src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecRFCut_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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "kernelutil_sparc64_hpc_ace_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_VF_sparc64_hpc_ace_double
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_VF_sparc64_hpc_ace_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      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->epsfac);
 112     charge           = mdatoms->chargeA;
 113     krf              = gmx_fjsp_set1_v2r8(fr->ic->k_rf);
 114     krf2             = gmx_fjsp_set1_v2r8(fr->ic->k_rf*2.0);
 115     crf              = gmx_fjsp_set1_v2r8(fr->ic->c_rf);
 116     nvdwtype         = fr->ntype;
 117     vdwparam         = fr->nbfp;
 118     vdwtype          = mdatoms->typeA;
 119
 120     vftab            = kernel_data->table_vdw->data;
 121     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_vdw->scale);
 122
 123     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 124     rcutoff_scalar   = fr->rcoulomb;
 125     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 126     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 127
 128     /* Avoid stupid compiler warnings */
 129     jnrA = jnrB = 0;
 130     j_coord_offsetA = 0;
 131     j_coord_offsetB = 0;
 132
 133     outeriter        = 0;
 134     inneriter        = 0;
 135
 136     /* Start outer loop over neighborlists */
 137     for(iidx=0; iidx<nri; iidx++)
 138     {
 139         /* Load shift vector for this list */
 140         i_shift_offset   = DIM*shiftidx[iidx];
 141
 142         /* Load limits for loop over neighbors */
 143         j_index_start    = jindex[iidx];
 144         j_index_end      = jindex[iidx+1];
 145
 146         /* Get outer coordinate index */
 147         inr              = iinr[iidx];
 148         i_coord_offset   = DIM*inr;
 149
 150         /* Load i particle coords and add shift vector */
 151         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 152
 153         fix0             = _fjsp_setzero_v2r8();
 154         fiy0             = _fjsp_setzero_v2r8();
 155         fiz0             = _fjsp_setzero_v2r8();
 156
 157         /* Load parameters for i particles */
 158         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 159         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 160
 161         /* Reset potential sums */
 162         velecsum         = _fjsp_setzero_v2r8();
 163         vvdwsum          = _fjsp_setzero_v2r8();
 164
 165         /* Start inner kernel loop */
 166         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 167         {
 168
 169             /* Get j neighbor index, and coordinate index */
 170             jnrA             = jjnr[jidx];
 171             jnrB             = jjnr[jidx+1];
 172             j_coord_offsetA  = DIM*jnrA;
 173             j_coord_offsetB  = DIM*jnrB;
 174
 175             /* load j atom coordinates */
 176             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 177                                               &jx0,&jy0,&jz0);
 178
 179             /* Calculate displacement vector */
 180             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 181             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 182             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 183
 184             /* Calculate squared distance and things based on it */
 185             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 186
 187             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 188
 189             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 190
 191             /* Load parameters for j particles */
 192             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 193             vdwjidx0A        = 2*vdwtype[jnrA+0];
 194             vdwjidx0B        = 2*vdwtype[jnrB+0];
 195
 196             /**************************
 197              * CALCULATE INTERACTIONS *
 198              **************************/
 199
 200             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 201             {
 202
 203             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 204
 205             /* Compute parameters for interactions between i and j atoms */
 206             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 207             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 208                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 209
 210             /* Calculate table index by multiplying r with table scale and truncate to integer */
 211             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 212             itab_tmp         = _fjsp_dtox_v2r8(rt);
 213             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 214             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 215             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 216
 217             vfconv.i[0]     *= 8;
 218             vfconv.i[1]     *= 8;
 219
 220             /* REACTION-FIELD ELECTROSTATICS */
 221             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(_fjsp_madd_v2r8(krf,rsq00,rinv00),crf));
 222             felec            = _fjsp_mul_v2r8(qq00,_fjsp_msub_v2r8(rinv00,rinvsq00,krf2));
 223
 224             /* CUBIC SPLINE TABLE DISPERSION */
 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             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 253
 254             /* Update potential sum for this i atom from the interaction with this j atom. */
 255             velec            = _fjsp_and_v2r8(velec,cutoff_mask);
 256             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 257             vvdw             = _fjsp_and_v2r8(vvdw,cutoff_mask);
 258             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 259
 260             fscal            = _fjsp_add_v2r8(felec,fvdw);
 261
 262             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 263
 264             /* Update vectorial force */
 265             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 266             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 267             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 268
 269             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 270
 271             }
 272
 273             /* Inner loop uses 75 flops */
 274         }
 275
 276         if(jidx<j_index_end)
 277         {
 278
 279             jnrA             = jjnr[jidx];
 280             j_coord_offsetA  = DIM*jnrA;
 281
 282             /* load j atom coordinates */
 283             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 284                                               &jx0,&jy0,&jz0);
 285
 286             /* Calculate displacement vector */
 287             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 288             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 289             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 290
 291             /* Calculate squared distance and things based on it */
 292             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 293
 294             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 295
 296             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 297
 298             /* Load parameters for j particles */
 299             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 300             vdwjidx0A        = 2*vdwtype[jnrA+0];
 301
 302             /**************************
 303              * CALCULATE INTERACTIONS *
 304              **************************/
 305
 306             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 307             {
 308
 309             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 310
 311             /* Compute parameters for interactions between i and j atoms */
 312             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 313             gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 314
 315             /* Calculate table index by multiplying r with table scale and truncate to integer */
 316             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 317             itab_tmp         = _fjsp_dtox_v2r8(rt);
 318             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 319             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 320             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 321
 322             vfconv.i[0]     *= 8;
 323             vfconv.i[1]     *= 8;
 324
 325             /* REACTION-FIELD ELECTROSTATICS */
 326             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(_fjsp_madd_v2r8(krf,rsq00,rinv00),crf));
 327             felec            = _fjsp_mul_v2r8(qq00,_fjsp_msub_v2r8(rinv00,rinvsq00,krf2));
 328
 329             /* CUBIC SPLINE TABLE DISPERSION */
 330             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 331             F                = _fjsp_setzero_v2r8();
 332             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 333             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 334             H                = _fjsp_setzero_v2r8();
 335             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 336             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 337             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 338             vvdw6            = _fjsp_mul_v2r8(c6_00,VV);
 339             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 340             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 341
 342             /* CUBIC SPLINE TABLE REPULSION */
 343             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 344             F                = _fjsp_setzero_v2r8();
 345             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 346             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 347             H                = _fjsp_setzero_v2r8();
 348             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 349             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 350             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 351             vvdw12           = _fjsp_mul_v2r8(c12_00,VV);
 352             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 353             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 354             vvdw             = _fjsp_add_v2r8(vvdw12,vvdw6);
 355             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 356
 357             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 358
 359             /* Update potential sum for this i atom from the interaction with this j atom. */
 360             velec            = _fjsp_and_v2r8(velec,cutoff_mask);
 361             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 362             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 363             vvdw             = _fjsp_and_v2r8(vvdw,cutoff_mask);
 364             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 365             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 366
 367             fscal            = _fjsp_add_v2r8(felec,fvdw);
 368
 369             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 370
 371             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 372
 373             /* Update vectorial force */
 374             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 375             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 376             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 377
 378             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 379
 380             }
 381
 382             /* Inner loop uses 75 flops */
 383         }
 384
 385         /* End of innermost loop */
 386
 387         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 388                                               f+i_coord_offset,fshift+i_shift_offset);
 389
 390         ggid                        = gid[iidx];
 391         /* Update potential energies */
 392         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 393         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 394
 395         /* Increment number of inner iterations */
 396         inneriter                  += j_index_end - j_index_start;
 397
 398         /* Outer loop uses 9 flops */
 399     }
 400
 401     /* Increment number of outer iterations */
 402     outeriter        += nri;
 403
 404     /* Update outer/inner flops */
 405
 406     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*75);
 407 }
 408 /*
 409  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_F_sparc64_hpc_ace_double
 410  * Electrostatics interaction: ReactionField
 411  * VdW interaction:            CubicSplineTable
 412  * Geometry:                   Particle-Particle
 413  * Calculate force/pot:        Force
 414  */
 415 void
 416 nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_F_sparc64_hpc_ace_double
 417                     (t_nblist                    * gmx_restrict       nlist,
 418                      rvec                        * gmx_restrict          xx,
 419                      rvec                        * gmx_restrict          ff,
 420                      t_forcerec                  * gmx_restrict          fr,
 421                      t_mdatoms                   * gmx_restrict     mdatoms,
 422                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 423                      t_nrnb                      * gmx_restrict        nrnb)
 424 {
 425     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 426      * just 0 for non-waters.
 427      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 428      * jnr indices corresponding to data put in the four positions in the SIMD register.
 429      */
 430     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 431     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 432     int              jnrA,jnrB;
 433     int              j_coord_offsetA,j_coord_offsetB;
 434     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 435     real             rcutoff_scalar;
 436     real             *shiftvec,*fshift,*x,*f;
 437     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 438     int              vdwioffset0;
 439     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 440     int              vdwjidx0A,vdwjidx0B;
 441     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 442     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 443     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 444     real             *charge;
 445     int              nvdwtype;
 446     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 447     int              *vdwtype;
 448     real             *vdwparam;
 449     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 450     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 451     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 452     real             *vftab;
 453     _fjsp_v2r8       itab_tmp;
 454     _fjsp_v2r8       dummy_mask,cutoff_mask;
 455     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 456     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 457     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 458
 459     x                = xx[0];
 460     f                = ff[0];
 461
 462     nri              = nlist->nri;
 463     iinr             = nlist->iinr;
 464     jindex           = nlist->jindex;
 465     jjnr             = nlist->jjnr;
 466     shiftidx         = nlist->shift;
 467     gid              = nlist->gid;
 468     shiftvec         = fr->shift_vec[0];
 469     fshift           = fr->fshift[0];
 470     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 471     charge           = mdatoms->chargeA;
 472     krf              = gmx_fjsp_set1_v2r8(fr->ic->k_rf);
 473     krf2             = gmx_fjsp_set1_v2r8(fr->ic->k_rf*2.0);
 474     crf              = gmx_fjsp_set1_v2r8(fr->ic->c_rf);
 475     nvdwtype         = fr->ntype;
 476     vdwparam         = fr->nbfp;
 477     vdwtype          = mdatoms->typeA;
 478
 479     vftab            = kernel_data->table_vdw->data;
 480     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_vdw->scale);
 481
 482     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 483     rcutoff_scalar   = fr->rcoulomb;
 484     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 485     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 486
 487     /* Avoid stupid compiler warnings */
 488     jnrA = jnrB = 0;
 489     j_coord_offsetA = 0;
 490     j_coord_offsetB = 0;
 491
 492     outeriter        = 0;
 493     inneriter        = 0;
 494
 495     /* Start outer loop over neighborlists */
 496     for(iidx=0; iidx<nri; iidx++)
 497     {
 498         /* Load shift vector for this list */
 499         i_shift_offset   = DIM*shiftidx[iidx];
 500
 501         /* Load limits for loop over neighbors */
 502         j_index_start    = jindex[iidx];
 503         j_index_end      = jindex[iidx+1];
 504
 505         /* Get outer coordinate index */
 506         inr              = iinr[iidx];
 507         i_coord_offset   = DIM*inr;
 508
 509         /* Load i particle coords and add shift vector */
 510         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 511
 512         fix0             = _fjsp_setzero_v2r8();
 513         fiy0             = _fjsp_setzero_v2r8();
 514         fiz0             = _fjsp_setzero_v2r8();
 515
 516         /* Load parameters for i particles */
 517         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 518         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 519
 520         /* Start inner kernel loop */
 521         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 522         {
 523
 524             /* Get j neighbor index, and coordinate index */
 525             jnrA             = jjnr[jidx];
 526             jnrB             = jjnr[jidx+1];
 527             j_coord_offsetA  = DIM*jnrA;
 528             j_coord_offsetB  = DIM*jnrB;
 529
 530             /* load j atom coordinates */
 531             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 532                                               &jx0,&jy0,&jz0);
 533
 534             /* Calculate displacement vector */
 535             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 536             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 537             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 538
 539             /* Calculate squared distance and things based on it */
 540             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 541
 542             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 543
 544             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 545
 546             /* Load parameters for j particles */
 547             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 548             vdwjidx0A        = 2*vdwtype[jnrA+0];
 549             vdwjidx0B        = 2*vdwtype[jnrB+0];
 550
 551             /**************************
 552              * CALCULATE INTERACTIONS *
 553              **************************/
 554
 555             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 556             {
 557
 558             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 559
 560             /* Compute parameters for interactions between i and j atoms */
 561             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 562             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 563                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 564
 565             /* Calculate table index by multiplying r with table scale and truncate to integer */
 566             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 567             itab_tmp         = _fjsp_dtox_v2r8(rt);
 568             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 569             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 570             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 571
 572             vfconv.i[0]     *= 8;
 573             vfconv.i[1]     *= 8;
 574
 575             /* REACTION-FIELD ELECTROSTATICS */
 576             felec            = _fjsp_mul_v2r8(qq00,_fjsp_msub_v2r8(rinv00,rinvsq00,krf2));
 577
 578             /* CUBIC SPLINE TABLE DISPERSION */
 579             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 580             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 581             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 582             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 583             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 2 );
 584             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 585             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 586             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 587             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 588
 589             /* CUBIC SPLINE TABLE REPULSION */
 590             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 591             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 4 );
 592             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 593             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 594             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 6 );
 595             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 596             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 597             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 598             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 599             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 600
 601             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 602
 603             fscal            = _fjsp_add_v2r8(felec,fvdw);
 604
 605             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 606
 607             /* Update vectorial force */
 608             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 609             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 610             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 611
 612             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 613
 614             }
 615
 616             /* Inner loop uses 60 flops */
 617         }
 618
 619         if(jidx<j_index_end)
 620         {
 621
 622             jnrA             = jjnr[jidx];
 623             j_coord_offsetA  = DIM*jnrA;
 624
 625             /* load j atom coordinates */
 626             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 627                                               &jx0,&jy0,&jz0);
 628
 629             /* Calculate displacement vector */
 630             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 631             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 632             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 633
 634             /* Calculate squared distance and things based on it */
 635             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 636
 637             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 638
 639             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 640
 641             /* Load parameters for j particles */
 642             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 643             vdwjidx0A        = 2*vdwtype[jnrA+0];
 644
 645             /**************************
 646              * CALCULATE INTERACTIONS *
 647              **************************/
 648
 649             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 650             {
 651
 652             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 653
 654             /* Compute parameters for interactions between i and j atoms */
 655             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 656             gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 657
 658             /* Calculate table index by multiplying r with table scale and truncate to integer */
 659             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 660             itab_tmp         = _fjsp_dtox_v2r8(rt);
 661             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 662             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 663             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 664
 665             vfconv.i[0]     *= 8;
 666             vfconv.i[1]     *= 8;
 667
 668             /* REACTION-FIELD ELECTROSTATICS */
 669             felec            = _fjsp_mul_v2r8(qq00,_fjsp_msub_v2r8(rinv00,rinvsq00,krf2));
 670
 671             /* CUBIC SPLINE TABLE DISPERSION */
 672             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 673             F                = _fjsp_setzero_v2r8();
 674             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 675             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 676             H                = _fjsp_setzero_v2r8();
 677             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 678             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 679             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 680             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 681
 682             /* CUBIC SPLINE TABLE REPULSION */
 683             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 684             F                = _fjsp_setzero_v2r8();
 685             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 686             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 687             H                = _fjsp_setzero_v2r8();
 688             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 689             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 690             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 691             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 692             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 693
 694             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 695
 696             fscal            = _fjsp_add_v2r8(felec,fvdw);
 697
 698             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 699
 700             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 701
 702             /* Update vectorial force */
 703             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 704             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 705             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 706
 707             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 708
 709             }
 710
 711             /* Inner loop uses 60 flops */
 712         }
 713
 714         /* End of innermost loop */
 715
 716         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 717                                               f+i_coord_offset,fshift+i_shift_offset);
 718
 719         /* Increment number of inner iterations */
 720         inneriter                  += j_index_end - j_index_start;
 721
 722         /* Outer loop uses 7 flops */
 723     }
 724
 725     /* Increment number of outer iterations */
 726     outeriter        += nri;
 727
 728     /* Update outer/inner flops */
 729
 730     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*60);
 731 }