdwparam;
  95     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
  96     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
  97     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  98     real             *vftab;
  99     _fjsp_v2r8       itab_tmp;
 100     _fjsp_v2r8       dummy_mask,cutoff_mask;
 101     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 102     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 103     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 104
 105     x                = xx[0];
 106     f                = ff[0];
 107
 108     nri              = nlist->nri;
 109     iinr             = nlist->iinr;
 110     jindex           = nlist->jindex;
 111     jjnr             = nlist->jjnr;
 112     shiftidx         = nlist->shift;
 113     gid              = nlist->gid;
 114     shiftvec         = fr->shift_vec[0];
 115     fshift           = fr->fshift[0];
 116     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 117     charge           = mdatoms->chargeA;
 118     nvdwtype         = fr->ntype;
 119     vdwparam         = fr->nbfp;
 120     vdwtype          = mdatoms->typeA;
 121
 122     vftab            = kernel_data->table_vdw->data;
 123     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_vdw->scale);
 124
 125     invsqrta         = fr->invsqrta;
 126     dvda             = fr->dvda;
 127     gbtabscale       = gmx_fjsp_set1_v2r8(fr->gbtab.scale);
 128     gbtab            = fr->gbtab.data;
 129     gbinvepsdiff     = gmx_fjsp_set1_v2r8((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 130
 131     /* Avoid stupid compiler warnings */
 132     jnrA = jnrB = 0;
 133     j_coord_offsetA = 0;
 134     j_coord_offsetB = 0;
 135
 136     outeriter        = 0;
 137     inneriter        = 0;
 138
 139     /* Start outer loop over neighborlists */
 140     for(iidx=0; iidx<nri; iidx++)
 141     {
 142         /* Load shift vector for this list */
 143         i_shift_offset   = DIM*shiftidx[iidx];
 144
 145         /* Load limits for loop over neighbors */
 146         j_index_start    = jindex[iidx];
 147         j_index_end      = jindex[iidx+1];
 148
 149         /* Get outer coordinate index */
 150         inr              = iinr[iidx];
 151         i_coord_offset   = DIM*inr;
 152
 153         /* Load i particle coords and add shift vector */
 154         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 155
 156         fix0             = _fjsp_setzero_v2r8();
 157         fiy0             = _fjsp_setzero_v2r8();
 158         fiz0             = _fjsp_setzero_v2r8();
 159
 160         /* Load parameters for i particles */
 161         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 162         isai0            = gmx_fjsp_load1_v2r8(invsqrta+inr+0);
 163         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 164
 165         /* Reset potential sums */
 166         velecsum         = _fjsp_setzero_v2r8();
 167         vgbsum           = _fjsp_setzero_v2r8();
 168         vvdwsum          = _fjsp_setzero_v2r8();
 169         dvdasum          = _fjsp_setzero_v2r8();
 170
 171         /* Start inner kernel loop */
 172         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 173         {
 174
 175             /* Get j neighbor index, and coordinate index */
 176             jnrA             = jjnr[jidx];
 177             jnrB             = jjnr[jidx+1];
 178             j_coord_offsetA  = DIM*jnrA;
 179             j_coord_offsetB  = DIM*jnrB;
 180
 181             /* load j atom coordinates */
 182             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 183                                               &jx0,&jy0,&jz0);
 184
 185             /* Calculate displacement vector */
 186             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 187             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 188             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 189
 190             /* Calculate squared distance and things based on it */
 191             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 192
 193             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 194
 195             /* Load parameters for j particles */
 196             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 197             isaj0            = gmx_fjsp_load_2real_swizzle_v2r8(invsqrta+jnrA+0,invsqrta+jnrB+0);
 198             vdwjidx0A        = 2*vdwtype[jnrA+0];
 199             vdwjidx0B        = 2*vdwtype[jnrB+0];
 200
 201             /**************************
 202              * CALCULATE INTERACTIONS *
 203              **************************/
 204
 205             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 206
 207             /* Compute parameters for interactions between i and j atoms */
 208             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 209             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 210                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 211
 212             /* Calculate table index by multiplying r with table scale and truncate to integer */
 213             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 214             itab_tmp         = _fjsp_dtox_v2r8(rt);
 215             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 216             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 217             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 218
 219             vfconv.i[0]     *= 8;
 220             vfconv.i[1]     *= 8;
 221
 222             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 223             isaprod          = _fjsp_mul_v2r8(isai0,isaj0);
 224             gbqqfactor       = _fjsp_neg_v2r8(_fjsp_mul_v2r8(qq00,_fjsp_mul_v2r8(isaprod,gbinvepsdiff)));
 225             gbscale          = _fjsp_mul_v2r8(isaprod,gbtabscale);
 226
 227             /* Calculate generalized born table index - this is a separate table from the normal one,
 228              * but we use the same procedure by multiplying r with scale and truncating to integer.
 229              */
 230             rt               = _fjsp_mul_v2r8(r00,gbscale);
 231             itab_tmp         = _fjsp_dtox_v2r8(rt);
 232             gbeps            = _fjsp_sub_v2r8(rt,_fjsp_xtod_v2r8(itab_tmp));
 233             _fjsp_store_v2r8(&gbconv.simd,itab_tmp);
 234
 235             Y                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] );
 236             F                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] );
 237             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 238             G                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] +2);
 239             H                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] +2);
 240             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 241             Fp               = _fjsp_madd_v2r8(gbeps,_fjsp_madd_v2r8(gbeps,H,G),F);
 242             VV               = _fjsp_madd_v2r8(gbeps,Fp,Y);
 243             vgb              = _fjsp_mul_v2r8(gbqqfactor,VV);
 244
 245             twogbeps         = _fjsp_add_v2r8(gbeps,gbeps);
 246             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twogbeps,H,G),gbeps,Fp);
 247             fgb              = _fjsp_mul_v2r8(gbqqfactor,_fjsp_mul_v2r8(FF,gbscale));
 248             dvdatmp          = _fjsp_mul_v2r8(minushalf,_fjsp_madd_v2r8(fgb,r00,vgb));
 249             dvdasum          = _fjsp_add_v2r8(dvdasum,dvdatmp);
 250             gmx_fjsp_increment_2real_swizzle_v2r8(dvda+jnrA,dvda+jnrB,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj0,isaj0)));
 251             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 252             felec            = _fjsp_mul_v2r8(_fjsp_msub_v2r8(velec,rinv00,fgb),rinv00);
 253
 254             /* CUBIC SPLINE TABLE DISPERSION */
 255             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 256             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 257             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 258             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 259             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 2 );
 260             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 261             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 262             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 263             vvdw6            = _fjsp_mul_v2r8(c6_00,VV);
 264             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 265             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 266
 267             /* CUBIC SPLINE TABLE REPULSION */
 268             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 269             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 4 );
 270             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 271             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 272             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 6 );
 273             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 274             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 275             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 276             vvdw12           = _fjsp_mul_v2r8(c12_00,VV);
 277             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 278             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 279             vvdw             = _fjsp_add_v2r8(vvdw12,vvdw6);
 280             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 281
 282             /* Update potential sum for this i atom from the interaction with this j atom. */
 283             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 284             vgbsum           = _fjsp_add_v2r8(vgbsum,vgb);
 285             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 286
 287             fscal            = _fjsp_add_v2r8(felec,fvdw);
 288
 289             /* Update vectorial force */
 290             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 291             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 292             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 293
 294             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 295
 296             /* Inner loop uses 95 flops */
 297         }
 298
 299         if(jidx<j_index_end)
 300         {
 301
 302             jnrA             = jjnr[jidx];
 303             j_coord_offsetA  = DIM*jnrA;
 304
 305             /* load j atom coordinates */
 306             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 307                                               &jx0,&jy0,&jz0);
 308
 309             /* Calculate displacement vector */
 310             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 311             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 312             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 313
 314             /* Calculate squared distance and things based on it */
 315             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 316
 317             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 318
 319             /* Load parameters for j particles */
 320             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 321             isaj0            = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),invsqrta+jnrA+0);
 322             vdwjidx0A        = 2*vdwtype[jnrA+0];
 323
 324             /**************************
 325              * CALCULATE INTERACTIONS *
 326              **************************/
 327
 328             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 329
 330             /* Compute parameters for interactions between i and j atoms */
 331             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 332             gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 333
 334             /* Calculate table index by multiplying r with table scale and truncate to integer */
 335             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 336             itab_tmp         = _fjsp_dtox_v2r8(rt);
 337             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 338             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 339             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 340
 341             vfconv.i[0]     *= 8;
 342             vfconv.i[1]     *= 8;
 343
 344             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 345             isaprod          = _fjsp_mul_v2r8(isai0,isaj0);
 346             gbqqfactor       = _fjsp_neg_v2r8(_fjsp_mul_v2r8(qq00,_fjsp_mul_v2r8(isaprod,gbinvepsdiff)));
 347             gbscale          = _fjsp_mul_v2r8(isaprod,gbtabscale);
 348
 349             /* Calculate generalized born table index - this is a separate table from the normal one,
 350              * but we use the same procedure by multiplying r with scale and truncating to integer.
 351              */
 352             rt               = _fjsp_mul_v2r8(r00,gbscale);
 353             itab_tmp         = _fjsp_dtox_v2r8(rt);
 354             gbeps            = _fjsp_sub_v2r8(rt,_fjsp_xtod_v2r8(itab_tmp));
 355             _fjsp_store_v2r8(&gbconv.simd,itab_tmp);
 356
 357             Y                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] );
 358             F                = _fjsp_setzero_v2r8();
 359             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 360             G                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] +2);
 361             H                = _fjsp_setzero_v2r8();
 362             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 363             Fp               = _fjsp_madd_v2r8(gbeps,_fjsp_madd_v2r8(gbeps,H,G),F);
 364             VV               = _fjsp_madd_v2r8(gbeps,Fp,Y);
 365             vgb              = _fjsp_mul_v2r8(gbqqfactor,VV);
 366
 367             twogbeps         = _fjsp_add_v2r8(gbeps,gbeps);
 368             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twogbeps,H,G),gbeps,Fp);
 369             fgb              = _fjsp_mul_v2r8(gbqqfactor,_fjsp_mul_v2r8(FF,gbscale));
 370             dvdatmp          = _fjsp_mul_v2r8(minushalf,_fjsp_madd_v2r8(fgb,r00,vgb));
 371             dvdasum          = _fjsp_add_v2r8(dvdasum,dvdatmp);
 372             gmx_fjsp_increment_1real_v2r8(dvda+jnrA,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj0,isaj0)));
 373             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 374             felec            = _fjsp_mul_v2r8(_fjsp_msub_v2r8(velec,rinv00,fgb),rinv00);
 375
 376             /* CUBIC SPLINE TABLE DISPERSION */
 377             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 378             F                = _fjsp_setzero_v2r8();
 379             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 380             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 381             H                = _fjsp_setzero_v2r8();
 382             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 383             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 384             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 385             vvdw6            = _fjsp_mul_v2r8(c6_00,VV);
 386             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 387             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 388
 389             /* CUBIC SPLINE TABLE REPULSION */
 390             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 391             F                = _fjsp_setzero_v2r8();
 392             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 393             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 394             H                = _fjsp_setzero_v2r8();
 395             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 396             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 397             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 398             vvdw12           = _fjsp_mul_v2r8(c12_00,VV);
 399             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 400             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 401             vvdw             = _fjsp_add_v2r8(vvdw12,vvdw6);
 402             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 403
 404             /* Update potential sum for this i atom from the interaction with this j atom. */
 405             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 406             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 407             vgb              = _fjsp_unpacklo_v2r8(vgb,_fjsp_setzero_v2r8());
 408             vgbsum           = _fjsp_add_v2r8(vgbsum,vgb);
 409             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 410             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 411
 412             fscal            = _fjsp_add_v2r8(felec,fvdw);
 413
 414             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 415
 416             /* Update vectorial force */
 417             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 418             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 419             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 420
 421             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 422
 423             /* Inner loop uses 95 flops */
 424         }
 425
 426         /* End of innermost loop */
 427
 428         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 429                                               f+i_coord_offset,fshift+i_shift_offset);
 430
 431         ggid                        = gid[iidx];
 432         /* Update potential energies */
 433         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 434         gmx_fjsp_update_1pot_v2r8(vgbsum,kernel_data->energygrp_polarization+ggid);
 435         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 436         dvdasum = _fjsp_mul_v2r8(dvdasum, _fjsp_mul_v2r8(isai0,isai0));
 437         gmx_fjsp_update_1pot_v2r8(dvdasum,dvda+inr);
 438
 439         /* Increment number of inner iterations */
 440         inneriter                  += j_index_end - j_index_start;
 441
 442         /* Outer loop uses 10 flops */
 443     }
 444
 445     /* Increment number of outer iterations */
 446     outeriter        += nri;
 447
 448     /* Update outer/inner flops */
 449
 450     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*10 + inneriter*95);
 451 }
 452 /*
 453  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwCSTab_GeomP1P1_F_sparc64_hpc_ace_double
 454  * Electrostatics interaction: GeneralizedBorn
 455  * VdW interaction:            CubicSplineTable
 456  * Geometry:                   Particle-Particle
 457  * Calculate force/pot:        Force
 458  */
 459 void
 460 nb_kernel_ElecGB_VdwCSTab_GeomP1P1_F_sparc64_hpc_ace_double
 461                     (t_nblist                    * gmx_restrict       nlist,
 462                      rvec                        * gmx_restrict          xx,
 463                      rvec                        * gmx_restrict          ff,
 464                      t_forcerec                  * gmx_restrict          fr,
 465                      t_mdatoms                   * gmx_restrict     mdatoms,
 466                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 467                      t_nrnb                      * gmx_restrict        nrnb)
 468 {
 469     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 470      * just 0 for non-waters.
 471      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 472      * jnr indices corresponding to data put in the four positions in the SIMD register.
 473      */
 474     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 475     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 476     int              jnrA,jnrB;
 477     int              j_coord_offsetA,j_coord_offsetB;
 478     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 479     real             rcutoff_scalar;
 480     real             *shiftvec,*fshift,*x,*f;
 481     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 482     int              vdwioffset0;
 483     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 484     int              vdwjidx0A,vdwjidx0B;
 485     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 486     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 487     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 488     real             *charge;
 489     _fjsp_v2r8       vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,twogbeps,dvdatmp;
 490     _fjsp_v2r8       minushalf = gmx_fjsp_set1_v2r8(-0.5);
 491     real             *invsqrta,*dvda,*gbtab;
 492     int              nvdwtype;
 493     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 494     int              *vdwtype;
 495     real             *vdwparam;
 496     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 497     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 498     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 499     real             *vftab;
 500     _fjsp_v2r8       itab_tmp;
 501     _fjsp_v2r8       dummy_mask,cutoff_mask;
 502     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 503     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 504     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 505
 506     x                = xx[0];
 507     f                = ff[0];
 508
 509     nri              = nlist->nri;
 510     iinr             = nlist->iinr;
 511     jindex           = nlist->jindex;
 512     jjnr             = nlist->jjnr;
 513     shiftidx         = nlist->shift;
 514     gid              = nlist->gid;
 515     shiftvec         = fr->shift_vec[0];
 516     fshift           = fr->fshift[0];
 517     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 518     charge           = mdatoms->chargeA;
 519     nvdwtype         = fr->ntype;
 520     vdwparam         = fr->nbfp;
 521     vdwtype          = mdatoms->typeA;
 522
 523     vftab            = kernel_data->table_vdw->data;
 524     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_vdw->scale);
 525
 526     invsqrta         = fr->invsqrta;
 527     dvda             = fr->dvda;
 528     gbtabscale       = gmx_fjsp_set1_v2r8(fr->gbtab.scale);
 529     gbtab            = fr->gbtab.data;
 530     gbinvepsdiff     = gmx_fjsp_set1_v2r8((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 531
 532     /* Avoid stupid compiler warnings */
 533     jnrA = jnrB = 0;
 534     j_coord_offsetA = 0;
 535     j_coord_offsetB = 0;
 536
 537     outeriter        = 0;
 538     inneriter        = 0;
 539
 540     /* Start outer loop over neighborlists */
 541     for(iidx=0; iidx<nri; iidx++)
 542     {
 543         /* Load shift vector for this list */
 544         i_shift_offset   = DIM*shiftidx[iidx];
 545
 546         /* Load limits for loop over neighbors */
 547         j_index_start    = jindex[iidx];
 548         j_index_end      = jindex[iidx+1];
 549
 550         /* Get outer coordinate index */
 551         inr              = iinr[iidx];
 552         i_coord_offset   = DIM*inr;
 553
 554         /* Load i particle coords and add shift vector */
 555         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 556
 557         fix0             = _fjsp_setzero_v2r8();
 558         fiy0             = _fjsp_setzero_v2r8();
 559         fiz0             = _fjsp_setzero_v2r8();
 560
 561         /* Load parameters for i particles */
 562         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 563         isai0            = gmx_fjsp_load1_v2r8(invsqrta+inr+0);
 564         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 565
 566         dvdasum          = _fjsp_setzero_v2r8();
 567
 568         /* Start inner kernel loop */
 569         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 570         {
 571
 572             /* Get j neighbor index, and coordinate index */
 573             jnrA             = jjnr[jidx];
 574             jnrB             = jjnr[jidx+1];
 575             j_coord_offsetA  = DIM*jnrA;
 576             j_coord_offsetB  = DIM*jnrB;
 577
 578             /* load j atom coordinates */
 579             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 580                                               &jx0,&jy0,&jz0);
 581
 582             /* Calculate displacement vector */
 583             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 584             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 585             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 586
 587             /* Calculate squared distance and things based on it */
 588             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 589
 590             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 591
 592             /* Load parameters for j particles */
 593             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 594             isaj0            = gmx_fjsp_load_2real_swizzle_v2r8(invsqrta+jnrA+0,invsqrta+jnrB+0);
 595             vdwjidx0A        = 2*vdwtype[jnrA+0];
 596             vdwjidx0B        = 2*vdwtype[jnrB+0];
 597
 598             /**************************
 599              * CALCULATE INTERACTIONS *
 600              **************************/
 601
 602             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 603
 604             /* Compute parameters for interactions between i and j atoms */
 605             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 606             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 607                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 608
 609             /* Calculate table index by multiplying r with table scale and truncate to integer */
 610             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 611             itab_tmp         = _fjsp_dtox_v2r8(rt);
 612             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 613             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 614             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 615
 616             vfconv.i[0]     *= 8;
 617             vfconv.i[1]     *= 8;
 618
 619             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 620             isaprod          = _fjsp_mul_v2r8(isai0,isaj0);
 621             gbqqfactor       = _fjsp_neg_v2r8(_fjsp_mul_v2r8(qq00,_fjsp_mul_v2r8(isaprod,gbinvepsdiff)));
 622             gbscale          = _fjsp_mul_v2r8(isaprod,gbtabscale);
 623
 624             /* Calculate generalized born table index - this is a separate table from the normal one,
 625              * but we use the same procedure by multiplying r with scale and truncating to integer.
 626              */
 627             rt               = _fjsp_mul_v2r8(r00,gbscale);
 628             itab_tmp         = _fjsp_dtox_v2r8(rt);
 629             gbeps            = _fjsp_sub_v2r8(rt,_fjsp_xtod_v2r8(itab_tmp));
 630             _fjsp_store_v2r8(&gbconv.simd,itab_tmp);
 631
 632             Y                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] );
 633             F                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] );
 634             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 635             G                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] +2);
 636             H                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] +2);
 637             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 638             Fp               = _fjsp_madd_v2r8(gbeps,_fjsp_madd_v2r8(gbeps,H,G),F);
 639             VV               = _fjsp_madd_v2r8(gbeps,Fp,Y);
 640             vgb              = _fjsp_mul_v2r8(gbqqfactor,VV);
 641
 642             twogbeps         = _fjsp_add_v2r8(gbeps,gbeps);
 643             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twogbeps,H,G),gbeps,Fp);
 644             fgb              = _fjsp_mul_v2r8(gbqqfactor,_fjsp_mul_v2r8(FF,gbscale));
 645             dvdatmp          = _fjsp_mul_v2r8(minushalf,_fjsp_madd_v2r8(fgb,r00,vgb));
 646             dvdasum          = _fjsp_add_v2r8(dvdasum,dvdatmp);
 647             gmx_fjsp_increment_2real_swizzle_v2r8(dvda+jnrA,dvda+jnrB,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj0,isaj0)));
 648             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 649             felec            = _fjsp_mul_v2r8(_fjsp_msub_v2r8(velec,rinv00,fgb),rinv00);
 650
 651             /* CUBIC SPLINE TABLE DISPERSION */
 652             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 653             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 654             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 655             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 656             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 2 );
 657             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 658             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 659             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 660             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 661
 662             /* CUBIC SPLINE TABLE REPULSION */
 663             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 664             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 4 );
 665             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 666             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 667             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 6 );
 668             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 669             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 670             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 671             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 672             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 673
 674             fscal            = _fjsp_add_v2r8(felec,fvdw);
 675
 676             /* Update vectorial force */
 677             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 678             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 679             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 680
 681             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 682
 683             /* Inner loop uses 85 flops */
 684         }
 685
 686         if(jidx<j_index_end)
 687         {
 688
 689             jnrA             = jjnr[jidx];
 690             j_coord_offsetA  = DIM*jnrA;
 691
 692             /* load j atom coordinates */
 693             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 694                                               &jx0,&jy0,&jz0);
 695
 696             /* Calculate displacement vector */
 697             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 698             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 699             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 700
 701             /* Calculate squared distance and things based on it */
 702             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 703
 704             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 705
 706             /* Load parameters for j particles */
 707             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 708             isaj0            = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),invsqrta+jnrA+0);
 709             vdwjidx0A        = 2*vdwtype[jnrA+0];
 710
 711             /**************************
 712              * CALCULATE INTERACTIONS *
 713              **************************/
 714
 715             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 716
 717             /* Compute parameters for interactions between i and j atoms */
 718             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 719             gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 720
 721             /* Calculate table index by multiplying r with table scale and truncate to integer */
 722             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 723             itab_tmp         = _fjsp_dtox_v2r8(rt);
 724             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 725             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 726             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 727
 728             vfconv.i[0]     *= 8;
 729             vfconv.i[1]     *= 8;
 730
 731             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 732             isaprod          = _fjsp_mul_v2r8(isai0,isaj0);
 733             gbqqfactor       = _fjsp_neg_v2r8(_fjsp_mul_v2r8(qq00,_fjsp_mul_v2r8(isaprod,gbinvepsdiff)));
 734             gbscale          = _fjsp_mul_v2r8(isaprod,gbtabscale);
 735
 736             /* Calculate generalized born table index - this is a separate table from the normal one,
 737              * but we use the same procedure by multiplying r with scale and truncating to integer.
 738              */
 739             rt               = _fjsp_mul_v2r8(r00,gbscale);
 740             itab_tmp         = _fjsp_dtox_v2r8(rt);
 741             gbeps            = _fjsp_sub_v2r8(rt,_fjsp_xtod_v2r8(itab_tmp));
 742             _fjsp_store_v2r8(&gbconv.simd,itab_tmp);
 743
 744             Y                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] );
 745             F                = _fjsp_setzero_v2r8();
 746             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 747             G                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] +2);
 748             H                = _fjsp_setzero_v2r8();
 749             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 750             Fp               = _fjsp_madd_v2r8(gbeps,_fjsp_madd_v2r8(gbeps,H,G),F);
 751             VV               = _fjsp_madd_v2r8(gbeps,Fp,Y);
 752             vgb              = _fjsp_mul_v2r8(gbqqfactor,VV);
 753
 754             twogbeps         = _fjsp_add_v2r8(gbeps,gbeps);
 755             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twogbeps,H,G),gbeps,Fp);
 756             fgb              = _fjsp_mul_v2r8(gbqqfactor,_fjsp_mul_v2r8(FF,gbscale));
 757             dvdatmp          = _fjsp_mul_v2r8(minushalf,_fjsp_madd_v2r8(fgb,r00,vgb));
 758             dvdasum          = _fjsp_add_v2r8(dvdasum,dvdatmp);
 759             gmx_fjsp_increment_1real_v2r8(dvda+jnrA,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj0,isaj0)));
 760             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 761             felec            = _fjsp_mul_v2r8(_fjsp_msub_v2r8(velec,rinv00,fgb),rinv00);
 762
 763             /* CUBIC SPLINE TABLE DISPERSION */
 764             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 765             F                = _fjsp_setzero_v2r8();
 766             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 767             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 768             H                = _fjsp_setzero_v2r8();
 769             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 770             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 771             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 772             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 773
 774             /* CUBIC SPLINE TABLE REPULSION */
 775             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 776             F                = _fjsp_setzero_v2r8();
 777             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 778             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 779             H                = _fjsp_setzero_v2r8();
 780             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 781             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 782             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 783             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 784             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 785
 786             fscal            = _fjsp_add_v2r8(felec,fvdw);
 787
 788             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 789
 790             /* Update vectorial force */
 791             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 792             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 793             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 794
 795             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 796
 797             /* Inner loop uses 85 flops */
 798         }
 799
 800         /* End of innermost loop */
 801
 802         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 803                                               f+i_coord_offset,fshift+i_shift_offset);
 804
 805         dvdasum = _fjsp_mul_v2r8(dvdasum, _fjsp_mul_v2r8(isai0,isai0));
 806         gmx_fjsp_update_1pot_v2r8(dvdasum,dvda+inr);
 807
 808         /* Increment number of inner iterations */
 809         inneriter                  += j_index_end - j_index_start;
 810
 811         /* Outer loop uses 7 flops */
 812     }
 813
 814     /* Increment number of outer iterations */
 815     outeriter        += nri;
 816
 817     /* Update outer/inner flops */
 818
 819     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*85);
 820 }