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) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sparc64_hpc_ace_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "kernelutil_sparc64_hpc_ace_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwCSTab_GeomP1P1_VF_sparc64_hpc_ace_double
  51  * Electrostatics interaction: GeneralizedBorn
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecGB_VdwCSTab_GeomP1P1_VF_sparc64_hpc_ace_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      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     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
  85     real             *charge;
  86     _fjsp_v2r8       vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,twogbeps,dvdatmp;
  87     _fjsp_v2r8       minushalf = gmx_fjsp_set1_v2r8(-0.5);
  88     real             *invsqrta,*dvda,*gbtab;
  89     int              nvdwtype;
  90     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  91     int              *vdwtype;
  92     real             *vdwparam;
  93     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
  94     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
  95     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  96     real             *vftab;
  97     _fjsp_v2r8       itab_tmp;
  98     _fjsp_v2r8       dummy_mask,cutoff_mask;
  99     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 100     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 101     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 102
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 115     charge           = mdatoms->chargeA;
 116     nvdwtype         = fr->ntype;
 117     vdwparam         = fr->nbfp;
 118     vdwtype          = mdatoms->typeA;
 119
 120     vftab            = kernel_data->table_vdw->data;
 121     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_vdw->scale);
 122
 123     invsqrta         = fr->invsqrta;
 124     dvda             = fr->dvda;
 125     gbtabscale       = gmx_fjsp_set1_v2r8(fr->gbtab.scale);
 126     gbtab            = fr->gbtab.data;
 127     gbinvepsdiff     = gmx_fjsp_set1_v2r8((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 128
 129     /* Avoid stupid compiler warnings */
 130     jnrA = jnrB = 0;
 131     j_coord_offsetA = 0;
 132     j_coord_offsetB = 0;
 133
 134     outeriter        = 0;
 135     inneriter        = 0;
 136
 137     /* Start outer loop over neighborlists */
 138     for(iidx=0; iidx<nri; iidx++)
 139     {
 140         /* Load shift vector for this list */
 141         i_shift_offset   = DIM*shiftidx[iidx];
 142
 143         /* Load limits for loop over neighbors */
 144         j_index_start    = jindex[iidx];
 145         j_index_end      = jindex[iidx+1];
 146
 147         /* Get outer coordinate index */
 148         inr              = iinr[iidx];
 149         i_coord_offset   = DIM*inr;
 150
 151         /* Load i particle coords and add shift vector */
 152         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 153
 154         fix0             = _fjsp_setzero_v2r8();
 155         fiy0             = _fjsp_setzero_v2r8();
 156         fiz0             = _fjsp_setzero_v2r8();
 157
 158         /* Load parameters for i particles */
 159         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 160         isai0            = gmx_fjsp_load1_v2r8(invsqrta+inr+0);
 161         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 162
 163         /* Reset potential sums */
 164         velecsum         = _fjsp_setzero_v2r8();
 165         vgbsum           = _fjsp_setzero_v2r8();
 166         vvdwsum          = _fjsp_setzero_v2r8();
 167         dvdasum          = _fjsp_setzero_v2r8();
 168
 169         /* Start inner kernel loop */
 170         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 171         {
 172
 173             /* Get j neighbor index, and coordinate index */
 174             jnrA             = jjnr[jidx];
 175             jnrB             = jjnr[jidx+1];
 176             j_coord_offsetA  = DIM*jnrA;
 177             j_coord_offsetB  = DIM*jnrB;
 178
 179             /* load j atom coordinates */
 180             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 181                                               &jx0,&jy0,&jz0);
 182
 183             /* Calculate displacement vector */
 184             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 185             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 186             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 187
 188             /* Calculate squared distance and things based on it */
 189             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 190
 191             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 192
 193             /* Load parameters for j particles */
 194             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 195             isaj0            = gmx_fjsp_load_2real_swizzle_v2r8(invsqrta+jnrA+0,invsqrta+jnrB+0);
 196             vdwjidx0A        = 2*vdwtype[jnrA+0];
 197             vdwjidx0B        = 2*vdwtype[jnrB+0];
 198
 199             /**************************
 200              * CALCULATE INTERACTIONS *
 201              **************************/
 202
 203             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 204
 205             /* Compute parameters for interactions between i and j atoms */
 206             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 207             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 208                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 209
 210             /* Calculate table index by multiplying r with table scale and truncate to integer */
 211             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 212             itab_tmp         = _fjsp_dtox_v2r8(rt);
 213             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 214             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 215             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 216
 217             vfconv.i[0]     *= 8;
 218             vfconv.i[1]     *= 8;
 219
 220             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 221             isaprod          = _fjsp_mul_v2r8(isai0,isaj0);
 222             gbqqfactor       = _fjsp_neg_v2r8(_fjsp_mul_v2r8(qq00,_fjsp_mul_v2r8(isaprod,gbinvepsdiff)));
 223             gbscale          = _fjsp_mul_v2r8(isaprod,gbtabscale);
 224
 225             /* Calculate generalized born table index - this is a separate table from the normal one,
 226              * but we use the same procedure by multiplying r with scale and truncating to integer.
 227              */
 228             rt               = _fjsp_mul_v2r8(r00,gbscale);
 229             itab_tmp         = _fjsp_dtox_v2r8(rt);
 230             gbeps            = _fjsp_sub_v2r8(rt,_fjsp_xtod_v2r8(itab_tmp));
 231             _fjsp_store_v2r8(&gbconv.simd,itab_tmp);
 232
 233             Y                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] );
 234             F                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] );
 235             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 236             G                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] +2);
 237             H                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] +2);
 238             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 239             Fp               = _fjsp_madd_v2r8(gbeps,_fjsp_madd_v2r8(gbeps,H,G),F);
 240             VV               = _fjsp_madd_v2r8(gbeps,Fp,Y);
 241             vgb              = _fjsp_mul_v2r8(gbqqfactor,VV);
 242
 243             twogbeps         = _fjsp_add_v2r8(gbeps,gbeps);
 244             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twogbeps,H,G),gbeps,Fp);
 245             fgb              = _fjsp_mul_v2r8(gbqqfactor,_fjsp_mul_v2r8(FF,gbscale));
 246             dvdatmp          = _fjsp_mul_v2r8(minushalf,_fjsp_madd_v2r8(fgb,r00,vgb));
 247             dvdasum          = _fjsp_add_v2r8(dvdasum,dvdatmp);
 248             gmx_fjsp_increment_2real_swizzle_v2r8(dvda+jnrA,dvda+jnrB,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj0,isaj0)));
 249             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 250             felec            = _fjsp_mul_v2r8(_fjsp_msub_v2r8(velec,rinv00,fgb),rinv00);
 251
 252             /* CUBIC SPLINE TABLE DISPERSION */
 253             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 254             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 255             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 256             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 257             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 2 );
 258             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 259             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 260             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 261             vvdw6            = _fjsp_mul_v2r8(c6_00,VV);
 262             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 263             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 264
 265             /* CUBIC SPLINE TABLE REPULSION */
 266             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 267             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 4 );
 268             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 269             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 270             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 6 );
 271             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 272             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 273             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 274             vvdw12           = _fjsp_mul_v2r8(c12_00,VV);
 275             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 276             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 277             vvdw             = _fjsp_add_v2r8(vvdw12,vvdw6);
 278             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 279
 280             /* Update potential sum for this i atom from the interaction with this j atom. */
 281             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 282             vgbsum           = _fjsp_add_v2r8(vgbsum,vgb);
 283             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 284
 285             fscal            = _fjsp_add_v2r8(felec,fvdw);
 286
 287             /* Update vectorial force */
 288             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 289             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 290             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 291
 292             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 293
 294             /* Inner loop uses 95 flops */
 295         }
 296
 297         if(jidx<j_index_end)
 298         {
 299
 300             jnrA             = jjnr[jidx];
 301             j_coord_offsetA  = DIM*jnrA;
 302
 303             /* load j atom coordinates */
 304             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 305                                               &jx0,&jy0,&jz0);
 306
 307             /* Calculate displacement vector */
 308             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 309             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 310             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 311
 312             /* Calculate squared distance and things based on it */
 313             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 314
 315             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 316
 317             /* Load parameters for j particles */
 318             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 319             isaj0            = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),invsqrta+jnrA+0);
 320             vdwjidx0A        = 2*vdwtype[jnrA+0];
 321
 322             /**************************
 323              * CALCULATE INTERACTIONS *
 324              **************************/
 325
 326             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 327
 328             /* Compute parameters for interactions between i and j atoms */
 329             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 330             gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 331
 332             /* Calculate table index by multiplying r with table scale and truncate to integer */
 333             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 334             itab_tmp         = _fjsp_dtox_v2r8(rt);
 335             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 336             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 337             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 338
 339             vfconv.i[0]     *= 8;
 340             vfconv.i[1]     *= 8;
 341
 342             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 343             isaprod          = _fjsp_mul_v2r8(isai0,isaj0);
 344             gbqqfactor       = _fjsp_neg_v2r8(_fjsp_mul_v2r8(qq00,_fjsp_mul_v2r8(isaprod,gbinvepsdiff)));
 345             gbscale          = _fjsp_mul_v2r8(isaprod,gbtabscale);
 346
 347             /* Calculate generalized born table index - this is a separate table from the normal one,
 348              * but we use the same procedure by multiplying r with scale and truncating to integer.
 349              */
 350             rt               = _fjsp_mul_v2r8(r00,gbscale);
 351             itab_tmp         = _fjsp_dtox_v2r8(rt);
 352             gbeps            = _fjsp_sub_v2r8(rt,_fjsp_xtod_v2r8(itab_tmp));
 353             _fjsp_store_v2r8(&gbconv.simd,itab_tmp);
 354
 355             Y                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] );
 356             F                = _fjsp_setzero_v2r8();
 357             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 358             G                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] +2);
 359             H                = _fjsp_setzero_v2r8();
 360             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 361             Fp               = _fjsp_madd_v2r8(gbeps,_fjsp_madd_v2r8(gbeps,H,G),F);
 362             VV               = _fjsp_madd_v2r8(gbeps,Fp,Y);
 363             vgb              = _fjsp_mul_v2r8(gbqqfactor,VV);
 364
 365             twogbeps         = _fjsp_add_v2r8(gbeps,gbeps);
 366             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twogbeps,H,G),gbeps,Fp);
 367             fgb              = _fjsp_mul_v2r8(gbqqfactor,_fjsp_mul_v2r8(FF,gbscale));
 368             dvdatmp          = _fjsp_mul_v2r8(minushalf,_fjsp_madd_v2r8(fgb,r00,vgb));
 369             dvdasum          = _fjsp_add_v2r8(dvdasum,dvdatmp);
 370             gmx_fjsp_increment_1real_v2r8(dvda+jnrA,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj0,isaj0)));
 371             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 372             felec            = _fjsp_mul_v2r8(_fjsp_msub_v2r8(velec,rinv00,fgb),rinv00);
 373
 374             /* CUBIC SPLINE TABLE DISPERSION */
 375             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 376             F                = _fjsp_setzero_v2r8();
 377             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 378             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 379             H                = _fjsp_setzero_v2r8();
 380             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 381             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 382             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 383             vvdw6            = _fjsp_mul_v2r8(c6_00,VV);
 384             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 385             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 386
 387             /* CUBIC SPLINE TABLE REPULSION */
 388             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 389             F                = _fjsp_setzero_v2r8();
 390             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 391             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 392             H                = _fjsp_setzero_v2r8();
 393             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 394             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 395             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 396             vvdw12           = _fjsp_mul_v2r8(c12_00,VV);
 397             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 398             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 399             vvdw             = _fjsp_add_v2r8(vvdw12,vvdw6);
 400             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 401
 402             /* Update potential sum for this i atom from the interaction with this j atom. */
 403             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 404             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 405             vgb              = _fjsp_unpacklo_v2r8(vgb,_fjsp_setzero_v2r8());
 406             vgbsum           = _fjsp_add_v2r8(vgbsum,vgb);
 407             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 408             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 409
 410             fscal            = _fjsp_add_v2r8(felec,fvdw);
 411
 412             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 413
 414             /* Update vectorial force */
 415             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 416             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 417             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 418
 419             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 420
 421             /* Inner loop uses 95 flops */
 422         }
 423
 424         /* End of innermost loop */
 425
 426         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 427                                               f+i_coord_offset,fshift+i_shift_offset);
 428
 429         ggid                        = gid[iidx];
 430         /* Update potential energies */
 431         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 432         gmx_fjsp_update_1pot_v2r8(vgbsum,kernel_data->energygrp_polarization+ggid);
 433         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 434         dvdasum = _fjsp_mul_v2r8(dvdasum, _fjsp_mul_v2r8(isai0,isai0));
 435         gmx_fjsp_update_1pot_v2r8(dvdasum,dvda+inr);
 436
 437         /* Increment number of inner iterations */
 438         inneriter                  += j_index_end - j_index_start;
 439
 440         /* Outer loop uses 10 flops */
 441     }
 442
 443     /* Increment number of outer iterations */
 444     outeriter        += nri;
 445
 446     /* Update outer/inner flops */
 447
 448     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*10 + inneriter*95);
 449 }
 450 /*
 451  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwCSTab_GeomP1P1_F_sparc64_hpc_ace_double
 452  * Electrostatics interaction: GeneralizedBorn
 453  * VdW interaction:            CubicSplineTable
 454  * Geometry:                   Particle-Particle
 455  * Calculate force/pot:        Force
 456  */
 457 void
 458 nb_kernel_ElecGB_VdwCSTab_GeomP1P1_F_sparc64_hpc_ace_double
 459                     (t_nblist                    * gmx_restrict       nlist,
 460                      rvec                        * gmx_restrict          xx,
 461                      rvec                        * gmx_restrict          ff,
 462                      t_forcerec                  * gmx_restrict          fr,
 463                      t_mdatoms                   * gmx_restrict     mdatoms,
 464                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 465                      t_nrnb                      * gmx_restrict        nrnb)
 466 {
 467     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 468      * just 0 for non-waters.
 469      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 470      * jnr indices corresponding to data put in the four positions in the SIMD register.
 471      */
 472     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 473     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 474     int              jnrA,jnrB;
 475     int              j_coord_offsetA,j_coord_offsetB;
 476     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 477     real             rcutoff_scalar;
 478     real             *shiftvec,*fshift,*x,*f;
 479     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 480     int              vdwioffset0;
 481     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 482     int              vdwjidx0A,vdwjidx0B;
 483     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 484     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 485     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 486     real             *charge;
 487     _fjsp_v2r8       vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,twogbeps,dvdatmp;
 488     _fjsp_v2r8       minushalf = gmx_fjsp_set1_v2r8(-0.5);
 489     real             *invsqrta,*dvda,*gbtab;
 490     int              nvdwtype;
 491     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 492     int              *vdwtype;
 493     real             *vdwparam;
 494     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 495     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 496     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 497     real             *vftab;
 498     _fjsp_v2r8       itab_tmp;
 499     _fjsp_v2r8       dummy_mask,cutoff_mask;
 500     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 501     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 502     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 503
 504     x                = xx[0];
 505     f                = ff[0];
 506
 507     nri              = nlist->nri;
 508     iinr             = nlist->iinr;
 509     jindex           = nlist->jindex;
 510     jjnr             = nlist->jjnr;
 511     shiftidx         = nlist->shift;
 512     gid              = nlist->gid;
 513     shiftvec         = fr->shift_vec[0];
 514     fshift           = fr->fshift[0];
 515     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 516     charge           = mdatoms->chargeA;
 517     nvdwtype         = fr->ntype;
 518     vdwparam         = fr->nbfp;
 519     vdwtype          = mdatoms->typeA;
 520
 521     vftab            = kernel_data->table_vdw->data;
 522     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_vdw->scale);
 523
 524     invsqrta         = fr->invsqrta;
 525     dvda             = fr->dvda;
 526     gbtabscale       = gmx_fjsp_set1_v2r8(fr->gbtab.scale);
 527     gbtab            = fr->gbtab.data;
 528     gbinvepsdiff     = gmx_fjsp_set1_v2r8((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 529
 530     /* Avoid stupid compiler warnings */
 531     jnrA = jnrB = 0;
 532     j_coord_offsetA = 0;
 533     j_coord_offsetB = 0;
 534
 535     outeriter        = 0;
 536     inneriter        = 0;
 537
 538     /* Start outer loop over neighborlists */
 539     for(iidx=0; iidx<nri; iidx++)
 540     {
 541         /* Load shift vector for this list */
 542         i_shift_offset   = DIM*shiftidx[iidx];
 543
 544         /* Load limits for loop over neighbors */
 545         j_index_start    = jindex[iidx];
 546         j_index_end      = jindex[iidx+1];
 547
 548         /* Get outer coordinate index */
 549         inr              = iinr[iidx];
 550         i_coord_offset   = DIM*inr;
 551
 552         /* Load i particle coords and add shift vector */
 553         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 554
 555         fix0             = _fjsp_setzero_v2r8();
 556         fiy0             = _fjsp_setzero_v2r8();
 557         fiz0             = _fjsp_setzero_v2r8();
 558
 559         /* Load parameters for i particles */
 560         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 561         isai0            = gmx_fjsp_load1_v2r8(invsqrta+inr+0);
 562         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 563
 564         dvdasum          = _fjsp_setzero_v2r8();
 565
 566         /* Start inner kernel loop */
 567         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 568         {
 569
 570             /* Get j neighbor index, and coordinate index */
 571             jnrA             = jjnr[jidx];
 572             jnrB             = jjnr[jidx+1];
 573             j_coord_offsetA  = DIM*jnrA;
 574             j_coord_offsetB  = DIM*jnrB;
 575
 576             /* load j atom coordinates */
 577             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 578                                               &jx0,&jy0,&jz0);
 579
 580             /* Calculate displacement vector */
 581             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 582             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 583             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 584
 585             /* Calculate squared distance and things based on it */
 586             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 587
 588             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 589
 590             /* Load parameters for j particles */
 591             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 592             isaj0            = gmx_fjsp_load_2real_swizzle_v2r8(invsqrta+jnrA+0,invsqrta+jnrB+0);
 593             vdwjidx0A        = 2*vdwtype[jnrA+0];
 594             vdwjidx0B        = 2*vdwtype[jnrB+0];
 595
 596             /**************************
 597              * CALCULATE INTERACTIONS *
 598              **************************/
 599
 600             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 601
 602             /* Compute parameters for interactions between i and j atoms */
 603             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 604             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 605                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 606
 607             /* Calculate table index by multiplying r with table scale and truncate to integer */
 608             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 609             itab_tmp         = _fjsp_dtox_v2r8(rt);
 610             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 611             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 612             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 613
 614             vfconv.i[0]     *= 8;
 615             vfconv.i[1]     *= 8;
 616
 617             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 618             isaprod          = _fjsp_mul_v2r8(isai0,isaj0);
 619             gbqqfactor       = _fjsp_neg_v2r8(_fjsp_mul_v2r8(qq00,_fjsp_mul_v2r8(isaprod,gbinvepsdiff)));
 620             gbscale          = _fjsp_mul_v2r8(isaprod,gbtabscale);
 621
 622             /* Calculate generalized born table index - this is a separate table from the normal one,
 623              * but we use the same procedure by multiplying r with scale and truncating to integer.
 624              */
 625             rt               = _fjsp_mul_v2r8(r00,gbscale);
 626             itab_tmp         = _fjsp_dtox_v2r8(rt);
 627             gbeps            = _fjsp_sub_v2r8(rt,_fjsp_xtod_v2r8(itab_tmp));
 628             _fjsp_store_v2r8(&gbconv.simd,itab_tmp);
 629
 630             Y                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] );
 631             F                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] );
 632             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 633             G                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] +2);
 634             H                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] +2);
 635             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 636             Fp               = _fjsp_madd_v2r8(gbeps,_fjsp_madd_v2r8(gbeps,H,G),F);
 637             VV               = _fjsp_madd_v2r8(gbeps,Fp,Y);
 638             vgb              = _fjsp_mul_v2r8(gbqqfactor,VV);
 639
 640             twogbeps         = _fjsp_add_v2r8(gbeps,gbeps);
 641             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twogbeps,H,G),gbeps,Fp);
 642             fgb              = _fjsp_mul_v2r8(gbqqfactor,_fjsp_mul_v2r8(FF,gbscale));
 643             dvdatmp          = _fjsp_mul_v2r8(minushalf,_fjsp_madd_v2r8(fgb,r00,vgb));
 644             dvdasum          = _fjsp_add_v2r8(dvdasum,dvdatmp);
 645             gmx_fjsp_increment_2real_swizzle_v2r8(dvda+jnrA,dvda+jnrB,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj0,isaj0)));
 646             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 647             felec            = _fjsp_mul_v2r8(_fjsp_msub_v2r8(velec,rinv00,fgb),rinv00);
 648
 649             /* CUBIC SPLINE TABLE DISPERSION */
 650             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 651             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 652             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 653             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 654             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 2 );
 655             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 656             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 657             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 658             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 659
 660             /* CUBIC SPLINE TABLE REPULSION */
 661             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 662             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 4 );
 663             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 664             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 665             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 6 );
 666             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 667             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 668             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 669             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 670             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 671
 672             fscal            = _fjsp_add_v2r8(felec,fvdw);
 673
 674             /* Update vectorial force */
 675             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 676             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 677             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 678
 679             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 680
 681             /* Inner loop uses 85 flops */
 682         }
 683
 684         if(jidx<j_index_end)
 685         {
 686
 687             jnrA             = jjnr[jidx];
 688             j_coord_offsetA  = DIM*jnrA;
 689
 690             /* load j atom coordinates */
 691             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 692                                               &jx0,&jy0,&jz0);
 693
 694             /* Calculate displacement vector */
 695             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 696             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 697             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 698
 699             /* Calculate squared distance and things based on it */
 700             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 701
 702             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 703
 704             /* Load parameters for j particles */
 705             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 706             isaj0            = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),invsqrta+jnrA+0);
 707             vdwjidx0A        = 2*vdwtype[jnrA+0];
 708
 709             /**************************
 710              * CALCULATE INTERACTIONS *
 711              **************************/
 712
 713             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 714
 715             /* Compute parameters for interactions between i and j atoms */
 716             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 717             gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 718
 719             /* Calculate table index by multiplying r with table scale and truncate to integer */
 720             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 721             itab_tmp         = _fjsp_dtox_v2r8(rt);
 722             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 723             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 724             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 725
 726             vfconv.i[0]     *= 8;
 727             vfconv.i[1]     *= 8;
 728
 729             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 730             isaprod          = _fjsp_mul_v2r8(isai0,isaj0);
 731             gbqqfactor       = _fjsp_neg_v2r8(_fjsp_mul_v2r8(qq00,_fjsp_mul_v2r8(isaprod,gbinvepsdiff)));
 732             gbscale          = _fjsp_mul_v2r8(isaprod,gbtabscale);
 733
 734             /* Calculate generalized born table index - this is a separate table from the normal one,
 735              * but we use the same procedure by multiplying r with scale and truncating to integer.
 736              */
 737             rt               = _fjsp_mul_v2r8(r00,gbscale);
 738             itab_tmp         = _fjsp_dtox_v2r8(rt);
 739             gbeps            = _fjsp_sub_v2r8(rt,_fjsp_xtod_v2r8(itab_tmp));
 740             _fjsp_store_v2r8(&gbconv.simd,itab_tmp);
 741
 742             Y                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] );
 743             F                = _fjsp_setzero_v2r8();
 744             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 745             G                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] +2);
 746             H                = _fjsp_setzero_v2r8();
 747             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 748             Fp               = _fjsp_madd_v2r8(gbeps,_fjsp_madd_v2r8(gbeps,H,G),F);
 749             VV               = _fjsp_madd_v2r8(gbeps,Fp,Y);
 750             vgb              = _fjsp_mul_v2r8(gbqqfactor,VV);
 751
 752             twogbeps         = _fjsp_add_v2r8(gbeps,gbeps);
 753             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twogbeps,H,G),gbeps,Fp);
 754             fgb              = _fjsp_mul_v2r8(gbqqfactor,_fjsp_mul_v2r8(FF,gbscale));
 755             dvdatmp          = _fjsp_mul_v2r8(minushalf,_fjsp_madd_v2r8(fgb,r00,vgb));
 756             dvdasum          = _fjsp_add_v2r8(dvdasum,dvdatmp);
 757             gmx_fjsp_increment_1real_v2r8(dvda+jnrA,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj0,isaj0)));
 758             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 759             felec            = _fjsp_mul_v2r8(_fjsp_msub_v2r8(velec,rinv00,fgb),rinv00);
 760
 761             /* CUBIC SPLINE TABLE DISPERSION */
 762             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 763             F                = _fjsp_setzero_v2r8();
 764             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 765             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 766             H                = _fjsp_setzero_v2r8();
 767             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 768             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 769             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 770             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 771
 772             /* CUBIC SPLINE TABLE REPULSION */
 773             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 774             F                = _fjsp_setzero_v2r8();
 775             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 776             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 777             H                = _fjsp_setzero_v2r8();
 778             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 779             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 780             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 781             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 782             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 783
 784             fscal            = _fjsp_add_v2r8(felec,fvdw);
 785
 786             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 787
 788             /* Update vectorial force */
 789             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 790             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 791             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 792
 793             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 794
 795             /* Inner loop uses 85 flops */
 796         }
 797
 798         /* End of innermost loop */
 799
 800         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 801                                               f+i_coord_offset,fshift+i_shift_offset);
 802
 803         dvdasum = _fjsp_mul_v2r8(dvdasum, _fjsp_mul_v2r8(isai0,isai0));
 804         gmx_fjsp_update_1pot_v2r8(dvdasum,dvda+inr);
 805
 806         /* Increment number of inner iterations */
 807         inneriter                  += j_index_end - j_index_start;
 808
 809         /* Outer loop uses 7 flops */
 810     }
 811
 812     /* Increment number of outer iterations */
 813     outeriter        += nri;
 814
 815     /* Update outer/inner flops */
 816
 817     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*85);
 818 }