src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecEw_VdwLJEw_GeomW3P1_sparc64_hpc_ace_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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_ElecEw_VdwLJEw_GeomW3P1_VF_sparc64_hpc_ace_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LJEwald
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_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              vdwioffset1;
  82     _fjsp_v2r8       ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  83     int              vdwioffset2;
  84     _fjsp_v2r8       ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  85     int              vdwjidx0A,vdwjidx0B;
  86     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     _fjsp_v2r8       dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  89     _fjsp_v2r8       dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  90     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     int              nvdwtype;
  93     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
  97     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
  98     _fjsp_v2r8           c6grid_00;
  99     _fjsp_v2r8           c6grid_10;
 100     _fjsp_v2r8           c6grid_20;
 101     real                 *vdwgridparam;
 102     _fjsp_v2r8           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 103     _fjsp_v2r8           one_half = gmx_fjsp_set1_v2r8(0.5);
 104     _fjsp_v2r8           minus_one = gmx_fjsp_set1_v2r8(-1.0);
 105     _fjsp_v2r8       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 106     real             *ewtab;
 107     _fjsp_v2r8       itab_tmp;
 108     _fjsp_v2r8       dummy_mask,cutoff_mask;
 109     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 110     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 111     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 112
 113     x                = xx[0];
 114     f                = ff[0];
 115
 116     nri              = nlist->nri;
 117     iinr             = nlist->iinr;
 118     jindex           = nlist->jindex;
 119     jjnr             = nlist->jjnr;
 120     shiftidx         = nlist->shift;
 121     gid              = nlist->gid;
 122     shiftvec         = fr->shift_vec[0];
 123     fshift           = fr->fshift[0];
 124     facel            = gmx_fjsp_set1_v2r8(fr->ic->epsfac);
 125     charge           = mdatoms->chargeA;
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129     vdwgridparam     = fr->ljpme_c6grid;
 130     sh_lj_ewald      = gmx_fjsp_set1_v2r8(fr->ic->sh_lj_ewald);
 131     ewclj            = gmx_fjsp_set1_v2r8(fr->ic->ewaldcoeff_lj);
 132     ewclj2           = _fjsp_mul_v2r8(minus_one,_fjsp_mul_v2r8(ewclj,ewclj));
 133
 134     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 135     ewtab            = fr->ic->tabq_coul_FDV0;
 136     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 137     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 138
 139     /* Setup water-specific parameters */
 140     inr              = nlist->iinr[0];
 141     iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+0]));
 142     iq1              = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+1]));
 143     iq2              = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+2]));
 144     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 145
 146     /* Avoid stupid compiler warnings */
 147     jnrA = jnrB = 0;
 148     j_coord_offsetA = 0;
 149     j_coord_offsetB = 0;
 150
 151     outeriter        = 0;
 152     inneriter        = 0;
 153
 154     /* Start outer loop over neighborlists */
 155     for(iidx=0; iidx<nri; iidx++)
 156     {
 157         /* Load shift vector for this list */
 158         i_shift_offset   = DIM*shiftidx[iidx];
 159
 160         /* Load limits for loop over neighbors */
 161         j_index_start    = jindex[iidx];
 162         j_index_end      = jindex[iidx+1];
 163
 164         /* Get outer coordinate index */
 165         inr              = iinr[iidx];
 166         i_coord_offset   = DIM*inr;
 167
 168         /* Load i particle coords and add shift vector */
 169         gmx_fjsp_load_shift_and_3rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,
 170                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 171
 172         fix0             = _fjsp_setzero_v2r8();
 173         fiy0             = _fjsp_setzero_v2r8();
 174         fiz0             = _fjsp_setzero_v2r8();
 175         fix1             = _fjsp_setzero_v2r8();
 176         fiy1             = _fjsp_setzero_v2r8();
 177         fiz1             = _fjsp_setzero_v2r8();
 178         fix2             = _fjsp_setzero_v2r8();
 179         fiy2             = _fjsp_setzero_v2r8();
 180         fiz2             = _fjsp_setzero_v2r8();
 181
 182         /* Reset potential sums */
 183         velecsum         = _fjsp_setzero_v2r8();
 184         vvdwsum          = _fjsp_setzero_v2r8();
 185
 186         /* Start inner kernel loop */
 187         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 188         {
 189
 190             /* Get j neighbor index, and coordinate index */
 191             jnrA             = jjnr[jidx];
 192             jnrB             = jjnr[jidx+1];
 193             j_coord_offsetA  = DIM*jnrA;
 194             j_coord_offsetB  = DIM*jnrB;
 195
 196             /* load j atom coordinates */
 197             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 198                                               &jx0,&jy0,&jz0);
 199
 200             /* Calculate displacement vector */
 201             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 202             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 203             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 204             dx10             = _fjsp_sub_v2r8(ix1,jx0);
 205             dy10             = _fjsp_sub_v2r8(iy1,jy0);
 206             dz10             = _fjsp_sub_v2r8(iz1,jz0);
 207             dx20             = _fjsp_sub_v2r8(ix2,jx0);
 208             dy20             = _fjsp_sub_v2r8(iy2,jy0);
 209             dz20             = _fjsp_sub_v2r8(iz2,jz0);
 210
 211             /* Calculate squared distance and things based on it */
 212             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 213             rsq10            = gmx_fjsp_calc_rsq_v2r8(dx10,dy10,dz10);
 214             rsq20            = gmx_fjsp_calc_rsq_v2r8(dx20,dy20,dz20);
 215
 216             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 217             rinv10           = gmx_fjsp_invsqrt_v2r8(rsq10);
 218             rinv20           = gmx_fjsp_invsqrt_v2r8(rsq20);
 219
 220             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 221             rinvsq10         = _fjsp_mul_v2r8(rinv10,rinv10);
 222             rinvsq20         = _fjsp_mul_v2r8(rinv20,rinv20);
 223
 224             /* Load parameters for j particles */
 225             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 226             vdwjidx0A        = 2*vdwtype[jnrA+0];
 227             vdwjidx0B        = 2*vdwtype[jnrB+0];
 228
 229             fjx0             = _fjsp_setzero_v2r8();
 230             fjy0             = _fjsp_setzero_v2r8();
 231             fjz0             = _fjsp_setzero_v2r8();
 232
 233             /**************************
 234              * CALCULATE INTERACTIONS *
 235              **************************/
 236
 237             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 238
 239             /* Compute parameters for interactions between i and j atoms */
 240             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 241             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 242                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 243
 244             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 245                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 246
 247             /* EWALD ELECTROSTATICS */
 248
 249             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 250             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 251             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 252             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 253             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 254
 255             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 256             ewtabD           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
 257             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 258             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 259             ewtabFn          = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
 260             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 261             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 262             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 263             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 264             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 265
 266             /* Analytical LJ-PME */
 267             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 268             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 269             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 270             exponent         = gmx_simd_exp_d(ewcljrsq);
 271             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 272             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 273             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 274             vvdw6            = _fjsp_mul_v2r8(_fjsp_madd_v2r8(c6grid_00,_fjsp_sub_v2r8(poly,one),c6_00),rinvsix);
 275             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 276             vvdw             = _fjsp_msub_v2r8(vvdw12,one_twelfth,_fjsp_mul_v2r8(vvdw6,one_sixth));
 277             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 278             fvdw             = _fjsp_mul_v2r8(_fjsp_add_v2r8(vvdw12,_fjsp_msub_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6),vvdw6)),rinvsq00);
 279
 280             /* Update potential sum for this i atom from the interaction with this j atom. */
 281             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 282             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 283
 284             fscal            = _fjsp_add_v2r8(felec,fvdw);
 285
 286             /* Update vectorial force */
 287             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 288             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 289             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 290
 291             fjx0             = _fjsp_madd_v2r8(dx00,fscal,fjx0);
 292             fjy0             = _fjsp_madd_v2r8(dy00,fscal,fjy0);
 293             fjz0             = _fjsp_madd_v2r8(dz00,fscal,fjz0);
 294
 295             /**************************
 296              * CALCULATE INTERACTIONS *
 297              **************************/
 298
 299             r10              = _fjsp_mul_v2r8(rsq10,rinv10);
 300
 301             /* Compute parameters for interactions between i and j atoms */
 302             qq10             = _fjsp_mul_v2r8(iq1,jq0);
 303
 304             /* EWALD ELECTROSTATICS */
 305
 306             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 307             ewrt             = _fjsp_mul_v2r8(r10,ewtabscale);
 308             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 309             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 310             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 311
 312             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 313             ewtabD           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
 314             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 315             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 316             ewtabFn          = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
 317             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 318             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 319             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 320             velec            = _fjsp_mul_v2r8(qq10,_fjsp_sub_v2r8(rinv10,velec));
 321             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq10,rinv10),_fjsp_sub_v2r8(rinvsq10,felec));
 322
 323             /* Update potential sum for this i atom from the interaction with this j atom. */
 324             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 325
 326             fscal            = felec;
 327
 328             /* Update vectorial force */
 329             fix1             = _fjsp_madd_v2r8(dx10,fscal,fix1);
 330             fiy1             = _fjsp_madd_v2r8(dy10,fscal,fiy1);
 331             fiz1             = _fjsp_madd_v2r8(dz10,fscal,fiz1);
 332
 333             fjx0             = _fjsp_madd_v2r8(dx10,fscal,fjx0);
 334             fjy0             = _fjsp_madd_v2r8(dy10,fscal,fjy0);
 335             fjz0             = _fjsp_madd_v2r8(dz10,fscal,fjz0);
 336
 337             /**************************
 338              * CALCULATE INTERACTIONS *
 339              **************************/
 340
 341             r20              = _fjsp_mul_v2r8(rsq20,rinv20);
 342
 343             /* Compute parameters for interactions between i and j atoms */
 344             qq20             = _fjsp_mul_v2r8(iq2,jq0);
 345
 346             /* EWALD ELECTROSTATICS */
 347
 348             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 349             ewrt             = _fjsp_mul_v2r8(r20,ewtabscale);
 350             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 351             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 352             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 353
 354             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 355             ewtabD           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
 356             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 357             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 358             ewtabFn          = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
 359             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 360             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 361             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 362             velec            = _fjsp_mul_v2r8(qq20,_fjsp_sub_v2r8(rinv20,velec));
 363             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq20,rinv20),_fjsp_sub_v2r8(rinvsq20,felec));
 364
 365             /* Update potential sum for this i atom from the interaction with this j atom. */
 366             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 367
 368             fscal            = felec;
 369
 370             /* Update vectorial force */
 371             fix2             = _fjsp_madd_v2r8(dx20,fscal,fix2);
 372             fiy2             = _fjsp_madd_v2r8(dy20,fscal,fiy2);
 373             fiz2             = _fjsp_madd_v2r8(dz20,fscal,fiz2);
 374
 375             fjx0             = _fjsp_madd_v2r8(dx20,fscal,fjx0);
 376             fjy0             = _fjsp_madd_v2r8(dy20,fscal,fjy0);
 377             fjz0             = _fjsp_madd_v2r8(dz20,fscal,fjz0);
 378
 379             gmx_fjsp_decrement_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 380
 381             /* Inner loop uses 159 flops */
 382         }
 383
 384         if(jidx<j_index_end)
 385         {
 386
 387             jnrA             = jjnr[jidx];
 388             j_coord_offsetA  = DIM*jnrA;
 389
 390             /* load j atom coordinates */
 391             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 392                                               &jx0,&jy0,&jz0);
 393
 394             /* Calculate displacement vector */
 395             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 396             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 397             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 398             dx10             = _fjsp_sub_v2r8(ix1,jx0);
 399             dy10             = _fjsp_sub_v2r8(iy1,jy0);
 400             dz10             = _fjsp_sub_v2r8(iz1,jz0);
 401             dx20             = _fjsp_sub_v2r8(ix2,jx0);
 402             dy20             = _fjsp_sub_v2r8(iy2,jy0);
 403             dz20             = _fjsp_sub_v2r8(iz2,jz0);
 404
 405             /* Calculate squared distance and things based on it */
 406             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 407             rsq10            = gmx_fjsp_calc_rsq_v2r8(dx10,dy10,dz10);
 408             rsq20            = gmx_fjsp_calc_rsq_v2r8(dx20,dy20,dz20);
 409
 410             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 411             rinv10           = gmx_fjsp_invsqrt_v2r8(rsq10);
 412             rinv20           = gmx_fjsp_invsqrt_v2r8(rsq20);
 413
 414             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 415             rinvsq10         = _fjsp_mul_v2r8(rinv10,rinv10);
 416             rinvsq20         = _fjsp_mul_v2r8(rinv20,rinv20);
 417
 418             /* Load parameters for j particles */
 419             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 420             vdwjidx0A        = 2*vdwtype[jnrA+0];
 421
 422             fjx0             = _fjsp_setzero_v2r8();
 423             fjy0             = _fjsp_setzero_v2r8();
 424             fjz0             = _fjsp_setzero_v2r8();
 425
 426             /**************************
 427              * CALCULATE INTERACTIONS *
 428              **************************/
 429
 430             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 431
 432             /* Compute parameters for interactions between i and j atoms */
 433             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 434             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 435                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 436
 437             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 438                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 439
 440             /* EWALD ELECTROSTATICS */
 441
 442             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 443             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 444             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 445             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 446             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 447
 448             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 449             ewtabD           = _fjsp_setzero_v2r8();
 450             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 451             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 452             ewtabFn          = _fjsp_setzero_v2r8();
 453             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 454             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 455             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 456             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 457             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 458
 459             /* Analytical LJ-PME */
 460             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 461             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 462             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 463             exponent         = gmx_simd_exp_d(ewcljrsq);
 464             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 465             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 466             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 467             vvdw6            = _fjsp_mul_v2r8(_fjsp_madd_v2r8(c6grid_00,_fjsp_sub_v2r8(poly,one),c6_00),rinvsix);
 468             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 469             vvdw             = _fjsp_msub_v2r8(vvdw12,one_twelfth,_fjsp_mul_v2r8(vvdw6,one_sixth));
 470             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 471             fvdw             = _fjsp_mul_v2r8(_fjsp_add_v2r8(vvdw12,_fjsp_msub_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6),vvdw6)),rinvsq00);
 472
 473             /* Update potential sum for this i atom from the interaction with this j atom. */
 474             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 475             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 476             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 477             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 478
 479             fscal            = _fjsp_add_v2r8(felec,fvdw);
 480
 481             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 482
 483             /* Update vectorial force */
 484             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 485             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 486             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 487
 488             fjx0             = _fjsp_madd_v2r8(dx00,fscal,fjx0);
 489             fjy0             = _fjsp_madd_v2r8(dy00,fscal,fjy0);
 490             fjz0             = _fjsp_madd_v2r8(dz00,fscal,fjz0);
 491
 492             /**************************
 493              * CALCULATE INTERACTIONS *
 494              **************************/
 495
 496             r10              = _fjsp_mul_v2r8(rsq10,rinv10);
 497
 498             /* Compute parameters for interactions between i and j atoms */
 499             qq10             = _fjsp_mul_v2r8(iq1,jq0);
 500
 501             /* EWALD ELECTROSTATICS */
 502
 503             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 504             ewrt             = _fjsp_mul_v2r8(r10,ewtabscale);
 505             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 506             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 507             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 508
 509             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 510             ewtabD           = _fjsp_setzero_v2r8();
 511             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 512             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 513             ewtabFn          = _fjsp_setzero_v2r8();
 514             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 515             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 516             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 517             velec            = _fjsp_mul_v2r8(qq10,_fjsp_sub_v2r8(rinv10,velec));
 518             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq10,rinv10),_fjsp_sub_v2r8(rinvsq10,felec));
 519
 520             /* Update potential sum for this i atom from the interaction with this j atom. */
 521             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 522             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 523
 524             fscal            = felec;
 525
 526             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 527
 528             /* Update vectorial force */
 529             fix1             = _fjsp_madd_v2r8(dx10,fscal,fix1);
 530             fiy1             = _fjsp_madd_v2r8(dy10,fscal,fiy1);
 531             fiz1             = _fjsp_madd_v2r8(dz10,fscal,fiz1);
 532
 533             fjx0             = _fjsp_madd_v2r8(dx10,fscal,fjx0);
 534             fjy0             = _fjsp_madd_v2r8(dy10,fscal,fjy0);
 535             fjz0             = _fjsp_madd_v2r8(dz10,fscal,fjz0);
 536
 537             /**************************
 538              * CALCULATE INTERACTIONS *
 539              **************************/
 540
 541             r20              = _fjsp_mul_v2r8(rsq20,rinv20);
 542
 543             /* Compute parameters for interactions between i and j atoms */
 544             qq20             = _fjsp_mul_v2r8(iq2,jq0);
 545
 546             /* EWALD ELECTROSTATICS */
 547
 548             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 549             ewrt             = _fjsp_mul_v2r8(r20,ewtabscale);
 550             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 551             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 552             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 553
 554             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 555             ewtabD           = _fjsp_setzero_v2r8();
 556             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 557             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 558             ewtabFn          = _fjsp_setzero_v2r8();
 559             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 560             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 561             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 562             velec            = _fjsp_mul_v2r8(qq20,_fjsp_sub_v2r8(rinv20,velec));
 563             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq20,rinv20),_fjsp_sub_v2r8(rinvsq20,felec));
 564
 565             /* Update potential sum for this i atom from the interaction with this j atom. */
 566             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 567             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 568
 569             fscal            = felec;
 570
 571             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 572
 573             /* Update vectorial force */
 574             fix2             = _fjsp_madd_v2r8(dx20,fscal,fix2);
 575             fiy2             = _fjsp_madd_v2r8(dy20,fscal,fiy2);
 576             fiz2             = _fjsp_madd_v2r8(dz20,fscal,fiz2);
 577
 578             fjx0             = _fjsp_madd_v2r8(dx20,fscal,fjx0);
 579             fjy0             = _fjsp_madd_v2r8(dy20,fscal,fjy0);
 580             fjz0             = _fjsp_madd_v2r8(dz20,fscal,fjz0);
 581
 582             gmx_fjsp_decrement_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 583
 584             /* Inner loop uses 159 flops */
 585         }
 586
 587         /* End of innermost loop */
 588
 589         gmx_fjsp_update_iforce_3atom_swizzle_v2r8(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 590                                               f+i_coord_offset,fshift+i_shift_offset);
 591
 592         ggid                        = gid[iidx];
 593         /* Update potential energies */
 594         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 595         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 596
 597         /* Increment number of inner iterations */
 598         inneriter                  += j_index_end - j_index_start;
 599
 600         /* Outer loop uses 20 flops */
 601     }
 602
 603     /* Increment number of outer iterations */
 604     outeriter        += nri;
 605
 606     /* Update outer/inner flops */
 607
 608     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*159);
 609 }
 610 /*
 611  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_sparc64_hpc_ace_double
 612  * Electrostatics interaction: Ewald
 613  * VdW interaction:            LJEwald
 614  * Geometry:                   Water3-Particle
 615  * Calculate force/pot:        Force
 616  */
 617 void
 618 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_sparc64_hpc_ace_double
 619                     (t_nblist                    * gmx_restrict       nlist,
 620                      rvec                        * gmx_restrict          xx,
 621                      rvec                        * gmx_restrict          ff,
 622                      struct t_forcerec           * gmx_restrict          fr,
 623                      t_mdatoms                   * gmx_restrict     mdatoms,
 624                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 625                      t_nrnb                      * gmx_restrict        nrnb)
 626 {
 627     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 628      * just 0 for non-waters.
 629      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 630      * jnr indices corresponding to data put in the four positions in the SIMD register.
 631      */
 632     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 633     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 634     int              jnrA,jnrB;
 635     int              j_coord_offsetA,j_coord_offsetB;
 636     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 637     real             rcutoff_scalar;
 638     real             *shiftvec,*fshift,*x,*f;
 639     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 640     int              vdwioffset0;
 641     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 642     int              vdwioffset1;
 643     _fjsp_v2r8       ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 644     int              vdwioffset2;
 645     _fjsp_v2r8       ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 646     int              vdwjidx0A,vdwjidx0B;
 647     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 648     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 649     _fjsp_v2r8       dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 650     _fjsp_v2r8       dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 651     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 652     real             *charge;
 653     int              nvdwtype;
 654     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 655     int              *vdwtype;
 656     real             *vdwparam;
 657     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 658     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 659     _fjsp_v2r8           c6grid_00;
 660     _fjsp_v2r8           c6grid_10;
 661     _fjsp_v2r8           c6grid_20;
 662     real                 *vdwgridparam;
 663     _fjsp_v2r8           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 664     _fjsp_v2r8           one_half = gmx_fjsp_set1_v2r8(0.5);
 665     _fjsp_v2r8           minus_one = gmx_fjsp_set1_v2r8(-1.0);
 666     _fjsp_v2r8       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 667     real             *ewtab;
 668     _fjsp_v2r8       itab_tmp;
 669     _fjsp_v2r8       dummy_mask,cutoff_mask;
 670     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 671     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 672     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 673
 674     x                = xx[0];
 675     f                = ff[0];
 676
 677     nri              = nlist->nri;
 678     iinr             = nlist->iinr;
 679     jindex           = nlist->jindex;
 680     jjnr             = nlist->jjnr;
 681     shiftidx         = nlist->shift;
 682     gid              = nlist->gid;
 683     shiftvec         = fr->shift_vec[0];
 684     fshift           = fr->fshift[0];
 685     facel            = gmx_fjsp_set1_v2r8(fr->ic->epsfac);
 686     charge           = mdatoms->chargeA;
 687     nvdwtype         = fr->ntype;
 688     vdwparam         = fr->nbfp;
 689     vdwtype          = mdatoms->typeA;
 690     vdwgridparam     = fr->ljpme_c6grid;
 691     sh_lj_ewald      = gmx_fjsp_set1_v2r8(fr->ic->sh_lj_ewald);
 692     ewclj            = gmx_fjsp_set1_v2r8(fr->ic->ewaldcoeff_lj);
 693     ewclj2           = _fjsp_mul_v2r8(minus_one,_fjsp_mul_v2r8(ewclj,ewclj));
 694
 695     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 696     ewtab            = fr->ic->tabq_coul_F;
 697     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 698     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 699
 700     /* Setup water-specific parameters */
 701     inr              = nlist->iinr[0];
 702     iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+0]));
 703     iq1              = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+1]));
 704     iq2              = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+2]));
 705     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 706
 707     /* Avoid stupid compiler warnings */
 708     jnrA = jnrB = 0;
 709     j_coord_offsetA = 0;
 710     j_coord_offsetB = 0;
 711
 712     outeriter        = 0;
 713     inneriter        = 0;
 714
 715     /* Start outer loop over neighborlists */
 716     for(iidx=0; iidx<nri; iidx++)
 717     {
 718         /* Load shift vector for this list */
 719         i_shift_offset   = DIM*shiftidx[iidx];
 720
 721         /* Load limits for loop over neighbors */
 722         j_index_start    = jindex[iidx];
 723         j_index_end      = jindex[iidx+1];
 724
 725         /* Get outer coordinate index */
 726         inr              = iinr[iidx];
 727         i_coord_offset   = DIM*inr;
 728
 729         /* Load i particle coords and add shift vector */
 730         gmx_fjsp_load_shift_and_3rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,
 731                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 732
 733         fix0             = _fjsp_setzero_v2r8();
 734         fiy0             = _fjsp_setzero_v2r8();
 735         fiz0             = _fjsp_setzero_v2r8();
 736         fix1             = _fjsp_setzero_v2r8();
 737         fiy1             = _fjsp_setzero_v2r8();
 738         fiz1             = _fjsp_setzero_v2r8();
 739         fix2             = _fjsp_setzero_v2r8();
 740         fiy2             = _fjsp_setzero_v2r8();
 741         fiz2             = _fjsp_setzero_v2r8();
 742
 743         /* Start inner kernel loop */
 744         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 745         {
 746
 747             /* Get j neighbor index, and coordinate index */
 748             jnrA             = jjnr[jidx];
 749             jnrB             = jjnr[jidx+1];
 750             j_coord_offsetA  = DIM*jnrA;
 751             j_coord_offsetB  = DIM*jnrB;
 752
 753             /* load j atom coordinates */
 754             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 755                                               &jx0,&jy0,&jz0);
 756
 757             /* Calculate displacement vector */
 758             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 759             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 760             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 761             dx10             = _fjsp_sub_v2r8(ix1,jx0);
 762             dy10             = _fjsp_sub_v2r8(iy1,jy0);
 763             dz10             = _fjsp_sub_v2r8(iz1,jz0);
 764             dx20             = _fjsp_sub_v2r8(ix2,jx0);
 765             dy20             = _fjsp_sub_v2r8(iy2,jy0);
 766             dz20             = _fjsp_sub_v2r8(iz2,jz0);
 767
 768             /* Calculate squared distance and things based on it */
 769             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 770             rsq10            = gmx_fjsp_calc_rsq_v2r8(dx10,dy10,dz10);
 771             rsq20            = gmx_fjsp_calc_rsq_v2r8(dx20,dy20,dz20);
 772
 773             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 774             rinv10           = gmx_fjsp_invsqrt_v2r8(rsq10);
 775             rinv20           = gmx_fjsp_invsqrt_v2r8(rsq20);
 776
 777             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 778             rinvsq10         = _fjsp_mul_v2r8(rinv10,rinv10);
 779             rinvsq20         = _fjsp_mul_v2r8(rinv20,rinv20);
 780
 781             /* Load parameters for j particles */
 782             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 783             vdwjidx0A        = 2*vdwtype[jnrA+0];
 784             vdwjidx0B        = 2*vdwtype[jnrB+0];
 785
 786             fjx0             = _fjsp_setzero_v2r8();
 787             fjy0             = _fjsp_setzero_v2r8();
 788             fjz0             = _fjsp_setzero_v2r8();
 789
 790             /**************************
 791              * CALCULATE INTERACTIONS *
 792              **************************/
 793
 794             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 795
 796             /* Compute parameters for interactions between i and j atoms */
 797             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 798             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 799                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 800
 801             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 802                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 803
 804             /* EWALD ELECTROSTATICS */
 805
 806             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 807             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 808             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 809             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 810             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 811
 812             gmx_fjsp_load_2pair_swizzle_v2r8(ewtab+ewconv.i[0],ewtab+ewconv.i[1],
 813                                          &ewtabF,&ewtabFn);
 814             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 815             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 816
 817             /* Analytical LJ-PME */
 818             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 819             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 820             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 821             exponent         = gmx_simd_exp_d(ewcljrsq);
 822             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 823             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 824             /* f6A = 6 * C6grid * (1 - poly) */
 825             f6A              = _fjsp_mul_v2r8(c6grid_00,_fjsp_sub_v2r8(one,poly));
 826             /* f6B = C6grid * exponent * beta^6 */
 827             f6B              = _fjsp_mul_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6));
 828             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 829             fvdw              = _fjsp_mul_v2r8(_fjsp_madd_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,_fjsp_sub_v2r8(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 830
 831             fscal            = _fjsp_add_v2r8(felec,fvdw);
 832
 833             /* Update vectorial force */
 834             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 835             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 836             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 837
 838             fjx0             = _fjsp_madd_v2r8(dx00,fscal,fjx0);
 839             fjy0             = _fjsp_madd_v2r8(dy00,fscal,fjy0);
 840             fjz0             = _fjsp_madd_v2r8(dz00,fscal,fjz0);
 841
 842             /**************************
 843              * CALCULATE INTERACTIONS *
 844              **************************/
 845
 846             r10              = _fjsp_mul_v2r8(rsq10,rinv10);
 847
 848             /* Compute parameters for interactions between i and j atoms */
 849             qq10             = _fjsp_mul_v2r8(iq1,jq0);
 850
 851             /* EWALD ELECTROSTATICS */
 852
 853             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 854             ewrt             = _fjsp_mul_v2r8(r10,ewtabscale);
 855             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 856             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 857             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 858
 859             gmx_fjsp_load_2pair_swizzle_v2r8(ewtab+ewconv.i[0],ewtab+ewconv.i[1],
 860                                          &ewtabF,&ewtabFn);
 861             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 862             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq10,rinv10),_fjsp_sub_v2r8(rinvsq10,felec));
 863
 864             fscal            = felec;
 865
 866             /* Update vectorial force */
 867             fix1             = _fjsp_madd_v2r8(dx10,fscal,fix1);
 868             fiy1             = _fjsp_madd_v2r8(dy10,fscal,fiy1);
 869             fiz1             = _fjsp_madd_v2r8(dz10,fscal,fiz1);
 870
 871             fjx0             = _fjsp_madd_v2r8(dx10,fscal,fjx0);
 872             fjy0             = _fjsp_madd_v2r8(dy10,fscal,fjy0);
 873             fjz0             = _fjsp_madd_v2r8(dz10,fscal,fjz0);
 874
 875             /**************************
 876              * CALCULATE INTERACTIONS *
 877              **************************/
 878
 879             r20              = _fjsp_mul_v2r8(rsq20,rinv20);
 880
 881             /* Compute parameters for interactions between i and j atoms */
 882             qq20             = _fjsp_mul_v2r8(iq2,jq0);
 883
 884             /* EWALD ELECTROSTATICS */
 885
 886             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 887             ewrt             = _fjsp_mul_v2r8(r20,ewtabscale);
 888             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 889             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 890             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 891
 892             gmx_fjsp_load_2pair_swizzle_v2r8(ewtab+ewconv.i[0],ewtab+ewconv.i[1],
 893                                          &ewtabF,&ewtabFn);
 894             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 895             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq20,rinv20),_fjsp_sub_v2r8(rinvsq20,felec));
 896
 897             fscal            = felec;
 898
 899             /* Update vectorial force */
 900             fix2             = _fjsp_madd_v2r8(dx20,fscal,fix2);
 901             fiy2             = _fjsp_madd_v2r8(dy20,fscal,fiy2);
 902             fiz2             = _fjsp_madd_v2r8(dz20,fscal,fiz2);
 903
 904             fjx0             = _fjsp_madd_v2r8(dx20,fscal,fjx0);
 905             fjy0             = _fjsp_madd_v2r8(dy20,fscal,fjy0);
 906             fjz0             = _fjsp_madd_v2r8(dz20,fscal,fjz0);
 907
 908             gmx_fjsp_decrement_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 909
 910             /* Inner loop uses 142 flops */
 911         }
 912
 913         if(jidx<j_index_end)
 914         {
 915
 916             jnrA             = jjnr[jidx];
 917             j_coord_offsetA  = DIM*jnrA;
 918
 919             /* load j atom coordinates */
 920             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 921                                               &jx0,&jy0,&jz0);
 922
 923             /* Calculate displacement vector */
 924             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 925             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 926             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 927             dx10             = _fjsp_sub_v2r8(ix1,jx0);
 928             dy10             = _fjsp_sub_v2r8(iy1,jy0);
 929             dz10             = _fjsp_sub_v2r8(iz1,jz0);
 930             dx20             = _fjsp_sub_v2r8(ix2,jx0);
 931             dy20             = _fjsp_sub_v2r8(iy2,jy0);
 932             dz20             = _fjsp_sub_v2r8(iz2,jz0);
 933
 934             /* Calculate squared distance and things based on it */
 935             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 936             rsq10            = gmx_fjsp_calc_rsq_v2r8(dx10,dy10,dz10);
 937             rsq20            = gmx_fjsp_calc_rsq_v2r8(dx20,dy20,dz20);
 938
 939             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 940             rinv10           = gmx_fjsp_invsqrt_v2r8(rsq10);
 941             rinv20           = gmx_fjsp_invsqrt_v2r8(rsq20);
 942
 943             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 944             rinvsq10         = _fjsp_mul_v2r8(rinv10,rinv10);
 945             rinvsq20         = _fjsp_mul_v2r8(rinv20,rinv20);
 946
 947             /* Load parameters for j particles */
 948             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 949             vdwjidx0A        = 2*vdwtype[jnrA+0];
 950
 951             fjx0             = _fjsp_setzero_v2r8();
 952             fjy0             = _fjsp_setzero_v2r8();
 953             fjz0             = _fjsp_setzero_v2r8();
 954
 955             /**************************
 956              * CALCULATE INTERACTIONS *
 957              **************************/
 958
 959             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 960
 961             /* Compute parameters for interactions between i and j atoms */
 962             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 963             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 964                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 965
 966             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 967                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 968
 969             /* EWALD ELECTROSTATICS */
 970
 971             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 972             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 973             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 974             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 975             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 976
 977             gmx_fjsp_load_1pair_swizzle_v2r8(ewtab+ewconv.i[0],&ewtabF,&ewtabFn);
 978             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 979             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 980
 981             /* Analytical LJ-PME */
 982             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 983             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 984             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 985             exponent         = gmx_simd_exp_d(ewcljrsq);
 986             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 987             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 988             /* f6A = 6 * C6grid * (1 - poly) */
 989             f6A              = _fjsp_mul_v2r8(c6grid_00,_fjsp_sub_v2r8(one,poly));
 990             /* f6B = C6grid * exponent * beta^6 */
 991             f6B              = _fjsp_mul_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6));
 992             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 993             fvdw              = _fjsp_mul_v2r8(_fjsp_madd_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,_fjsp_sub_v2r8(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 994
 995             fscal            = _fjsp_add_v2r8(felec,fvdw);
 996
 997             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 998
 999             /* Update vectorial force */
1000             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
1001             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
1002             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
1003
1004             fjx0             = _fjsp_madd_v2r8(dx00,fscal,fjx0);
1005             fjy0             = _fjsp_madd_v2r8(dy00,fscal,fjy0);
1006             fjz0             = _fjsp_madd_v2r8(dz00,fscal,fjz0);
1007
1008             /**************************
1009              * CALCULATE INTERACTIONS *
1010              **************************/
1011
1012             r10              = _fjsp_mul_v2r8(rsq10,rinv10);
1013
1014             /* Compute parameters for interactions between i and j atoms */
1015             qq10             = _fjsp_mul_v2r8(iq1,jq0);
1016
1017             /* EWALD ELECTROSTATICS */
1018
1019             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1020             ewrt             = _fjsp_mul_v2r8(r10,ewtabscale);
1021             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
1022             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
1023             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
1024
1025             gmx_fjsp_load_1pair_swizzle_v2r8(ewtab+ewconv.i[0],&ewtabF,&ewtabFn);
1026             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
1027             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq10,rinv10),_fjsp_sub_v2r8(rinvsq10,felec));
1028
1029             fscal            = felec;
1030
1031             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
1032
1033             /* Update vectorial force */
1034             fix1             = _fjsp_madd_v2r8(dx10,fscal,fix1);
1035             fiy1             = _fjsp_madd_v2r8(dy10,fscal,fiy1);
1036             fiz1             = _fjsp_madd_v2r8(dz10,fscal,fiz1);
1037
1038             fjx0             = _fjsp_madd_v2r8(dx10,fscal,fjx0);
1039             fjy0             = _fjsp_madd_v2r8(dy10,fscal,fjy0);
1040             fjz0             = _fjsp_madd_v2r8(dz10,fscal,fjz0);
1041
1042             /**************************
1043              * CALCULATE INTERACTIONS *
1044              **************************/
1045
1046             r20              = _fjsp_mul_v2r8(rsq20,rinv20);
1047
1048             /* Compute parameters for interactions between i and j atoms */
1049             qq20             = _fjsp_mul_v2r8(iq2,jq0);
1050
1051             /* EWALD ELECTROSTATICS */
1052
1053             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1054             ewrt             = _fjsp_mul_v2r8(r20,ewtabscale);
1055             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
1056             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
1057             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
1058
1059             gmx_fjsp_load_1pair_swizzle_v2r8(ewtab+ewconv.i[0],&ewtabF,&ewtabFn);
1060             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
1061             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq20,rinv20),_fjsp_sub_v2r8(rinvsq20,felec));
1062
1063             fscal            = felec;
1064
1065             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
1066
1067             /* Update vectorial force */
1068             fix2             = _fjsp_madd_v2r8(dx20,fscal,fix2);
1069             fiy2             = _fjsp_madd_v2r8(dy20,fscal,fiy2);
1070             fiz2             = _fjsp_madd_v2r8(dz20,fscal,fiz2);
1071
1072             fjx0             = _fjsp_madd_v2r8(dx20,fscal,fjx0);
1073             fjy0             = _fjsp_madd_v2r8(dy20,fscal,fjy0);
1074             fjz0             = _fjsp_madd_v2r8(dz20,fscal,fjz0);
1075
1076             gmx_fjsp_decrement_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1077
1078             /* Inner loop uses 142 flops */
1079         }
1080
1081         /* End of innermost loop */
1082
1083         gmx_fjsp_update_iforce_3atom_swizzle_v2r8(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1084                                               f+i_coord_offset,fshift+i_shift_offset);
1085
1086         /* Increment number of inner iterations */
1087         inneriter                  += j_index_end - j_index_start;
1088
1089         /* Outer loop uses 18 flops */
1090     }
1091
1092     /* Increment number of outer iterations */
1093     outeriter        += nri;
1094
1095     /* Update outer/inner flops */
1096
1097     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*142);
1098 }