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) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sparc64_hpc_ace_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_sparc64_hpc_ace_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_VF_sparc64_hpc_ace_double
  51  * Electrostatics interaction: CubicSplineTable
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_VF_sparc64_hpc_ace_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwjidx0A,vdwjidx0B;
  82     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  83     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  84     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
  85     real             *charge;
  86     int              nvdwtype;
  87     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  88     int              *vdwtype;
  89     real             *vdwparam;
  90     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
  91     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
  92     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  93     real             *vftab;
  94     _fjsp_v2r8       itab_tmp;
  95     _fjsp_v2r8       dummy_mask,cutoff_mask;
  96     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
  97     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
  98     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
  99
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     facel            = gmx_fjsp_set1_v2r8(fr->ic->epsfac);
 112     charge           = mdatoms->chargeA;
 113     nvdwtype         = fr->ntype;
 114     vdwparam         = fr->nbfp;
 115     vdwtype          = mdatoms->typeA;
 116
 117     vftab            = kernel_data->table_elec_vdw->data;
 118     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_elec_vdw->scale);
 119
 120     /* Avoid stupid compiler warnings */
 121     jnrA = jnrB = 0;
 122     j_coord_offsetA = 0;
 123     j_coord_offsetB = 0;
 124
 125     outeriter        = 0;
 126     inneriter        = 0;
 127
 128     /* Start outer loop over neighborlists */
 129     for(iidx=0; iidx<nri; iidx++)
 130     {
 131         /* Load shift vector for this list */
 132         i_shift_offset   = DIM*shiftidx[iidx];
 133
 134         /* Load limits for loop over neighbors */
 135         j_index_start    = jindex[iidx];
 136         j_index_end      = jindex[iidx+1];
 137
 138         /* Get outer coordinate index */
 139         inr              = iinr[iidx];
 140         i_coord_offset   = DIM*inr;
 141
 142         /* Load i particle coords and add shift vector */
 143         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 144
 145         fix0             = _fjsp_setzero_v2r8();
 146         fiy0             = _fjsp_setzero_v2r8();
 147         fiz0             = _fjsp_setzero_v2r8();
 148
 149         /* Load parameters for i particles */
 150         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 151         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 152
 153         /* Reset potential sums */
 154         velecsum         = _fjsp_setzero_v2r8();
 155         vvdwsum          = _fjsp_setzero_v2r8();
 156
 157         /* Start inner kernel loop */
 158         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 159         {
 160
 161             /* Get j neighbor index, and coordinate index */
 162             jnrA             = jjnr[jidx];
 163             jnrB             = jjnr[jidx+1];
 164             j_coord_offsetA  = DIM*jnrA;
 165             j_coord_offsetB  = DIM*jnrB;
 166
 167             /* load j atom coordinates */
 168             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 169                                               &jx0,&jy0,&jz0);
 170
 171             /* Calculate displacement vector */
 172             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 173             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 174             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 175
 176             /* Calculate squared distance and things based on it */
 177             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 178
 179             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 180
 181             /* Load parameters for j particles */
 182             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 183             vdwjidx0A        = 2*vdwtype[jnrA+0];
 184             vdwjidx0B        = 2*vdwtype[jnrB+0];
 185
 186             /**************************
 187              * CALCULATE INTERACTIONS *
 188              **************************/
 189
 190             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 191
 192             /* Compute parameters for interactions between i and j atoms */
 193             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 194             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 195                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 196
 197             /* Calculate table index by multiplying r with table scale and truncate to integer */
 198             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 199             itab_tmp         = _fjsp_dtox_v2r8(rt);
 200             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 201             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 202             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 203
 204             vfconv.i[0]     *= 12;
 205             vfconv.i[1]     *= 12;
 206
 207             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 208             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 209             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 210             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 211             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] +2);
 212             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] +2);
 213             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 214             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(vfeps,H,G),F);
 215             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 216             velec            = _fjsp_mul_v2r8(qq00,VV);
 217             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twovfeps,H,G),vfeps,Fp);
 218             felec            = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,FF),_fjsp_mul_v2r8(vftabscale,rinv00)));
 219
 220             /* CUBIC SPLINE TABLE DISPERSION */
 221             vfconv.i[0]       += 4;
 222             vfconv.i[1]       += 4;
 223             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 224             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 225             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 226             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 227             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 2 );
 228             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 229             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 230             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 231             vvdw6            = _fjsp_mul_v2r8(c6_00,VV);
 232             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 233             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 234
 235             /* CUBIC SPLINE TABLE REPULSION */
 236             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 237             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 4 );
 238             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 239             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 240             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 6 );
 241             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 242             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 243             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 244             vvdw12           = _fjsp_mul_v2r8(c12_00,VV);
 245             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 246             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 247             vvdw             = _fjsp_add_v2r8(vvdw12,vvdw6);
 248             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 249
 250             /* Update potential sum for this i atom from the interaction with this j atom. */
 251             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 252             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 253
 254             fscal            = _fjsp_add_v2r8(felec,fvdw);
 255
 256             /* Update vectorial force */
 257             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 258             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 259             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 260
 261             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 262
 263             /* Inner loop uses 76 flops */
 264         }
 265
 266         if(jidx<j_index_end)
 267         {
 268
 269             jnrA             = jjnr[jidx];
 270             j_coord_offsetA  = DIM*jnrA;
 271
 272             /* load j atom coordinates */
 273             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 274                                               &jx0,&jy0,&jz0);
 275
 276             /* Calculate displacement vector */
 277             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 278             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 279             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 280
 281             /* Calculate squared distance and things based on it */
 282             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 283
 284             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 285
 286             /* Load parameters for j particles */
 287             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 288             vdwjidx0A        = 2*vdwtype[jnrA+0];
 289
 290             /**************************
 291              * CALCULATE INTERACTIONS *
 292              **************************/
 293
 294             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 295
 296             /* Compute parameters for interactions between i and j atoms */
 297             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 298             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 299                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 300
 301             /* Calculate table index by multiplying r with table scale and truncate to integer */
 302             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 303             itab_tmp         = _fjsp_dtox_v2r8(rt);
 304             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 305             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 306             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 307
 308             vfconv.i[0]     *= 12;
 309             vfconv.i[1]     *= 12;
 310
 311             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 312             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 313             F                = _fjsp_setzero_v2r8();
 314             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 315             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] +2);
 316             H                = _fjsp_setzero_v2r8();
 317             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 318             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(vfeps,H,G),F);
 319             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 320             velec            = _fjsp_mul_v2r8(qq00,VV);
 321             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twovfeps,H,G),vfeps,Fp);
 322             felec            = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,FF),_fjsp_mul_v2r8(vftabscale,rinv00)));
 323
 324             /* CUBIC SPLINE TABLE DISPERSION */
 325             vfconv.i[0]       += 4;
 326             vfconv.i[1]       += 4;
 327             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 328             F                = _fjsp_setzero_v2r8();
 329             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 330             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 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             vvdw6            = _fjsp_mul_v2r8(c6_00,VV);
 336             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 337             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 338
 339             /* CUBIC SPLINE TABLE REPULSION */
 340             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 341             F                = _fjsp_setzero_v2r8();
 342             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 343             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 344             H                = _fjsp_setzero_v2r8();
 345             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 346             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 347             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 348             vvdw12           = _fjsp_mul_v2r8(c12_00,VV);
 349             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 350             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 351             vvdw             = _fjsp_add_v2r8(vvdw12,vvdw6);
 352             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 353
 354             /* Update potential sum for this i atom from the interaction with this j atom. */
 355             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 356             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 357             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 358             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 359
 360             fscal            = _fjsp_add_v2r8(felec,fvdw);
 361
 362             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 363
 364             /* Update vectorial force */
 365             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 366             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 367             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 368
 369             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 370
 371             /* Inner loop uses 76 flops */
 372         }
 373
 374         /* End of innermost loop */
 375
 376         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 377                                               f+i_coord_offset,fshift+i_shift_offset);
 378
 379         ggid                        = gid[iidx];
 380         /* Update potential energies */
 381         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 382         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 383
 384         /* Increment number of inner iterations */
 385         inneriter                  += j_index_end - j_index_start;
 386
 387         /* Outer loop uses 9 flops */
 388     }
 389
 390     /* Increment number of outer iterations */
 391     outeriter        += nri;
 392
 393     /* Update outer/inner flops */
 394
 395     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*76);
 396 }
 397 /*
 398  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_F_sparc64_hpc_ace_double
 399  * Electrostatics interaction: CubicSplineTable
 400  * VdW interaction:            CubicSplineTable
 401  * Geometry:                   Particle-Particle
 402  * Calculate force/pot:        Force
 403  */
 404 void
 405 nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_F_sparc64_hpc_ace_double
 406                     (t_nblist                    * gmx_restrict       nlist,
 407                      rvec                        * gmx_restrict          xx,
 408                      rvec                        * gmx_restrict          ff,
 409                      struct t_forcerec           * gmx_restrict          fr,
 410                      t_mdatoms                   * gmx_restrict     mdatoms,
 411                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 412                      t_nrnb                      * gmx_restrict        nrnb)
 413 {
 414     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 415      * just 0 for non-waters.
 416      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 417      * jnr indices corresponding to data put in the four positions in the SIMD register.
 418      */
 419     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 420     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 421     int              jnrA,jnrB;
 422     int              j_coord_offsetA,j_coord_offsetB;
 423     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 424     real             rcutoff_scalar;
 425     real             *shiftvec,*fshift,*x,*f;
 426     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 427     int              vdwioffset0;
 428     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 429     int              vdwjidx0A,vdwjidx0B;
 430     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 431     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 432     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 433     real             *charge;
 434     int              nvdwtype;
 435     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 436     int              *vdwtype;
 437     real             *vdwparam;
 438     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 439     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 440     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 441     real             *vftab;
 442     _fjsp_v2r8       itab_tmp;
 443     _fjsp_v2r8       dummy_mask,cutoff_mask;
 444     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 445     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 446     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 447
 448     x                = xx[0];
 449     f                = ff[0];
 450
 451     nri              = nlist->nri;
 452     iinr             = nlist->iinr;
 453     jindex           = nlist->jindex;
 454     jjnr             = nlist->jjnr;
 455     shiftidx         = nlist->shift;
 456     gid              = nlist->gid;
 457     shiftvec         = fr->shift_vec[0];
 458     fshift           = fr->fshift[0];
 459     facel            = gmx_fjsp_set1_v2r8(fr->ic->epsfac);
 460     charge           = mdatoms->chargeA;
 461     nvdwtype         = fr->ntype;
 462     vdwparam         = fr->nbfp;
 463     vdwtype          = mdatoms->typeA;
 464
 465     vftab            = kernel_data->table_elec_vdw->data;
 466     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_elec_vdw->scale);
 467
 468     /* Avoid stupid compiler warnings */
 469     jnrA = jnrB = 0;
 470     j_coord_offsetA = 0;
 471     j_coord_offsetB = 0;
 472
 473     outeriter        = 0;
 474     inneriter        = 0;
 475
 476     /* Start outer loop over neighborlists */
 477     for(iidx=0; iidx<nri; iidx++)
 478     {
 479         /* Load shift vector for this list */
 480         i_shift_offset   = DIM*shiftidx[iidx];
 481
 482         /* Load limits for loop over neighbors */
 483         j_index_start    = jindex[iidx];
 484         j_index_end      = jindex[iidx+1];
 485
 486         /* Get outer coordinate index */
 487         inr              = iinr[iidx];
 488         i_coord_offset   = DIM*inr;
 489
 490         /* Load i particle coords and add shift vector */
 491         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 492
 493         fix0             = _fjsp_setzero_v2r8();
 494         fiy0             = _fjsp_setzero_v2r8();
 495         fiz0             = _fjsp_setzero_v2r8();
 496
 497         /* Load parameters for i particles */
 498         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 499         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 500
 501         /* Start inner kernel loop */
 502         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 503         {
 504
 505             /* Get j neighbor index, and coordinate index */
 506             jnrA             = jjnr[jidx];
 507             jnrB             = jjnr[jidx+1];
 508             j_coord_offsetA  = DIM*jnrA;
 509             j_coord_offsetB  = DIM*jnrB;
 510
 511             /* load j atom coordinates */
 512             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 513                                               &jx0,&jy0,&jz0);
 514
 515             /* Calculate displacement vector */
 516             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 517             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 518             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 519
 520             /* Calculate squared distance and things based on it */
 521             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 522
 523             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 524
 525             /* Load parameters for j particles */
 526             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 527             vdwjidx0A        = 2*vdwtype[jnrA+0];
 528             vdwjidx0B        = 2*vdwtype[jnrB+0];
 529
 530             /**************************
 531              * CALCULATE INTERACTIONS *
 532              **************************/
 533
 534             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 535
 536             /* Compute parameters for interactions between i and j atoms */
 537             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 538             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 539                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 540
 541             /* Calculate table index by multiplying r with table scale and truncate to integer */
 542             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 543             itab_tmp         = _fjsp_dtox_v2r8(rt);
 544             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 545             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 546             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 547
 548             vfconv.i[0]     *= 12;
 549             vfconv.i[1]     *= 12;
 550
 551             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 552             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 553             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 554             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 555             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] +2);
 556             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] +2);
 557             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 558             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(vfeps,H,G),F);
 559             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twovfeps,H,G),vfeps,Fp);
 560             felec            = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,FF),_fjsp_mul_v2r8(vftabscale,rinv00)));
 561
 562             /* CUBIC SPLINE TABLE DISPERSION */
 563             vfconv.i[0]       += 4;
 564             vfconv.i[1]       += 4;
 565             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 566             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 567             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 568             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 569             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 2 );
 570             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 571             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 572             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 573             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 574
 575             /* CUBIC SPLINE TABLE REPULSION */
 576             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 577             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 4 );
 578             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 579             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 580             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 6 );
 581             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 582             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 583             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 584             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 585             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 586
 587             fscal            = _fjsp_add_v2r8(felec,fvdw);
 588
 589             /* Update vectorial force */
 590             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 591             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 592             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 593
 594             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 595
 596             /* Inner loop uses 64 flops */
 597         }
 598
 599         if(jidx<j_index_end)
 600         {
 601
 602             jnrA             = jjnr[jidx];
 603             j_coord_offsetA  = DIM*jnrA;
 604
 605             /* load j atom coordinates */
 606             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 607                                               &jx0,&jy0,&jz0);
 608
 609             /* Calculate displacement vector */
 610             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 611             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 612             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 613
 614             /* Calculate squared distance and things based on it */
 615             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 616
 617             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 618
 619             /* Load parameters for j particles */
 620             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 621             vdwjidx0A        = 2*vdwtype[jnrA+0];
 622
 623             /**************************
 624              * CALCULATE INTERACTIONS *
 625              **************************/
 626
 627             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 628
 629             /* Compute parameters for interactions between i and j atoms */
 630             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 631             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 632                                          vdwparam+vdwioffset0+vdwjidx0B,&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 }