src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecEw_VdwLJEw_GeomW3P1_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 "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/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                      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->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->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(one,poly),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_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 435
 436             c6grid_00       = gmx_fjsp_load_1real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A);
 437
 438             /* EWALD ELECTROSTATICS */
 439
 440             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 441             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 442             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 443             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 444             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 445
 446             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 447             ewtabD           = _fjsp_setzero_v2r8();
 448             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 449             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 450             ewtabFn          = _fjsp_setzero_v2r8();
 451             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 452             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 453             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 454             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 455             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 456
 457             /* Analytical LJ-PME */
 458             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 459             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 460             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 461             exponent         = gmx_simd_exp_d(-ewcljrsq);
 462             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 463             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 464             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 465             vvdw6            = _fjsp_mul_v2r8(_fjsp_madd_v2r8(-c6grid_00,_fjsp_sub_v2r8(one,poly),c6_00),rinvsix);
 466             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 467             vvdw             = _fjsp_msub_v2r8(vvdw12,one_twelfth,_fjsp_mul_v2r8(vvdw6,one_sixth));
 468             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 469             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);
 470
 471             /* Update potential sum for this i atom from the interaction with this j atom. */
 472             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 473             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 474             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 475             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 476
 477             fscal            = _fjsp_add_v2r8(felec,fvdw);
 478
 479             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 480
 481             /* Update vectorial force */
 482             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 483             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 484             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 485
 486             fjx0             = _fjsp_madd_v2r8(dx00,fscal,fjx0);
 487             fjy0             = _fjsp_madd_v2r8(dy00,fscal,fjy0);
 488             fjz0             = _fjsp_madd_v2r8(dz00,fscal,fjz0);
 489
 490             /**************************
 491              * CALCULATE INTERACTIONS *
 492              **************************/
 493
 494             r10              = _fjsp_mul_v2r8(rsq10,rinv10);
 495
 496             /* Compute parameters for interactions between i and j atoms */
 497             qq10             = _fjsp_mul_v2r8(iq1,jq0);
 498
 499             /* EWALD ELECTROSTATICS */
 500
 501             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 502             ewrt             = _fjsp_mul_v2r8(r10,ewtabscale);
 503             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 504             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 505             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 506
 507             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 508             ewtabD           = _fjsp_setzero_v2r8();
 509             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 510             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 511             ewtabFn          = _fjsp_setzero_v2r8();
 512             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 513             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 514             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 515             velec            = _fjsp_mul_v2r8(qq10,_fjsp_sub_v2r8(rinv10,velec));
 516             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq10,rinv10),_fjsp_sub_v2r8(rinvsq10,felec));
 517
 518             /* Update potential sum for this i atom from the interaction with this j atom. */
 519             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 520             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 521
 522             fscal            = felec;
 523
 524             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 525
 526             /* Update vectorial force */
 527             fix1             = _fjsp_madd_v2r8(dx10,fscal,fix1);
 528             fiy1             = _fjsp_madd_v2r8(dy10,fscal,fiy1);
 529             fiz1             = _fjsp_madd_v2r8(dz10,fscal,fiz1);
 530
 531             fjx0             = _fjsp_madd_v2r8(dx10,fscal,fjx0);
 532             fjy0             = _fjsp_madd_v2r8(dy10,fscal,fjy0);
 533             fjz0             = _fjsp_madd_v2r8(dz10,fscal,fjz0);
 534
 535             /**************************
 536              * CALCULATE INTERACTIONS *
 537              **************************/
 538
 539             r20              = _fjsp_mul_v2r8(rsq20,rinv20);
 540
 541             /* Compute parameters for interactions between i and j atoms */
 542             qq20             = _fjsp_mul_v2r8(iq2,jq0);
 543
 544             /* EWALD ELECTROSTATICS */
 545
 546             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 547             ewrt             = _fjsp_mul_v2r8(r20,ewtabscale);
 548             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 549             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 550             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 551
 552             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 553             ewtabD           = _fjsp_setzero_v2r8();
 554             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 555             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 556             ewtabFn          = _fjsp_setzero_v2r8();
 557             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 558             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 559             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 560             velec            = _fjsp_mul_v2r8(qq20,_fjsp_sub_v2r8(rinv20,velec));
 561             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq20,rinv20),_fjsp_sub_v2r8(rinvsq20,felec));
 562
 563             /* Update potential sum for this i atom from the interaction with this j atom. */
 564             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 565             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 566
 567             fscal            = felec;
 568
 569             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 570
 571             /* Update vectorial force */
 572             fix2             = _fjsp_madd_v2r8(dx20,fscal,fix2);
 573             fiy2             = _fjsp_madd_v2r8(dy20,fscal,fiy2);
 574             fiz2             = _fjsp_madd_v2r8(dz20,fscal,fiz2);
 575
 576             fjx0             = _fjsp_madd_v2r8(dx20,fscal,fjx0);
 577             fjy0             = _fjsp_madd_v2r8(dy20,fscal,fjy0);
 578             fjz0             = _fjsp_madd_v2r8(dz20,fscal,fjz0);
 579
 580             gmx_fjsp_decrement_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 581
 582             /* Inner loop uses 159 flops */
 583         }
 584
 585         /* End of innermost loop */
 586
 587         gmx_fjsp_update_iforce_3atom_swizzle_v2r8(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 588                                               f+i_coord_offset,fshift+i_shift_offset);
 589
 590         ggid                        = gid[iidx];
 591         /* Update potential energies */
 592         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 593         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 594
 595         /* Increment number of inner iterations */
 596         inneriter                  += j_index_end - j_index_start;
 597
 598         /* Outer loop uses 20 flops */
 599     }
 600
 601     /* Increment number of outer iterations */
 602     outeriter        += nri;
 603
 604     /* Update outer/inner flops */
 605
 606     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*159);
 607 }
 608 /*
 609  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_sparc64_hpc_ace_double
 610  * Electrostatics interaction: Ewald
 611  * VdW interaction:            LJEwald
 612  * Geometry:                   Water3-Particle
 613  * Calculate force/pot:        Force
 614  */
 615 void
 616 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_sparc64_hpc_ace_double
 617                     (t_nblist                    * gmx_restrict       nlist,
 618                      rvec                        * gmx_restrict          xx,
 619                      rvec                        * gmx_restrict          ff,
 620                      t_forcerec                  * gmx_restrict          fr,
 621                      t_mdatoms                   * gmx_restrict     mdatoms,
 622                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 623                      t_nrnb                      * gmx_restrict        nrnb)
 624 {
 625     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 626      * just 0 for non-waters.
 627      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 628      * jnr indices corresponding to data put in the four positions in the SIMD register.
 629      */
 630     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 631     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 632     int              jnrA,jnrB;
 633     int              j_coord_offsetA,j_coord_offsetB;
 634     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 635     real             rcutoff_scalar;
 636     real             *shiftvec,*fshift,*x,*f;
 637     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 638     int              vdwioffset0;
 639     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 640     int              vdwioffset1;
 641     _fjsp_v2r8       ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 642     int              vdwioffset2;
 643     _fjsp_v2r8       ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 644     int              vdwjidx0A,vdwjidx0B;
 645     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 646     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 647     _fjsp_v2r8       dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 648     _fjsp_v2r8       dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 649     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 650     real             *charge;
 651     int              nvdwtype;
 652     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 653     int              *vdwtype;
 654     real             *vdwparam;
 655     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 656     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 657     _fjsp_v2r8           c6grid_00;
 658     _fjsp_v2r8           c6grid_10;
 659     _fjsp_v2r8           c6grid_20;
 660     real                 *vdwgridparam;
 661     _fjsp_v2r8           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 662     _fjsp_v2r8           one_half = gmx_fjsp_set1_v2r8(0.5);
 663     _fjsp_v2r8           minus_one = gmx_fjsp_set1_v2r8(-1.0);
 664     _fjsp_v2r8       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 665     real             *ewtab;
 666     _fjsp_v2r8       itab_tmp;
 667     _fjsp_v2r8       dummy_mask,cutoff_mask;
 668     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 669     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 670     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 671
 672     x                = xx[0];
 673     f                = ff[0];
 674
 675     nri              = nlist->nri;
 676     iinr             = nlist->iinr;
 677     jindex           = nlist->jindex;
 678     jjnr             = nlist->jjnr;
 679     shiftidx         = nlist->shift;
 680     gid              = nlist->gid;
 681     shiftvec         = fr->shift_vec[0];
 682     fshift           = fr->fshift[0];
 683     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 684     charge           = mdatoms->chargeA;
 685     nvdwtype         = fr->ntype;
 686     vdwparam         = fr->nbfp;
 687     vdwtype          = mdatoms->typeA;
 688     vdwgridparam     = fr->ljpme_c6grid;
 689     sh_lj_ewald      = gmx_fjsp_set1_v2r8(fr->ic->sh_lj_ewald);
 690     ewclj            = gmx_fjsp_set1_v2r8(fr->ewaldcoeff_lj);
 691     ewclj2           = _fjsp_mul_v2r8(minus_one,_fjsp_mul_v2r8(ewclj,ewclj));
 692
 693     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 694     ewtab            = fr->ic->tabq_coul_F;
 695     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 696     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 697
 698     /* Setup water-specific parameters */
 699     inr              = nlist->iinr[0];
 700     iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+0]));
 701     iq1              = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+1]));
 702     iq2              = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+2]));
 703     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 704
 705     /* Avoid stupid compiler warnings */
 706     jnrA = jnrB = 0;
 707     j_coord_offsetA = 0;
 708     j_coord_offsetB = 0;
 709
 710     outeriter        = 0;
 711     inneriter        = 0;
 712
 713     /* Start outer loop over neighborlists */
 714     for(iidx=0; iidx<nri; iidx++)
 715     {
 716         /* Load shift vector for this list */
 717         i_shift_offset   = DIM*shiftidx[iidx];
 718
 719         /* Load limits for loop over neighbors */
 720         j_index_start    = jindex[iidx];
 721         j_index_end      = jindex[iidx+1];
 722
 723         /* Get outer coordinate index */
 724         inr              = iinr[iidx];
 725         i_coord_offset   = DIM*inr;
 726
 727         /* Load i particle coords and add shift vector */
 728         gmx_fjsp_load_shift_and_3rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,
 729                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 730
 731         fix0             = _fjsp_setzero_v2r8();
 732         fiy0             = _fjsp_setzero_v2r8();
 733         fiz0             = _fjsp_setzero_v2r8();
 734         fix1             = _fjsp_setzero_v2r8();
 735         fiy1             = _fjsp_setzero_v2r8();
 736         fiz1             = _fjsp_setzero_v2r8();
 737         fix2             = _fjsp_setzero_v2r8();
 738         fiy2             = _fjsp_setzero_v2r8();
 739         fiz2             = _fjsp_setzero_v2r8();
 740
 741         /* Start inner kernel loop */
 742         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 743         {
 744
 745             /* Get j neighbor index, and coordinate index */
 746             jnrA             = jjnr[jidx];
 747             jnrB             = jjnr[jidx+1];
 748             j_coord_offsetA  = DIM*jnrA;
 749             j_coord_offsetB  = DIM*jnrB;
 750
 751             /* load j atom coordinates */
 752             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 753                                               &jx0,&jy0,&jz0);
 754
 755             /* Calculate displacement vector */
 756             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 757             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 758             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 759             dx10             = _fjsp_sub_v2r8(ix1,jx0);
 760             dy10             = _fjsp_sub_v2r8(iy1,jy0);
 761             dz10             = _fjsp_sub_v2r8(iz1,jz0);
 762             dx20             = _fjsp_sub_v2r8(ix2,jx0);
 763             dy20             = _fjsp_sub_v2r8(iy2,jy0);
 764             dz20             = _fjsp_sub_v2r8(iz2,jz0);
 765
 766             /* Calculate squared distance and things based on it */
 767             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 768             rsq10            = gmx_fjsp_calc_rsq_v2r8(dx10,dy10,dz10);
 769             rsq20            = gmx_fjsp_calc_rsq_v2r8(dx20,dy20,dz20);
 770
 771             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 772             rinv10           = gmx_fjsp_invsqrt_v2r8(rsq10);
 773             rinv20           = gmx_fjsp_invsqrt_v2r8(rsq20);
 774
 775             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 776             rinvsq10         = _fjsp_mul_v2r8(rinv10,rinv10);
 777             rinvsq20         = _fjsp_mul_v2r8(rinv20,rinv20);
 778
 779             /* Load parameters for j particles */
 780             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 781             vdwjidx0A        = 2*vdwtype[jnrA+0];
 782             vdwjidx0B        = 2*vdwtype[jnrB+0];
 783
 784             fjx0             = _fjsp_setzero_v2r8();
 785             fjy0             = _fjsp_setzero_v2r8();
 786             fjz0             = _fjsp_setzero_v2r8();
 787
 788             /**************************
 789              * CALCULATE INTERACTIONS *
 790              **************************/
 791
 792             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 793
 794             /* Compute parameters for interactions between i and j atoms */
 795             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 796             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 797                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 798
 799             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 800                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 801
 802             /* EWALD ELECTROSTATICS */
 803
 804             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 805             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 806             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 807             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 808             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 809
 810             gmx_fjsp_load_2pair_swizzle_v2r8(ewtab+ewconv.i[0],ewtab+ewconv.i[1],
 811                                          &ewtabF,&ewtabFn);
 812             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 813             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 814
 815             /* Analytical LJ-PME */
 816             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 817             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 818             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 819             exponent         = gmx_simd_exp_d(-ewcljrsq);
 820             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 821             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 822             /* f6A = 6 * C6grid * (1 - poly) */
 823             f6A              = _fjsp_mul_v2r8(c6grid_00,_fjsp_msub_v2r8(one,poly));
 824             /* f6B = C6grid * exponent * beta^6 */
 825             f6B              = _fjsp_mul_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6));
 826             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 827             fvdw              = _fjsp_mul_v2r8(_fjsp_madd_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,_fjsp_sub_v2r8(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 828
 829             fscal            = _fjsp_add_v2r8(felec,fvdw);
 830
 831             /* Update vectorial force */
 832             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 833             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 834             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 835
 836             fjx0             = _fjsp_madd_v2r8(dx00,fscal,fjx0);
 837             fjy0             = _fjsp_madd_v2r8(dy00,fscal,fjy0);
 838             fjz0             = _fjsp_madd_v2r8(dz00,fscal,fjz0);
 839
 840             /**************************
 841              * CALCULATE INTERACTIONS *
 842              **************************/
 843
 844             r10              = _fjsp_mul_v2r8(rsq10,rinv10);
 845
 846             /* Compute parameters for interactions between i and j atoms */
 847             qq10             = _fjsp_mul_v2r8(iq1,jq0);
 848
 849             /* EWALD ELECTROSTATICS */
 850
 851             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 852             ewrt             = _fjsp_mul_v2r8(r10,ewtabscale);
 853             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 854             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 855             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 856
 857             gmx_fjsp_load_2pair_swizzle_v2r8(ewtab+ewconv.i[0],ewtab+ewconv.i[1],
 858                                          &ewtabF,&ewtabFn);
 859             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 860             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq10,rinv10),_fjsp_sub_v2r8(rinvsq10,felec));
 861
 862             fscal            = felec;
 863
 864             /* Update vectorial force */
 865             fix1             = _fjsp_madd_v2r8(dx10,fscal,fix1);
 866             fiy1             = _fjsp_madd_v2r8(dy10,fscal,fiy1);
 867             fiz1             = _fjsp_madd_v2r8(dz10,fscal,fiz1);
 868
 869             fjx0             = _fjsp_madd_v2r8(dx10,fscal,fjx0);
 870             fjy0             = _fjsp_madd_v2r8(dy10,fscal,fjy0);
 871             fjz0             = _fjsp_madd_v2r8(dz10,fscal,fjz0);
 872
 873             /**************************
 874              * CALCULATE INTERACTIONS *
 875              **************************/
 876
 877             r20              = _fjsp_mul_v2r8(rsq20,rinv20);
 878
 879             /* Compute parameters for interactions between i and j atoms */
 880             qq20             = _fjsp_mul_v2r8(iq2,jq0);
 881
 882             /* EWALD ELECTROSTATICS */
 883
 884             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 885             ewrt             = _fjsp_mul_v2r8(r20,ewtabscale);
 886             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 887             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 888             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 889
 890             gmx_fjsp_load_2pair_swizzle_v2r8(ewtab+ewconv.i[0],ewtab+ewconv.i[1],
 891                                          &ewtabF,&ewtabFn);
 892             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 893             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq20,rinv20),_fjsp_sub_v2r8(rinvsq20,felec));
 894
 895             fscal            = felec;
 896
 897             /* Update vectorial force */
 898             fix2             = _fjsp_madd_v2r8(dx20,fscal,fix2);
 899             fiy2             = _fjsp_madd_v2r8(dy20,fscal,fiy2);
 900             fiz2             = _fjsp_madd_v2r8(dz20,fscal,fiz2);
 901
 902             fjx0             = _fjsp_madd_v2r8(dx20,fscal,fjx0);
 903             fjy0             = _fjsp_madd_v2r8(dy20,fscal,fjy0);
 904             fjz0             = _fjsp_madd_v2r8(dz20,fscal,fjz0);
 905
 906             gmx_fjsp_decrement_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 907
 908             /* Inner loop uses 142 flops */
 909         }
 910
 911         if(jidx<j_index_end)
 912         {
 913
 914             jnrA             = jjnr[jidx];
 915             j_coord_offsetA  = DIM*jnrA;
 916
 917             /* load j atom coordinates */
 918             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 919                                               &jx0,&jy0,&jz0);
 920
 921             /* Calculate displacement vector */
 922             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 923             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 924             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 925             dx10             = _fjsp_sub_v2r8(ix1,jx0);
 926             dy10             = _fjsp_sub_v2r8(iy1,jy0);
 927             dz10             = _fjsp_sub_v2r8(iz1,jz0);
 928             dx20             = _fjsp_sub_v2r8(ix2,jx0);
 929             dy20             = _fjsp_sub_v2r8(iy2,jy0);
 930             dz20             = _fjsp_sub_v2r8(iz2,jz0);
 931
 932             /* Calculate squared distance and things based on it */
 933             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 934             rsq10            = gmx_fjsp_calc_rsq_v2r8(dx10,dy10,dz10);
 935             rsq20            = gmx_fjsp_calc_rsq_v2r8(dx20,dy20,dz20);
 936
 937             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 938             rinv10           = gmx_fjsp_invsqrt_v2r8(rsq10);
 939             rinv20           = gmx_fjsp_invsqrt_v2r8(rsq20);
 940
 941             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 942             rinvsq10         = _fjsp_mul_v2r8(rinv10,rinv10);
 943             rinvsq20         = _fjsp_mul_v2r8(rinv20,rinv20);
 944
 945             /* Load parameters for j particles */
 946             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 947             vdwjidx0A        = 2*vdwtype[jnrA+0];
 948
 949             fjx0             = _fjsp_setzero_v2r8();
 950             fjy0             = _fjsp_setzero_v2r8();
 951             fjz0             = _fjsp_setzero_v2r8();
 952
 953             /**************************
 954              * CALCULATE INTERACTIONS *
 955              **************************/
 956
 957             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 958
 959             /* Compute parameters for interactions between i and j atoms */
 960             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 961             gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 962
 963             c6grid_00       = gmx_fjsp_load_1real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A);
 964
 965             /* EWALD ELECTROSTATICS */
 966
 967             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 968             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 969             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 970             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 971             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 972
 973             gmx_fjsp_load_1pair_swizzle_v2r8(ewtab+ewconv.i[0],&ewtabF,&ewtabFn);
 974             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 975             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 976
 977             /* Analytical LJ-PME */
 978             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 979             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 980             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 981             exponent         = gmx_simd_exp_d(-ewcljrsq);
 982             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 983             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 984             /* f6A = 6 * C6grid * (1 - poly) */
 985             f6A              = _fjsp_mul_v2r8(c6grid_00,_fjsp_msub_v2r8(one,poly));
 986             /* f6B = C6grid * exponent * beta^6 */
 987             f6B              = _fjsp_mul_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6));
 988             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 989             fvdw              = _fjsp_mul_v2r8(_fjsp_madd_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,_fjsp_sub_v2r8(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 990
 991             fscal            = _fjsp_add_v2r8(felec,fvdw);
 992
 993             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 994
 995             /* Update vectorial force */
 996             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 997             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 998             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 999
1000             fjx0             = _fjsp_madd_v2r8(dx00,fscal,fjx0);
1001             fjy0             = _fjsp_madd_v2r8(dy00,fscal,fjy0);
1002             fjz0             = _fjsp_madd_v2r8(dz00,fscal,fjz0);
1003
1004             /**************************
1005              * CALCULATE INTERACTIONS *
1006              **************************/
1007
1008             r10              = _fjsp_mul_v2r8(rsq10,rinv10);
1009
1010             /* Compute parameters for interactions between i and j atoms */
1011             qq10             = _fjsp_mul_v2r8(iq1,jq0);
1012
1013             /* EWALD ELECTROSTATICS */
1014
1015             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1016             ewrt             = _fjsp_mul_v2r8(r10,ewtabscale);
1017             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
1018             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
1019             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
1020
1021             gmx_fjsp_load_1pair_swizzle_v2r8(ewtab+ewconv.i[0],&ewtabF,&ewtabFn);
1022             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
1023             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq10,rinv10),_fjsp_sub_v2r8(rinvsq10,felec));
1024
1025             fscal            = felec;
1026
1027             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
1028
1029             /* Update vectorial force */
1030             fix1             = _fjsp_madd_v2r8(dx10,fscal,fix1);
1031             fiy1             = _fjsp_madd_v2r8(dy10,fscal,fiy1);
1032             fiz1             = _fjsp_madd_v2r8(dz10,fscal,fiz1);
1033
1034             fjx0             = _fjsp_madd_v2r8(dx10,fscal,fjx0);
1035             fjy0             = _fjsp_madd_v2r8(dy10,fscal,fjy0);
1036             fjz0             = _fjsp_madd_v2r8(dz10,fscal,fjz0);
1037
1038             /**************************
1039              * CALCULATE INTERACTIONS *
1040              **************************/
1041
1042             r20              = _fjsp_mul_v2r8(rsq20,rinv20);
1043
1044             /* Compute parameters for interactions between i and j atoms */
1045             qq20             = _fjsp_mul_v2r8(iq2,jq0);
1046
1047             /* EWALD ELECTROSTATICS */
1048
1049             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1050             ewrt             = _fjsp_mul_v2r8(r20,ewtabscale);
1051             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
1052             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
1053             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
1054
1055             gmx_fjsp_load_1pair_swizzle_v2r8(ewtab+ewconv.i[0],&ewtabF,&ewtabFn);
1056             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
1057             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq20,rinv20),_fjsp_sub_v2r8(rinvsq20,felec));
1058
1059             fscal            = felec;
1060
1061             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
1062
1063             /* Update vectorial force */
1064             fix2             = _fjsp_madd_v2r8(dx20,fscal,fix2);
1065             fiy2             = _fjsp_madd_v2r8(dy20,fscal,fiy2);
1066             fiz2             = _fjsp_madd_v2r8(dz20,fscal,fiz2);
1067
1068             fjx0             = _fjsp_madd_v2r8(dx20,fscal,fjx0);
1069             fjy0             = _fjsp_madd_v2r8(dy20,fscal,fjy0);
1070             fjz0             = _fjsp_madd_v2r8(dz20,fscal,fjz0);
1071
1072             gmx_fjsp_decrement_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1073
1074             /* Inner loop uses 142 flops */
1075         }
1076
1077         /* End of innermost loop */
1078
1079         gmx_fjsp_update_iforce_3atom_swizzle_v2r8(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1080                                               f+i_coord_offset,fshift+i_shift_offset);
1081
1082         /* Increment number of inner iterations */
1083         inneriter                  += j_index_end - j_index_start;
1084
1085         /* Outer loop uses 18 flops */
1086     }
1087
1088     /* Increment number of outer iterations */
1089     outeriter        += nri;
1090
1091     /* Update outer/inner flops */
1092
1093     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*142);
1094 }