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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sparc64_hpc_ace_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "gromacs/legacyheaders/vec.h"
  47 #include "nrnb.h"
  48
  49 #include "kernelutil_sparc64_hpc_ace_double.h"
  50
  51 /*
  52  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomP1P1_VF_sparc64_hpc_ace_double
  53  * Electrostatics interaction: ReactionField
  54  * VdW interaction:            CubicSplineTable
  55  * Geometry:                   Particle-Particle
  56  * Calculate force/pot:        PotentialAndForce
  57  */
  58 void
  59 nb_kernel_ElecRF_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     int              nvdwtype;
  89     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  90     int              *vdwtype;
  91     real             *vdwparam;
  92     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
  93     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
  94     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  95     real             *vftab;
  96     _fjsp_v2r8       itab_tmp;
  97     _fjsp_v2r8       dummy_mask,cutoff_mask;
  98     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
  99     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 100     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 101
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115     krf              = gmx_fjsp_set1_v2r8(fr->ic->k_rf);
 116     krf2             = gmx_fjsp_set1_v2r8(fr->ic->k_rf*2.0);
 117     crf              = gmx_fjsp_set1_v2r8(fr->ic->c_rf);
 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     /* Avoid stupid compiler warnings */
 126     jnrA = jnrB = 0;
 127     j_coord_offsetA = 0;
 128     j_coord_offsetB = 0;
 129
 130     outeriter        = 0;
 131     inneriter        = 0;
 132
 133     /* Start outer loop over neighborlists */
 134     for(iidx=0; iidx<nri; iidx++)
 135     {
 136         /* Load shift vector for this list */
 137         i_shift_offset   = DIM*shiftidx[iidx];
 138
 139         /* Load limits for loop over neighbors */
 140         j_index_start    = jindex[iidx];
 141         j_index_end      = jindex[iidx+1];
 142
 143         /* Get outer coordinate index */
 144         inr              = iinr[iidx];
 145         i_coord_offset   = DIM*inr;
 146
 147         /* Load i particle coords and add shift vector */
 148         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 149
 150         fix0             = _fjsp_setzero_v2r8();
 151         fiy0             = _fjsp_setzero_v2r8();
 152         fiz0             = _fjsp_setzero_v2r8();
 153
 154         /* Load parameters for i particles */
 155         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 156         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 157
 158         /* Reset potential sums */
 159         velecsum         = _fjsp_setzero_v2r8();
 160         vvdwsum          = _fjsp_setzero_v2r8();
 161
 162         /* Start inner kernel loop */
 163         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 164         {
 165
 166             /* Get j neighbor index, and coordinate index */
 167             jnrA             = jjnr[jidx];
 168             jnrB             = jjnr[jidx+1];
 169             j_coord_offsetA  = DIM*jnrA;
 170             j_coord_offsetB  = DIM*jnrB;
 171
 172             /* load j atom coordinates */
 173             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 174                                               &jx0,&jy0,&jz0);
 175
 176             /* Calculate displacement vector */
 177             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 178             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 179             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 180
 181             /* Calculate squared distance and things based on it */
 182             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 183
 184             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 185
 186             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 187
 188             /* Load parameters for j particles */
 189             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 190             vdwjidx0A        = 2*vdwtype[jnrA+0];
 191             vdwjidx0B        = 2*vdwtype[jnrB+0];
 192
 193             /**************************
 194              * CALCULATE INTERACTIONS *
 195              **************************/
 196
 197             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 198
 199             /* Compute parameters for interactions between i and j atoms */
 200             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 201             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 202                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 203
 204             /* Calculate table index by multiplying r with table scale and truncate to integer */
 205             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 206             itab_tmp         = _fjsp_dtox_v2r8(rt);
 207             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 208             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 209             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 210
 211             vfconv.i[0]     *= 8;
 212             vfconv.i[1]     *= 8;
 213
 214             /* REACTION-FIELD ELECTROSTATICS */
 215             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(_fjsp_madd_v2r8(krf,rsq00,rinv00),crf));
 216             felec            = _fjsp_mul_v2r8(qq00,_fjsp_msub_v2r8(rinv00,rinvsq00,krf2));
 217
 218             /* CUBIC SPLINE TABLE DISPERSION */
 219             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 220             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 221             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 222             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 223             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 2 );
 224             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 225             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 226             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 227             vvdw6            = _fjsp_mul_v2r8(c6_00,VV);
 228             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 229             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 230
 231             /* CUBIC SPLINE TABLE REPULSION */
 232             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 233             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 4 );
 234             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 235             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 236             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 6 );
 237             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 238             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 239             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 240             vvdw12           = _fjsp_mul_v2r8(c12_00,VV);
 241             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 242             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 243             vvdw             = _fjsp_add_v2r8(vvdw12,vvdw6);
 244             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 245
 246             /* Update potential sum for this i atom from the interaction with this j atom. */
 247             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 248             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 249
 250             fscal            = _fjsp_add_v2r8(felec,fvdw);
 251
 252             /* Update vectorial force */
 253             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 254             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 255             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 256
 257             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 258
 259             /* Inner loop uses 70 flops */
 260         }
 261
 262         if(jidx<j_index_end)
 263         {
 264
 265             jnrA             = jjnr[jidx];
 266             j_coord_offsetA  = DIM*jnrA;
 267
 268             /* load j atom coordinates */
 269             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 270                                               &jx0,&jy0,&jz0);
 271
 272             /* Calculate displacement vector */
 273             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 274             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 275             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 276
 277             /* Calculate squared distance and things based on it */
 278             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 279
 280             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 281
 282             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 283
 284             /* Load parameters for j particles */
 285             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 286             vdwjidx0A        = 2*vdwtype[jnrA+0];
 287
 288             /**************************
 289              * CALCULATE INTERACTIONS *
 290              **************************/
 291
 292             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 293
 294             /* Compute parameters for interactions between i and j atoms */
 295             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 296             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 297                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 298
 299             /* Calculate table index by multiplying r with table scale and truncate to integer */
 300             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 301             itab_tmp         = _fjsp_dtox_v2r8(rt);
 302             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 303             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 304             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 305
 306             vfconv.i[0]     *= 8;
 307             vfconv.i[1]     *= 8;
 308
 309             /* REACTION-FIELD ELECTROSTATICS */
 310             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(_fjsp_madd_v2r8(krf,rsq00,rinv00),crf));
 311             felec            = _fjsp_mul_v2r8(qq00,_fjsp_msub_v2r8(rinv00,rinvsq00,krf2));
 312
 313             /* CUBIC SPLINE TABLE DISPERSION */
 314             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 315             F                = _fjsp_setzero_v2r8();
 316             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 317             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 318             H                = _fjsp_setzero_v2r8();
 319             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 320             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 321             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 322             vvdw6            = _fjsp_mul_v2r8(c6_00,VV);
 323             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 324             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 325
 326             /* CUBIC SPLINE TABLE REPULSION */
 327             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 328             F                = _fjsp_setzero_v2r8();
 329             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 330             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 331             H                = _fjsp_setzero_v2r8();
 332             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 333             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 334             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 335             vvdw12           = _fjsp_mul_v2r8(c12_00,VV);
 336             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 337             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 338             vvdw             = _fjsp_add_v2r8(vvdw12,vvdw6);
 339             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 340
 341             /* Update potential sum for this i atom from the interaction with this j atom. */
 342             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 343             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 344             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 345             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 346
 347             fscal            = _fjsp_add_v2r8(felec,fvdw);
 348
 349             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 350
 351             /* Update vectorial force */
 352             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 353             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 354             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 355
 356             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 357
 358             /* Inner loop uses 70 flops */
 359         }
 360
 361         /* End of innermost loop */
 362
 363         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 364                                               f+i_coord_offset,fshift+i_shift_offset);
 365
 366         ggid                        = gid[iidx];
 367         /* Update potential energies */
 368         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 369         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 370
 371         /* Increment number of inner iterations */
 372         inneriter                  += j_index_end - j_index_start;
 373
 374         /* Outer loop uses 9 flops */
 375     }
 376
 377     /* Increment number of outer iterations */
 378     outeriter        += nri;
 379
 380     /* Update outer/inner flops */
 381
 382     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*70);
 383 }
 384 /*
 385  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_sparc64_hpc_ace_double
 386  * Electrostatics interaction: ReactionField
 387  * VdW interaction:            CubicSplineTable
 388  * Geometry:                   Particle-Particle
 389  * Calculate force/pot:        Force
 390  */
 391 void
 392 nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_sparc64_hpc_ace_double
 393                     (t_nblist                    * gmx_restrict       nlist,
 394                      rvec                        * gmx_restrict          xx,
 395                      rvec                        * gmx_restrict          ff,
 396                      t_forcerec                  * gmx_restrict          fr,
 397                      t_mdatoms                   * gmx_restrict     mdatoms,
 398                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 399                      t_nrnb                      * gmx_restrict        nrnb)
 400 {
 401     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 402      * just 0 for non-waters.
 403      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 404      * jnr indices corresponding to data put in the four positions in the SIMD register.
 405      */
 406     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 407     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 408     int              jnrA,jnrB;
 409     int              j_coord_offsetA,j_coord_offsetB;
 410     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 411     real             rcutoff_scalar;
 412     real             *shiftvec,*fshift,*x,*f;
 413     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 414     int              vdwioffset0;
 415     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 416     int              vdwjidx0A,vdwjidx0B;
 417     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 418     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 419     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 420     real             *charge;
 421     int              nvdwtype;
 422     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 423     int              *vdwtype;
 424     real             *vdwparam;
 425     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 426     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 427     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 428     real             *vftab;
 429     _fjsp_v2r8       itab_tmp;
 430     _fjsp_v2r8       dummy_mask,cutoff_mask;
 431     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 432     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 433     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 434
 435     x                = xx[0];
 436     f                = ff[0];
 437
 438     nri              = nlist->nri;
 439     iinr             = nlist->iinr;
 440     jindex           = nlist->jindex;
 441     jjnr             = nlist->jjnr;
 442     shiftidx         = nlist->shift;
 443     gid              = nlist->gid;
 444     shiftvec         = fr->shift_vec[0];
 445     fshift           = fr->fshift[0];
 446     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 447     charge           = mdatoms->chargeA;
 448     krf              = gmx_fjsp_set1_v2r8(fr->ic->k_rf);
 449     krf2             = gmx_fjsp_set1_v2r8(fr->ic->k_rf*2.0);
 450     crf              = gmx_fjsp_set1_v2r8(fr->ic->c_rf);
 451     nvdwtype         = fr->ntype;
 452     vdwparam         = fr->nbfp;
 453     vdwtype          = mdatoms->typeA;
 454
 455     vftab            = kernel_data->table_vdw->data;
 456     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_vdw->scale);
 457
 458     /* Avoid stupid compiler warnings */
 459     jnrA = jnrB = 0;
 460     j_coord_offsetA = 0;
 461     j_coord_offsetB = 0;
 462
 463     outeriter        = 0;
 464     inneriter        = 0;
 465
 466     /* Start outer loop over neighborlists */
 467     for(iidx=0; iidx<nri; iidx++)
 468     {
 469         /* Load shift vector for this list */
 470         i_shift_offset   = DIM*shiftidx[iidx];
 471
 472         /* Load limits for loop over neighbors */
 473         j_index_start    = jindex[iidx];
 474         j_index_end      = jindex[iidx+1];
 475
 476         /* Get outer coordinate index */
 477         inr              = iinr[iidx];
 478         i_coord_offset   = DIM*inr;
 479
 480         /* Load i particle coords and add shift vector */
 481         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 482
 483         fix0             = _fjsp_setzero_v2r8();
 484         fiy0             = _fjsp_setzero_v2r8();
 485         fiz0             = _fjsp_setzero_v2r8();
 486
 487         /* Load parameters for i particles */
 488         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 489         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 490
 491         /* Start inner kernel loop */
 492         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 493         {
 494
 495             /* Get j neighbor index, and coordinate index */
 496             jnrA             = jjnr[jidx];
 497             jnrB             = jjnr[jidx+1];
 498             j_coord_offsetA  = DIM*jnrA;
 499             j_coord_offsetB  = DIM*jnrB;
 500
 501             /* load j atom coordinates */
 502             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 503                                               &jx0,&jy0,&jz0);
 504
 505             /* Calculate displacement vector */
 506             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 507             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 508             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 509
 510             /* Calculate squared distance and things based on it */
 511             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 512
 513             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 514
 515             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 516
 517             /* Load parameters for j particles */
 518             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 519             vdwjidx0A        = 2*vdwtype[jnrA+0];
 520             vdwjidx0B        = 2*vdwtype[jnrB+0];
 521
 522             /**************************
 523              * CALCULATE INTERACTIONS *
 524              **************************/
 525
 526             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 527
 528             /* Compute parameters for interactions between i and j atoms */
 529             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 530             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 531                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 532
 533             /* Calculate table index by multiplying r with table scale and truncate to integer */
 534             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 535             itab_tmp         = _fjsp_dtox_v2r8(rt);
 536             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 537             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 538             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 539
 540             vfconv.i[0]     *= 8;
 541             vfconv.i[1]     *= 8;
 542
 543             /* REACTION-FIELD ELECTROSTATICS */
 544             felec            = _fjsp_mul_v2r8(qq00,_fjsp_msub_v2r8(rinv00,rinvsq00,krf2));
 545
 546             /* CUBIC SPLINE TABLE DISPERSION */
 547             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 548             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 549             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 550             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 551             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 2 );
 552             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 553             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 554             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 555             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 556
 557             /* CUBIC SPLINE TABLE REPULSION */
 558             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 559             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 4 );
 560             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 561             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 562             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 6 );
 563             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 564             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 565             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 566             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 567             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 568
 569             fscal            = _fjsp_add_v2r8(felec,fvdw);
 570
 571             /* Update vectorial force */
 572             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 573             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 574             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 575
 576             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 577
 578             /* Inner loop uses 57 flops */
 579         }
 580
 581         if(jidx<j_index_end)
 582         {
 583
 584             jnrA             = jjnr[jidx];
 585             j_coord_offsetA  = DIM*jnrA;
 586
 587             /* load j atom coordinates */
 588             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 589                                               &jx0,&jy0,&jz0);
 590
 591             /* Calculate displacement vector */
 592             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 593             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 594             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 595
 596             /* Calculate squared distance and things based on it */
 597             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 598
 599             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 600
 601             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 602
 603             /* Load parameters for j particles */
 604             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 605             vdwjidx0A        = 2*vdwtype[jnrA+0];
 606
 607             /**************************
 608              * CALCULATE INTERACTIONS *
 609              **************************/
 610
 611             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 612
 613             /* Compute parameters for interactions between i and j atoms */
 614             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 615             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 616                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 617
 618             /* Calculate table index by multiplying r with table scale and truncate to integer */
 619             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 620             itab_tmp         = _fjsp_dtox_v2r8(rt);
 621             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 622             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 623             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 624
 625             vfconv.i[0]     *= 8;
 626             vfconv.i[1]     *= 8;
 627
 628             /* REACTION-FIELD ELECTROSTATICS */
 629             felec            = _fjsp_mul_v2r8(qq00,_fjsp_msub_v2r8(rinv00,rinvsq00,krf2));
 630
 631             /* CUBIC SPLINE TABLE DISPERSION */
 632             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 633             F                = _fjsp_setzero_v2r8();
 634             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 635             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 636             H                = _fjsp_setzero_v2r8();
 637             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 638             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 639             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 640             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 641
 642             /* CUBIC SPLINE TABLE REPULSION */
 643             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 644             F                = _fjsp_setzero_v2r8();
 645             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 646             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 647             H                = _fjsp_setzero_v2r8();
 648             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 649             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 650             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 651             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 652             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 653
 654             fscal            = _fjsp_add_v2r8(felec,fvdw);
 655
 656             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 657
 658             /* Update vectorial force */
 659             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 660             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 661             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 662
 663             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 664
 665             /* Inner loop uses 57 flops */
 666         }
 667
 668         /* End of innermost loop */
 669
 670         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 671                                               f+i_coord_offset,fshift+i_shift_offset);
 672
 673         /* Increment number of inner iterations */
 674         inneriter                  += j_index_end - j_index_start;
 675
 676         /* Outer loop uses 7 flops */
 677     }
 678
 679     /* Increment number of outer iterations */
 680     outeriter        += nri;
 681
 682     /* Update outer/inner flops */
 683
 684     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*57);
 685 }