src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_sparc64_hpc_ace_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sparc64_hpc_ace_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "kernelutil_sparc64_hpc_ace_double.h"
  50
  51 /*
  52  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_VF_sparc64_hpc_ace_double
  53  * Electrostatics interaction: CubicSplineTable
  54  * VdW interaction:            CubicSplineTable
  55  * Geometry:                   Particle-Particle
  56  * Calculate force/pot:        PotentialAndForce
  57  */
  58 void
  59 nb_kernel_ElecCSTab_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_elec_vdw->data;
 120     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_elec_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             /* Load parameters for j particles */
 184             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 185             vdwjidx0A        = 2*vdwtype[jnrA+0];
 186             vdwjidx0B        = 2*vdwtype[jnrB+0];
 187
 188             /**************************
 189              * CALCULATE INTERACTIONS *
 190              **************************/
 191
 192             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 193
 194             /* Compute parameters for interactions between i and j atoms */
 195             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 196             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 197                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 198
 199             /* Calculate table index by multiplying r with table scale and truncate to integer */
 200             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 201             itab_tmp         = _fjsp_dtox_v2r8(rt);
 202             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 203             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 204             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 205
 206             vfconv.i[0]     *= 12;
 207             vfconv.i[1]     *= 12;
 208
 209             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 210             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 211             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 212             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 213             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] +2);
 214             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] +2);
 215             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 216             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(vfeps,H,G),F);
 217             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 218             velec            = _fjsp_mul_v2r8(qq00,VV);
 219             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twovfeps,H,G),vfeps,Fp);
 220             felec            = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,FF),_fjsp_mul_v2r8(vftabscale,rinv00)));
 221
 222             /* CUBIC SPLINE TABLE DISPERSION */
 223             vfconv.i[0]       += 4;
 224             vfconv.i[1]       += 4;
 225             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 226             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 227             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 228             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 229             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 2 );
 230             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 231             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 232             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 233             vvdw6            = _fjsp_mul_v2r8(c6_00,VV);
 234             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 235             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 236
 237             /* CUBIC SPLINE TABLE REPULSION */
 238             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 239             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 4 );
 240             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 241             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 242             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 6 );
 243             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 244             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 245             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 246             vvdw12           = _fjsp_mul_v2r8(c12_00,VV);
 247             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 248             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 249             vvdw             = _fjsp_add_v2r8(vvdw12,vvdw6);
 250             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 251
 252             /* Update potential sum for this i atom from the interaction with this j atom. */
 253             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 254             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 255
 256             fscal            = _fjsp_add_v2r8(felec,fvdw);
 257
 258             /* Update vectorial force */
 259             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 260             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 261             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 262
 263             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 264
 265             /* Inner loop uses 76 flops */
 266         }
 267
 268         if(jidx<j_index_end)
 269         {
 270
 271             jnrA             = jjnr[jidx];
 272             j_coord_offsetA  = DIM*jnrA;
 273
 274             /* load j atom coordinates */
 275             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 276                                               &jx0,&jy0,&jz0);
 277
 278             /* Calculate displacement vector */
 279             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 280             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 281             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 282
 283             /* Calculate squared distance and things based on it */
 284             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 285
 286             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 287
 288             /* Load parameters for j particles */
 289             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 290             vdwjidx0A        = 2*vdwtype[jnrA+0];
 291
 292             /**************************
 293              * CALCULATE INTERACTIONS *
 294              **************************/
 295
 296             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 297
 298             /* Compute parameters for interactions between i and j atoms */
 299             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 300             gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 301
 302             /* Calculate table index by multiplying r with table scale and truncate to integer */
 303             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 304             itab_tmp         = _fjsp_dtox_v2r8(rt);
 305             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 306             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 307             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 308
 309             vfconv.i[0]     *= 12;
 310             vfconv.i[1]     *= 12;
 311
 312             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 313             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 314             F                = _fjsp_setzero_v2r8();
 315             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 316             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] +2);
 317             H                = _fjsp_setzero_v2r8();
 318             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 319             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(vfeps,H,G),F);
 320             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 321             velec            = _fjsp_mul_v2r8(qq00,VV);
 322             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twovfeps,H,G),vfeps,Fp);
 323             felec            = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,FF),_fjsp_mul_v2r8(vftabscale,rinv00)));
 324
 325             /* CUBIC SPLINE TABLE DISPERSION */
 326             vfconv.i[0]       += 4;
 327             vfconv.i[1]       += 4;
 328             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 329             F                = _fjsp_setzero_v2r8();
 330             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 331             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 332             H                = _fjsp_setzero_v2r8();
 333             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 334             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 335             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 336             vvdw6            = _fjsp_mul_v2r8(c6_00,VV);
 337             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 338             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 339
 340             /* CUBIC SPLINE TABLE REPULSION */
 341             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 342             F                = _fjsp_setzero_v2r8();
 343             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 344             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 345             H                = _fjsp_setzero_v2r8();
 346             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 347             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 348             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 349             vvdw12           = _fjsp_mul_v2r8(c12_00,VV);
 350             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 351             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 352             vvdw             = _fjsp_add_v2r8(vvdw12,vvdw6);
 353             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 354
 355             /* Update potential sum for this i atom from the interaction with this j atom. */
 356             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 357             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 358             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 359             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 360
 361             fscal            = _fjsp_add_v2r8(felec,fvdw);
 362
 363             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 364
 365             /* Update vectorial force */
 366             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 367             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 368             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 369
 370             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 371
 372             /* Inner loop uses 76 flops */
 373         }
 374
 375         /* End of innermost loop */
 376
 377         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 378                                               f+i_coord_offset,fshift+i_shift_offset);
 379
 380         ggid                        = gid[iidx];
 381         /* Update potential energies */
 382         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 383         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 384
 385         /* Increment number of inner iterations */
 386         inneriter                  += j_index_end - j_index_start;
 387
 388         /* Outer loop uses 9 flops */
 389     }
 390
 391     /* Increment number of outer iterations */
 392     outeriter        += nri;
 393
 394     /* Update outer/inner flops */
 395
 396     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*76);
 397 }
 398 /*
 399  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_F_sparc64_hpc_ace_double
 400  * Electrostatics interaction: CubicSplineTable
 401  * VdW interaction:            CubicSplineTable
 402  * Geometry:                   Particle-Particle
 403  * Calculate force/pot:        Force
 404  */
 405 void
 406 nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_F_sparc64_hpc_ace_double
 407                     (t_nblist                    * gmx_restrict       nlist,
 408                      rvec                        * gmx_restrict          xx,
 409                      rvec                        * gmx_restrict          ff,
 410                      t_forcerec                  * gmx_restrict          fr,
 411                      t_mdatoms                   * gmx_restrict     mdatoms,
 412                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 413                      t_nrnb                      * gmx_restrict        nrnb)
 414 {
 415     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 416      * just 0 for non-waters.
 417      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 418      * jnr indices corresponding to data put in the four positions in the SIMD register.
 419      */
 420     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 421     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 422     int              jnrA,jnrB;
 423     int              j_coord_offsetA,j_coord_offsetB;
 424     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 425     real             rcutoff_scalar;
 426     real             *shiftvec,*fshift,*x,*f;
 427     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 428     int              vdwioffset0;
 429     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 430     int              vdwjidx0A,vdwjidx0B;
 431     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 432     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 433     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 434     real             *charge;
 435     int              nvdwtype;
 436     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 437     int              *vdwtype;
 438     real             *vdwparam;
 439     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 440     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 441     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 442     real             *vftab;
 443     _fjsp_v2r8       itab_tmp;
 444     _fjsp_v2r8       dummy_mask,cutoff_mask;
 445     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 446     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 447     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 448
 449     x                = xx[0];
 450     f                = ff[0];
 451
 452     nri              = nlist->nri;
 453     iinr             = nlist->iinr;
 454     jindex           = nlist->jindex;
 455     jjnr             = nlist->jjnr;
 456     shiftidx         = nlist->shift;
 457     gid              = nlist->gid;
 458     shiftvec         = fr->shift_vec[0];
 459     fshift           = fr->fshift[0];
 460     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 461     charge           = mdatoms->chargeA;
 462     nvdwtype         = fr->ntype;
 463     vdwparam         = fr->nbfp;
 464     vdwtype          = mdatoms->typeA;
 465
 466     vftab            = kernel_data->table_elec_vdw->data;
 467     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_elec_vdw->scale);
 468
 469     /* Avoid stupid compiler warnings */
 470     jnrA = jnrB = 0;
 471     j_coord_offsetA = 0;
 472     j_coord_offsetB = 0;
 473
 474     outeriter        = 0;
 475     inneriter        = 0;
 476
 477     /* Start outer loop over neighborlists */
 478     for(iidx=0; iidx<nri; iidx++)
 479     {
 480         /* Load shift vector for this list */
 481         i_shift_offset   = DIM*shiftidx[iidx];
 482
 483         /* Load limits for loop over neighbors */
 484         j_index_start    = jindex[iidx];
 485         j_index_end      = jindex[iidx+1];
 486
 487         /* Get outer coordinate index */
 488         inr              = iinr[iidx];
 489         i_coord_offset   = DIM*inr;
 490
 491         /* Load i particle coords and add shift vector */
 492         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 493
 494         fix0             = _fjsp_setzero_v2r8();
 495         fiy0             = _fjsp_setzero_v2r8();
 496         fiz0             = _fjsp_setzero_v2r8();
 497
 498         /* Load parameters for i particles */
 499         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 500         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 501
 502         /* Start inner kernel loop */
 503         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 504         {
 505
 506             /* Get j neighbor index, and coordinate index */
 507             jnrA             = jjnr[jidx];
 508             jnrB             = jjnr[jidx+1];
 509             j_coord_offsetA  = DIM*jnrA;
 510             j_coord_offsetB  = DIM*jnrB;
 511
 512             /* load j atom coordinates */
 513             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 514                                               &jx0,&jy0,&jz0);
 515
 516             /* Calculate displacement vector */
 517             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 518             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 519             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 520
 521             /* Calculate squared distance and things based on it */
 522             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 523
 524             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 525
 526             /* Load parameters for j particles */
 527             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 528             vdwjidx0A        = 2*vdwtype[jnrA+0];
 529             vdwjidx0B        = 2*vdwtype[jnrB+0];
 530
 531             /**************************
 532              * CALCULATE INTERACTIONS *
 533              **************************/
 534
 535             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 536
 537             /* Compute parameters for interactions between i and j atoms */
 538             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 539             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 540                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 541
 542             /* Calculate table index by multiplying r with table scale and truncate to integer */
 543             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 544             itab_tmp         = _fjsp_dtox_v2r8(rt);
 545             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 546             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 547             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 548
 549             vfconv.i[0]     *= 12;
 550             vfconv.i[1]     *= 12;
 551
 552             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 553             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 554             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 555             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 556             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] +2);
 557             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] +2);
 558             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 559             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(vfeps,H,G),F);
 560             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twovfeps,H,G),vfeps,Fp);
 561             felec            = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,FF),_fjsp_mul_v2r8(vftabscale,rinv00)));
 562
 563             /* CUBIC SPLINE TABLE DISPERSION */
 564             vfconv.i[0]       += 4;
 565             vfconv.i[1]       += 4;
 566             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 567             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 568             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 569             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 570             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 2 );
 571             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 572             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 573             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 574             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 575
 576             /* CUBIC SPLINE TABLE REPULSION */
 577             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 578             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 4 );
 579             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 580             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 581             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 6 );
 582             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 583             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 584             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 585             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 586             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 587
 588             fscal            = _fjsp_add_v2r8(felec,fvdw);
 589
 590             /* Update vectorial force */
 591             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 592             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 593             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 594
 595             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 596
 597             /* Inner loop uses 64 flops */
 598         }
 599
 600         if(jidx<j_index_end)
 601         {
 602
 603             jnrA             = jjnr[jidx];
 604             j_coord_offsetA  = DIM*jnrA;
 605
 606             /* load j atom coordinates */
 607             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 608                                               &jx0,&jy0,&jz0);
 609
 610             /* Calculate displacement vector */
 611             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 612             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 613             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 614
 615             /* Calculate squared distance and things based on it */
 616             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 617
 618             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 619
 620             /* Load parameters for j particles */
 621             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 622             vdwjidx0A        = 2*vdwtype[jnrA+0];
 623
 624             /**************************
 625              * CALCULATE INTERACTIONS *
 626              **************************/
 627
 628             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 629
 630             /* Compute parameters for interactions between i and j atoms */
 631             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 632             gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 633
 634             /* Calculate table index by multiplying r with table scale and truncate to integer */
 635             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 636             itab_tmp         = _fjsp_dtox_v2r8(rt);
 637             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 638             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 639             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 640
 641             vfconv.i[0]     *= 12;
 642             vfconv.i[1]     *= 12;
 643
 644             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 645             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 646             F                = _fjsp_setzero_v2r8();
 647             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 648             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] +2);
 649             H                = _fjsp_setzero_v2r8();
 650             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 651             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(vfeps,H,G),F);
 652             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twovfeps,H,G),vfeps,Fp);
 653             felec            = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,FF),_fjsp_mul_v2r8(vftabscale,rinv00)));
 654
 655             /* CUBIC SPLINE TABLE DISPERSION */
 656             vfconv.i[0]       += 4;
 657             vfconv.i[1]       += 4;
 658             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 659             F                = _fjsp_setzero_v2r8();
 660             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 661             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 662             H                = _fjsp_setzero_v2r8();
 663             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 664             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 665             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 666             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 667
 668             /* CUBIC SPLINE TABLE REPULSION */
 669             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 670             F                = _fjsp_setzero_v2r8();
 671             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 672             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 673             H                = _fjsp_setzero_v2r8();
 674             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 675             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 676             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 677             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 678             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 679
 680             fscal            = _fjsp_add_v2r8(felec,fvdw);
 681
 682             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 683
 684             /* Update vectorial force */
 685             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 686             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 687             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 688
 689             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 690
 691             /* Inner loop uses 64 flops */
 692         }
 693
 694         /* End of innermost loop */
 695
 696         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 697                                               f+i_coord_offset,fshift+i_shift_offset);
 698
 699         /* Increment number of inner iterations */
 700         inneriter                  += j_index_end - j_index_start;
 701
 702         /* Outer loop uses 7 flops */
 703     }
 704
 705     /* Increment number of outer iterations */
 706     outeriter        += nri;
 707
 708     /* Update outer/inner flops */
 709
 710     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*64);
 711 }