src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecRF_VdwCSTab_GeomP1P1_sparc64_hpc_ace_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sparc64_hpc_ace_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_sparc64_hpc_ace_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomP1P1_VF_sparc64_hpc_ace_double
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRF_VdwCSTab_GeomP1P1_VF_sparc64_hpc_ace_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwjidx0A,vdwjidx0B;
  82     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  83     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  84     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
  85     real             *charge;
  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     facel            = gmx_fjsp_set1_v2r8(fr->ic->epsfac);
 112     charge           = mdatoms->chargeA;
 113     krf              = gmx_fjsp_set1_v2r8(fr->ic->k_rf);
 114     krf2             = gmx_fjsp_set1_v2r8(fr->ic->k_rf*2.0);
 115     crf              = gmx_fjsp_set1_v2r8(fr->ic->c_rf);
 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     /* Avoid stupid compiler warnings */
 124     jnrA = jnrB = 0;
 125     j_coord_offsetA = 0;
 126     j_coord_offsetB = 0;
 127
 128     outeriter        = 0;
 129     inneriter        = 0;
 130
 131     /* Start outer loop over neighborlists */
 132     for(iidx=0; iidx<nri; iidx++)
 133     {
 134         /* Load shift vector for this list */
 135         i_shift_offset   = DIM*shiftidx[iidx];
 136
 137         /* Load limits for loop over neighbors */
 138         j_index_start    = jindex[iidx];
 139         j_index_end      = jindex[iidx+1];
 140
 141         /* Get outer coordinate index */
 142         inr              = iinr[iidx];
 143         i_coord_offset   = DIM*inr;
 144
 145         /* Load i particle coords and add shift vector */
 146         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 147
 148         fix0             = _fjsp_setzero_v2r8();
 149         fiy0             = _fjsp_setzero_v2r8();
 150         fiz0             = _fjsp_setzero_v2r8();
 151
 152         /* Load parameters for i particles */
 153         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 154         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 155
 156         /* Reset potential sums */
 157         velecsum         = _fjsp_setzero_v2r8();
 158         vvdwsum          = _fjsp_setzero_v2r8();
 159
 160         /* Start inner kernel loop */
 161         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 162         {
 163
 164             /* Get j neighbor index, and coordinate index */
 165             jnrA             = jjnr[jidx];
 166             jnrB             = jjnr[jidx+1];
 167             j_coord_offsetA  = DIM*jnrA;
 168             j_coord_offsetB  = DIM*jnrB;
 169
 170             /* load j atom coordinates */
 171             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 172                                               &jx0,&jy0,&jz0);
 173
 174             /* Calculate displacement vector */
 175             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 176             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 177             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 178
 179             /* Calculate squared distance and things based on it */
 180             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 181
 182             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 183
 184             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 185
 186             /* Load parameters for j particles */
 187             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 188             vdwjidx0A        = 2*vdwtype[jnrA+0];
 189             vdwjidx0B        = 2*vdwtype[jnrB+0];
 190
 191             /**************************
 192              * CALCULATE INTERACTIONS *
 193              **************************/
 194
 195             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 196
 197             /* Compute parameters for interactions between i and j atoms */
 198             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 199             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 200                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 201
 202             /* Calculate table index by multiplying r with table scale and truncate to integer */
 203             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 204             itab_tmp         = _fjsp_dtox_v2r8(rt);
 205             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 206             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 207             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 208
 209             vfconv.i[0]     *= 8;
 210             vfconv.i[1]     *= 8;
 211
 212             /* REACTION-FIELD ELECTROSTATICS */
 213             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(_fjsp_madd_v2r8(krf,rsq00,rinv00),crf));
 214             felec            = _fjsp_mul_v2r8(qq00,_fjsp_msub_v2r8(rinv00,rinvsq00,krf2));
 215
 216             /* CUBIC SPLINE TABLE DISPERSION */
 217             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 218             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 219             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 220             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 221             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 2 );
 222             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 223             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 224             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 225             vvdw6            = _fjsp_mul_v2r8(c6_00,VV);
 226             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 227             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 228
 229             /* CUBIC SPLINE TABLE REPULSION */
 230             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 231             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 4 );
 232             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 233             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 234             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 6 );
 235             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 236             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 237             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 238             vvdw12           = _fjsp_mul_v2r8(c12_00,VV);
 239             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 240             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 241             vvdw             = _fjsp_add_v2r8(vvdw12,vvdw6);
 242             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 243
 244             /* Update potential sum for this i atom from the interaction with this j atom. */
 245             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 246             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 247
 248             fscal            = _fjsp_add_v2r8(felec,fvdw);
 249
 250             /* Update vectorial force */
 251             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 252             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 253             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 254
 255             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 256
 257             /* Inner loop uses 70 flops */
 258         }
 259
 260         if(jidx<j_index_end)
 261         {
 262
 263             jnrA             = jjnr[jidx];
 264             j_coord_offsetA  = DIM*jnrA;
 265
 266             /* load j atom coordinates */
 267             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 268                                               &jx0,&jy0,&jz0);
 269
 270             /* Calculate displacement vector */
 271             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 272             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 273             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 274
 275             /* Calculate squared distance and things based on it */
 276             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 277
 278             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 279
 280             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 281
 282             /* Load parameters for j particles */
 283             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 284             vdwjidx0A        = 2*vdwtype[jnrA+0];
 285
 286             /**************************
 287              * CALCULATE INTERACTIONS *
 288              **************************/
 289
 290             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 291
 292             /* Compute parameters for interactions between i and j atoms */
 293             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 294             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 295                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 296
 297             /* Calculate table index by multiplying r with table scale and truncate to integer */
 298             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 299             itab_tmp         = _fjsp_dtox_v2r8(rt);
 300             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 301             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 302             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 303
 304             vfconv.i[0]     *= 8;
 305             vfconv.i[1]     *= 8;
 306
 307             /* REACTION-FIELD ELECTROSTATICS */
 308             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(_fjsp_madd_v2r8(krf,rsq00,rinv00),crf));
 309             felec            = _fjsp_mul_v2r8(qq00,_fjsp_msub_v2r8(rinv00,rinvsq00,krf2));
 310
 311             /* CUBIC SPLINE TABLE DISPERSION */
 312             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 313             F                = _fjsp_setzero_v2r8();
 314             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 315             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 316             H                = _fjsp_setzero_v2r8();
 317             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 318             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 319             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 320             vvdw6            = _fjsp_mul_v2r8(c6_00,VV);
 321             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 322             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 323
 324             /* CUBIC SPLINE TABLE REPULSION */
 325             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 326             F                = _fjsp_setzero_v2r8();
 327             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 328             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 329             H                = _fjsp_setzero_v2r8();
 330             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 331             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 332             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 333             vvdw12           = _fjsp_mul_v2r8(c12_00,VV);
 334             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 335             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 336             vvdw             = _fjsp_add_v2r8(vvdw12,vvdw6);
 337             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 338
 339             /* Update potential sum for this i atom from the interaction with this j atom. */
 340             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 341             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 342             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 343             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 344
 345             fscal            = _fjsp_add_v2r8(felec,fvdw);
 346
 347             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 348
 349             /* Update vectorial force */
 350             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 351             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 352             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 353
 354             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 355
 356             /* Inner loop uses 70 flops */
 357         }
 358
 359         /* End of innermost loop */
 360
 361         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 362                                               f+i_coord_offset,fshift+i_shift_offset);
 363
 364         ggid                        = gid[iidx];
 365         /* Update potential energies */
 366         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 367         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 368
 369         /* Increment number of inner iterations */
 370         inneriter                  += j_index_end - j_index_start;
 371
 372         /* Outer loop uses 9 flops */
 373     }
 374
 375     /* Increment number of outer iterations */
 376     outeriter        += nri;
 377
 378     /* Update outer/inner flops */
 379
 380     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*70);
 381 }
 382 /*
 383  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_sparc64_hpc_ace_double
 384  * Electrostatics interaction: ReactionField
 385  * VdW interaction:            CubicSplineTable
 386  * Geometry:                   Particle-Particle
 387  * Calculate force/pot:        Force
 388  */
 389 void
 390 nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_sparc64_hpc_ace_double
 391                     (t_nblist                    * gmx_restrict       nlist,
 392                      rvec                        * gmx_restrict          xx,
 393                      rvec                        * gmx_restrict          ff,
 394                      struct t_forcerec           * gmx_restrict          fr,
 395                      t_mdatoms                   * gmx_restrict     mdatoms,
 396                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 397                      t_nrnb                      * gmx_restrict        nrnb)
 398 {
 399     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 400      * just 0 for non-waters.
 401      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 402      * jnr indices corresponding to data put in the four positions in the SIMD register.
 403      */
 404     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 405     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 406     int              jnrA,jnrB;
 407     int              j_coord_offsetA,j_coord_offsetB;
 408     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 409     real             rcutoff_scalar;
 410     real             *shiftvec,*fshift,*x,*f;
 411     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 412     int              vdwioffset0;
 413     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 414     int              vdwjidx0A,vdwjidx0B;
 415     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 416     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 417     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 418     real             *charge;
 419     int              nvdwtype;
 420     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 421     int              *vdwtype;
 422     real             *vdwparam;
 423     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 424     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 425     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 426     real             *vftab;
 427     _fjsp_v2r8       itab_tmp;
 428     _fjsp_v2r8       dummy_mask,cutoff_mask;
 429     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 430     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 431     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 432
 433     x                = xx[0];
 434     f                = ff[0];
 435
 436     nri              = nlist->nri;
 437     iinr             = nlist->iinr;
 438     jindex           = nlist->jindex;
 439     jjnr             = nlist->jjnr;
 440     shiftidx         = nlist->shift;
 441     gid              = nlist->gid;
 442     shiftvec         = fr->shift_vec[0];
 443     fshift           = fr->fshift[0];
 444     facel            = gmx_fjsp_set1_v2r8(fr->ic->epsfac);
 445     charge           = mdatoms->chargeA;
 446     krf              = gmx_fjsp_set1_v2r8(fr->ic->k_rf);
 447     krf2             = gmx_fjsp_set1_v2r8(fr->ic->k_rf*2.0);
 448     crf              = gmx_fjsp_set1_v2r8(fr->ic->c_rf);
 449     nvdwtype         = fr->ntype;
 450     vdwparam         = fr->nbfp;
 451     vdwtype          = mdatoms->typeA;
 452
 453     vftab            = kernel_data->table_vdw->data;
 454     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_vdw->scale);
 455
 456     /* Avoid stupid compiler warnings */
 457     jnrA = jnrB = 0;
 458     j_coord_offsetA = 0;
 459     j_coord_offsetB = 0;
 460
 461     outeriter        = 0;
 462     inneriter        = 0;
 463
 464     /* Start outer loop over neighborlists */
 465     for(iidx=0; iidx<nri; iidx++)
 466     {
 467         /* Load shift vector for this list */
 468         i_shift_offset   = DIM*shiftidx[iidx];
 469
 470         /* Load limits for loop over neighbors */
 471         j_index_start    = jindex[iidx];
 472         j_index_end      = jindex[iidx+1];
 473
 474         /* Get outer coordinate index */
 475         inr              = iinr[iidx];
 476         i_coord_offset   = DIM*inr;
 477
 478         /* Load i particle coords and add shift vector */
 479         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 480
 481         fix0             = _fjsp_setzero_v2r8();
 482         fiy0             = _fjsp_setzero_v2r8();
 483         fiz0             = _fjsp_setzero_v2r8();
 484
 485         /* Load parameters for i particles */
 486         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 487         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 488
 489         /* Start inner kernel loop */
 490         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 491         {
 492
 493             /* Get j neighbor index, and coordinate index */
 494             jnrA             = jjnr[jidx];
 495             jnrB             = jjnr[jidx+1];
 496             j_coord_offsetA  = DIM*jnrA;
 497             j_coord_offsetB  = DIM*jnrB;
 498
 499             /* load j atom coordinates */
 500             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 501                                               &jx0,&jy0,&jz0);
 502
 503             /* Calculate displacement vector */
 504             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 505             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 506             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 507
 508             /* Calculate squared distance and things based on it */
 509             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 510
 511             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 512
 513             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 514
 515             /* Load parameters for j particles */
 516             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 517             vdwjidx0A        = 2*vdwtype[jnrA+0];
 518             vdwjidx0B        = 2*vdwtype[jnrB+0];
 519
 520             /**************************
 521              * CALCULATE INTERACTIONS *
 522              **************************/
 523
 524             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 525
 526             /* Compute parameters for interactions between i and j atoms */
 527             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 528             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 529                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 530
 531             /* Calculate table index by multiplying r with table scale and truncate to integer */
 532             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 533             itab_tmp         = _fjsp_dtox_v2r8(rt);
 534             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 535             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 536             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 537
 538             vfconv.i[0]     *= 8;
 539             vfconv.i[1]     *= 8;
 540
 541             /* REACTION-FIELD ELECTROSTATICS */
 542             felec            = _fjsp_mul_v2r8(qq00,_fjsp_msub_v2r8(rinv00,rinvsq00,krf2));
 543
 544             /* CUBIC SPLINE TABLE DISPERSION */
 545             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 546             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 547             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 548             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 549             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 2 );
 550             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 551             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 552             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 553             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 554
 555             /* CUBIC SPLINE TABLE REPULSION */
 556             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 557             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 4 );
 558             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 559             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 560             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 6 );
 561             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 562             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 563             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 564             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 565             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 566
 567             fscal            = _fjsp_add_v2r8(felec,fvdw);
 568
 569             /* Update vectorial force */
 570             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 571             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 572             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 573
 574             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 575
 576             /* Inner loop uses 57 flops */
 577         }
 578
 579         if(jidx<j_index_end)
 580         {
 581
 582             jnrA             = jjnr[jidx];
 583             j_coord_offsetA  = DIM*jnrA;
 584
 585             /* load j atom coordinates */
 586             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 587                                               &jx0,&jy0,&jz0);
 588
 589             /* Calculate displacement vector */
 590             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 591             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 592             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 593
 594             /* Calculate squared distance and things based on it */
 595             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 596
 597             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 598
 599             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 600
 601             /* Load parameters for j particles */
 602             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 603             vdwjidx0A        = 2*vdwtype[jnrA+0];
 604
 605             /**************************
 606              * CALCULATE INTERACTIONS *
 607              **************************/
 608
 609             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 610
 611             /* Compute parameters for interactions between i and j atoms */
 612             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 613             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 614                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 615
 616             /* Calculate table index by multiplying r with table scale and truncate to integer */
 617             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 618             itab_tmp         = _fjsp_dtox_v2r8(rt);
 619             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 620             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 621             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 622
 623             vfconv.i[0]     *= 8;
 624             vfconv.i[1]     *= 8;
 625
 626             /* REACTION-FIELD ELECTROSTATICS */
 627             felec            = _fjsp_mul_v2r8(qq00,_fjsp_msub_v2r8(rinv00,rinvsq00,krf2));
 628
 629             /* CUBIC SPLINE TABLE DISPERSION */
 630             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 631             F                = _fjsp_setzero_v2r8();
 632             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 633             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 634             H                = _fjsp_setzero_v2r8();
 635             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 636             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 637             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 638             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 639
 640             /* CUBIC SPLINE TABLE REPULSION */
 641             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 642             F                = _fjsp_setzero_v2r8();
 643             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 644             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 645             H                = _fjsp_setzero_v2r8();
 646             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 647             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 648             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 649             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 650             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 651
 652             fscal            = _fjsp_add_v2r8(felec,fvdw);
 653
 654             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 655
 656             /* Update vectorial force */
 657             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 658             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 659             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 660
 661             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 662
 663             /* Inner loop uses 57 flops */
 664         }
 665
 666         /* End of innermost loop */
 667
 668         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 669                                               f+i_coord_offset,fshift+i_shift_offset);
 670
 671         /* Increment number of inner iterations */
 672         inneriter                  += j_index_end - j_index_start;
 673
 674         /* Outer loop uses 7 flops */
 675     }
 676
 677     /* Increment number of outer iterations */
 678     outeriter        += nri;
 679
 680     /* Update outer/inner flops */
 681
 682     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*57);
 683 }