src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecCoul_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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "kernelutil_sparc64_hpc_ace_double.h"
  50
  51 /*
  52  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_VF_sparc64_hpc_ace_double
  53  * Electrostatics interaction: Coulomb
  54  * VdW interaction:            CubicSplineTable
  55  * Geometry:                   Particle-Particle
  56  * Calculate force/pot:        PotentialAndForce
  57  */
  58 void
  59 nb_kernel_ElecCoul_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     nvdwtype         = fr->ntype;
 116     vdwparam         = fr->nbfp;
 117     vdwtype          = mdatoms->typeA;
 118
 119     vftab            = kernel_data->table_vdw->data;
 120     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_vdw->scale);
 121
 122     /* Avoid stupid compiler warnings */
 123     jnrA = jnrB = 0;
 124     j_coord_offsetA = 0;
 125     j_coord_offsetB = 0;
 126
 127     outeriter        = 0;
 128     inneriter        = 0;
 129
 130     /* Start outer loop over neighborlists */
 131     for(iidx=0; iidx<nri; iidx++)
 132     {
 133         /* Load shift vector for this list */
 134         i_shift_offset   = DIM*shiftidx[iidx];
 135
 136         /* Load limits for loop over neighbors */
 137         j_index_start    = jindex[iidx];
 138         j_index_end      = jindex[iidx+1];
 139
 140         /* Get outer coordinate index */
 141         inr              = iinr[iidx];
 142         i_coord_offset   = DIM*inr;
 143
 144         /* Load i particle coords and add shift vector */
 145         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 146
 147         fix0             = _fjsp_setzero_v2r8();
 148         fiy0             = _fjsp_setzero_v2r8();
 149         fiz0             = _fjsp_setzero_v2r8();
 150
 151         /* Load parameters for i particles */
 152         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 153         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 154
 155         /* Reset potential sums */
 156         velecsum         = _fjsp_setzero_v2r8();
 157         vvdwsum          = _fjsp_setzero_v2r8();
 158
 159         /* Start inner kernel loop */
 160         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 161         {
 162
 163             /* Get j neighbor index, and coordinate index */
 164             jnrA             = jjnr[jidx];
 165             jnrB             = jjnr[jidx+1];
 166             j_coord_offsetA  = DIM*jnrA;
 167             j_coord_offsetB  = DIM*jnrB;
 168
 169             /* load j atom coordinates */
 170             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 171                                               &jx0,&jy0,&jz0);
 172
 173             /* Calculate displacement vector */
 174             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 175             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 176             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 177
 178             /* Calculate squared distance and things based on it */
 179             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 180
 181             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 182
 183             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 184
 185             /* Load parameters for j particles */
 186             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 187             vdwjidx0A        = 2*vdwtype[jnrA+0];
 188             vdwjidx0B        = 2*vdwtype[jnrB+0];
 189
 190             /**************************
 191              * CALCULATE INTERACTIONS *
 192              **************************/
 193
 194             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 195
 196             /* Compute parameters for interactions between i and j atoms */
 197             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 198             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 199                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 200
 201             /* Calculate table index by multiplying r with table scale and truncate to integer */
 202             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 203             itab_tmp         = _fjsp_dtox_v2r8(rt);
 204             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 205             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 206             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 207
 208             vfconv.i[0]     *= 8;
 209             vfconv.i[1]     *= 8;
 210
 211             /* COULOMB ELECTROSTATICS */
 212             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 213             felec            = _fjsp_mul_v2r8(velec,rinvsq00);
 214
 215             /* CUBIC SPLINE TABLE DISPERSION */
 216             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 217             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 218             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 219             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 220             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 2 );
 221             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 222             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 223             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 224             vvdw6            = _fjsp_mul_v2r8(c6_00,VV);
 225             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 226             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 227
 228             /* CUBIC SPLINE TABLE REPULSION */
 229             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 230             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 4 );
 231             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 232             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 233             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 6 );
 234             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 235             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 236             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 237             vvdw12           = _fjsp_mul_v2r8(c12_00,VV);
 238             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 239             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 240             vvdw             = _fjsp_add_v2r8(vvdw12,vvdw6);
 241             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 242
 243             /* Update potential sum for this i atom from the interaction with this j atom. */
 244             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 245             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 246
 247             fscal            = _fjsp_add_v2r8(felec,fvdw);
 248
 249             /* Update vectorial force */
 250             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 251             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 252             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 253
 254             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 255
 256             /* Inner loop uses 66 flops */
 257         }
 258
 259         if(jidx<j_index_end)
 260         {
 261
 262             jnrA             = jjnr[jidx];
 263             j_coord_offsetA  = DIM*jnrA;
 264
 265             /* load j atom coordinates */
 266             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 267                                               &jx0,&jy0,&jz0);
 268
 269             /* Calculate displacement vector */
 270             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 271             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 272             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 273
 274             /* Calculate squared distance and things based on it */
 275             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 276
 277             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 278
 279             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 280
 281             /* Load parameters for j particles */
 282             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 283             vdwjidx0A        = 2*vdwtype[jnrA+0];
 284
 285             /**************************
 286              * CALCULATE INTERACTIONS *
 287              **************************/
 288
 289             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 290
 291             /* Compute parameters for interactions between i and j atoms */
 292             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 293             gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 294
 295             /* Calculate table index by multiplying r with table scale and truncate to integer */
 296             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 297             itab_tmp         = _fjsp_dtox_v2r8(rt);
 298             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 299             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 300             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 301
 302             vfconv.i[0]     *= 8;
 303             vfconv.i[1]     *= 8;
 304
 305             /* COULOMB ELECTROSTATICS */
 306             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 307             felec            = _fjsp_mul_v2r8(velec,rinvsq00);
 308
 309             /* CUBIC SPLINE TABLE DISPERSION */
 310             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 311             F                = _fjsp_setzero_v2r8();
 312             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 313             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 314             H                = _fjsp_setzero_v2r8();
 315             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 316             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 317             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 318             vvdw6            = _fjsp_mul_v2r8(c6_00,VV);
 319             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 320             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 321
 322             /* CUBIC SPLINE TABLE REPULSION */
 323             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 324             F                = _fjsp_setzero_v2r8();
 325             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 326             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 327             H                = _fjsp_setzero_v2r8();
 328             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 329             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 330             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 331             vvdw12           = _fjsp_mul_v2r8(c12_00,VV);
 332             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 333             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 334             vvdw             = _fjsp_add_v2r8(vvdw12,vvdw6);
 335             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 336
 337             /* Update potential sum for this i atom from the interaction with this j atom. */
 338             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 339             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 340             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 341             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 342
 343             fscal            = _fjsp_add_v2r8(felec,fvdw);
 344
 345             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 346
 347             /* Update vectorial force */
 348             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 349             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 350             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 351
 352             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 353
 354             /* Inner loop uses 66 flops */
 355         }
 356
 357         /* End of innermost loop */
 358
 359         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 360                                               f+i_coord_offset,fshift+i_shift_offset);
 361
 362         ggid                        = gid[iidx];
 363         /* Update potential energies */
 364         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 365         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 366
 367         /* Increment number of inner iterations */
 368         inneriter                  += j_index_end - j_index_start;
 369
 370         /* Outer loop uses 9 flops */
 371     }
 372
 373     /* Increment number of outer iterations */
 374     outeriter        += nri;
 375
 376     /* Update outer/inner flops */
 377
 378     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*66);
 379 }
 380 /*
 381  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_sparc64_hpc_ace_double
 382  * Electrostatics interaction: Coulomb
 383  * VdW interaction:            CubicSplineTable
 384  * Geometry:                   Particle-Particle
 385  * Calculate force/pot:        Force
 386  */
 387 void
 388 nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_sparc64_hpc_ace_double
 389                     (t_nblist                    * gmx_restrict       nlist,
 390                      rvec                        * gmx_restrict          xx,
 391                      rvec                        * gmx_restrict          ff,
 392                      t_forcerec                  * gmx_restrict          fr,
 393                      t_mdatoms                   * gmx_restrict     mdatoms,
 394                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 395                      t_nrnb                      * gmx_restrict        nrnb)
 396 {
 397     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 398      * just 0 for non-waters.
 399      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 400      * jnr indices corresponding to data put in the four positions in the SIMD register.
 401      */
 402     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 403     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 404     int              jnrA,jnrB;
 405     int              j_coord_offsetA,j_coord_offsetB;
 406     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 407     real             rcutoff_scalar;
 408     real             *shiftvec,*fshift,*x,*f;
 409     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 410     int              vdwioffset0;
 411     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 412     int              vdwjidx0A,vdwjidx0B;
 413     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 414     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 415     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 416     real             *charge;
 417     int              nvdwtype;
 418     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 419     int              *vdwtype;
 420     real             *vdwparam;
 421     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 422     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 423     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 424     real             *vftab;
 425     _fjsp_v2r8       itab_tmp;
 426     _fjsp_v2r8       dummy_mask,cutoff_mask;
 427     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 428     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 429     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 430
 431     x                = xx[0];
 432     f                = ff[0];
 433
 434     nri              = nlist->nri;
 435     iinr             = nlist->iinr;
 436     jindex           = nlist->jindex;
 437     jjnr             = nlist->jjnr;
 438     shiftidx         = nlist->shift;
 439     gid              = nlist->gid;
 440     shiftvec         = fr->shift_vec[0];
 441     fshift           = fr->fshift[0];
 442     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 443     charge           = mdatoms->chargeA;
 444     nvdwtype         = fr->ntype;
 445     vdwparam         = fr->nbfp;
 446     vdwtype          = mdatoms->typeA;
 447
 448     vftab            = kernel_data->table_vdw->data;
 449     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_vdw->scale);
 450
 451     /* Avoid stupid compiler warnings */
 452     jnrA = jnrB = 0;
 453     j_coord_offsetA = 0;
 454     j_coord_offsetB = 0;
 455
 456     outeriter        = 0;
 457     inneriter        = 0;
 458
 459     /* Start outer loop over neighborlists */
 460     for(iidx=0; iidx<nri; iidx++)
 461     {
 462         /* Load shift vector for this list */
 463         i_shift_offset   = DIM*shiftidx[iidx];
 464
 465         /* Load limits for loop over neighbors */
 466         j_index_start    = jindex[iidx];
 467         j_index_end      = jindex[iidx+1];
 468
 469         /* Get outer coordinate index */
 470         inr              = iinr[iidx];
 471         i_coord_offset   = DIM*inr;
 472
 473         /* Load i particle coords and add shift vector */
 474         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 475
 476         fix0             = _fjsp_setzero_v2r8();
 477         fiy0             = _fjsp_setzero_v2r8();
 478         fiz0             = _fjsp_setzero_v2r8();
 479
 480         /* Load parameters for i particles */
 481         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 482         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 483
 484         /* Start inner kernel loop */
 485         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 486         {
 487
 488             /* Get j neighbor index, and coordinate index */
 489             jnrA             = jjnr[jidx];
 490             jnrB             = jjnr[jidx+1];
 491             j_coord_offsetA  = DIM*jnrA;
 492             j_coord_offsetB  = DIM*jnrB;
 493
 494             /* load j atom coordinates */
 495             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 496                                               &jx0,&jy0,&jz0);
 497
 498             /* Calculate displacement vector */
 499             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 500             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 501             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 502
 503             /* Calculate squared distance and things based on it */
 504             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 505
 506             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 507
 508             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 509
 510             /* Load parameters for j particles */
 511             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 512             vdwjidx0A        = 2*vdwtype[jnrA+0];
 513             vdwjidx0B        = 2*vdwtype[jnrB+0];
 514
 515             /**************************
 516              * CALCULATE INTERACTIONS *
 517              **************************/
 518
 519             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 520
 521             /* Compute parameters for interactions between i and j atoms */
 522             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 523             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 524                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 525
 526             /* Calculate table index by multiplying r with table scale and truncate to integer */
 527             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 528             itab_tmp         = _fjsp_dtox_v2r8(rt);
 529             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 530             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 531             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 532
 533             vfconv.i[0]     *= 8;
 534             vfconv.i[1]     *= 8;
 535
 536             /* COULOMB ELECTROSTATICS */
 537             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 538             felec            = _fjsp_mul_v2r8(velec,rinvsq00);
 539
 540             /* CUBIC SPLINE TABLE DISPERSION */
 541             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 542             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 543             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 544             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 545             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 2 );
 546             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 547             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 548             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 549             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 550
 551             /* CUBIC SPLINE TABLE REPULSION */
 552             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 553             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 4 );
 554             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 555             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 556             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 6 );
 557             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 558             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 559             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 560             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 561             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 562
 563             fscal            = _fjsp_add_v2r8(felec,fvdw);
 564
 565             /* Update vectorial force */
 566             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 567             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 568             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 569
 570             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 571
 572             /* Inner loop uses 57 flops */
 573         }
 574
 575         if(jidx<j_index_end)
 576         {
 577
 578             jnrA             = jjnr[jidx];
 579             j_coord_offsetA  = DIM*jnrA;
 580
 581             /* load j atom coordinates */
 582             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 583                                               &jx0,&jy0,&jz0);
 584
 585             /* Calculate displacement vector */
 586             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 587             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 588             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 589
 590             /* Calculate squared distance and things based on it */
 591             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 592
 593             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 594
 595             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 596
 597             /* Load parameters for j particles */
 598             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 599             vdwjidx0A        = 2*vdwtype[jnrA+0];
 600
 601             /**************************
 602              * CALCULATE INTERACTIONS *
 603              **************************/
 604
 605             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 606
 607             /* Compute parameters for interactions between i and j atoms */
 608             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 609             gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 610
 611             /* Calculate table index by multiplying r with table scale and truncate to integer */
 612             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 613             itab_tmp         = _fjsp_dtox_v2r8(rt);
 614             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 615             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 616             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 617
 618             vfconv.i[0]     *= 8;
 619             vfconv.i[1]     *= 8;
 620
 621             /* COULOMB ELECTROSTATICS */
 622             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 623             felec            = _fjsp_mul_v2r8(velec,rinvsq00);
 624
 625             /* CUBIC SPLINE TABLE DISPERSION */
 626             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 627             F                = _fjsp_setzero_v2r8();
 628             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 629             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 630             H                = _fjsp_setzero_v2r8();
 631             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 632             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 633             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 634             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 635
 636             /* CUBIC SPLINE TABLE REPULSION */
 637             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 638             F                = _fjsp_setzero_v2r8();
 639             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 640             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 641             H                = _fjsp_setzero_v2r8();
 642             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 643             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 644             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 645             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 646             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 647
 648             fscal            = _fjsp_add_v2r8(felec,fvdw);
 649
 650             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 651
 652             /* Update vectorial force */
 653             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 654             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 655             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 656
 657             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 658
 659             /* Inner loop uses 57 flops */
 660         }
 661
 662         /* End of innermost loop */
 663
 664         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 665                                               f+i_coord_offset,fshift+i_shift_offset);
 666
 667         /* Increment number of inner iterations */
 668         inneriter                  += j_index_end - j_index_start;
 669
 670         /* Outer loop uses 7 flops */
 671     }
 672
 673     /* Increment number of outer iterations */
 674     outeriter        += nri;
 675
 676     /* Update outer/inner flops */
 677
 678     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*57);
 679 }