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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "kernelutil_sparc64_hpc_ace_double.h"
  50
  51 /*
  52  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_VF_sparc64_hpc_ace_double
  53  * Electrostatics interaction: Ewald
  54  * VdW interaction:            LJEwald
  55  * Geometry:                   Water3-Particle
  56  * Calculate force/pot:        PotentialAndForce
  57  */
  58 void
  59 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_VF_sparc64_hpc_ace_double
  60                     (t_nblist                    * gmx_restrict       nlist,
  61                      rvec                        * gmx_restrict          xx,
  62                      rvec                        * gmx_restrict          ff,
  63                      t_forcerec                  * gmx_restrict          fr,
  64                      t_mdatoms                   * gmx_restrict     mdatoms,
  65                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  66                      t_nrnb                      * gmx_restrict        nrnb)
  67 {
  68     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  69      * just 0 for non-waters.
  70      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
  71      * jnr indices corresponding to data put in the four positions in the SIMD register.
  72      */
  73     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  74     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  75     int              jnrA,jnrB;
  76     int              j_coord_offsetA,j_coord_offsetB;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  81     int              vdwioffset0;
  82     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  83     int              vdwioffset1;
  84     _fjsp_v2r8       ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  85     int              vdwioffset2;
  86     _fjsp_v2r8       ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  87     int              vdwjidx0A,vdwjidx0B;
  88     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     _fjsp_v2r8       dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  91     _fjsp_v2r8       dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  92     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
  93     real             *charge;
  94     int              nvdwtype;
  95     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  96     int              *vdwtype;
  97     real             *vdwparam;
  98     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
  99     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 100     _fjsp_v2r8           c6grid_00;
 101     _fjsp_v2r8           c6grid_10;
 102     _fjsp_v2r8           c6grid_20;
 103     real                 *vdwgridparam;
 104     _fjsp_v2r8           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 105     _fjsp_v2r8           one_half = gmx_fjsp_set1_v2r8(0.5);
 106     _fjsp_v2r8           minus_one = gmx_fjsp_set1_v2r8(-1.0);
 107     _fjsp_v2r8       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 108     real             *ewtab;
 109     _fjsp_v2r8       itab_tmp;
 110     _fjsp_v2r8       dummy_mask,cutoff_mask;
 111     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 112     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 113     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 114
 115     x                = xx[0];
 116     f                = ff[0];
 117
 118     nri              = nlist->nri;
 119     iinr             = nlist->iinr;
 120     jindex           = nlist->jindex;
 121     jjnr             = nlist->jjnr;
 122     shiftidx         = nlist->shift;
 123     gid              = nlist->gid;
 124     shiftvec         = fr->shift_vec[0];
 125     fshift           = fr->fshift[0];
 126     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 127     charge           = mdatoms->chargeA;
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131     vdwgridparam     = fr->ljpme_c6grid;
 132     sh_lj_ewald      = gmx_fjsp_set1_v2r8(fr->ic->sh_lj_ewald);
 133     ewclj            = gmx_fjsp_set1_v2r8(fr->ewaldcoeff_lj);
 134     ewclj2           = _fjsp_mul_v2r8(minus_one,_fjsp_mul_v2r8(ewclj,ewclj));
 135
 136     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 137     ewtab            = fr->ic->tabq_coul_FDV0;
 138     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 139     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 140
 141     /* Setup water-specific parameters */
 142     inr              = nlist->iinr[0];
 143     iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+0]));
 144     iq1              = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+1]));
 145     iq2              = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+2]));
 146     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 147
 148     /* Avoid stupid compiler warnings */
 149     jnrA = jnrB = 0;
 150     j_coord_offsetA = 0;
 151     j_coord_offsetB = 0;
 152
 153     outeriter        = 0;
 154     inneriter        = 0;
 155
 156     /* Start outer loop over neighborlists */
 157     for(iidx=0; iidx<nri; iidx++)
 158     {
 159         /* Load shift vector for this list */
 160         i_shift_offset   = DIM*shiftidx[iidx];
 161
 162         /* Load limits for loop over neighbors */
 163         j_index_start    = jindex[iidx];
 164         j_index_end      = jindex[iidx+1];
 165
 166         /* Get outer coordinate index */
 167         inr              = iinr[iidx];
 168         i_coord_offset   = DIM*inr;
 169
 170         /* Load i particle coords and add shift vector */
 171         gmx_fjsp_load_shift_and_3rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,
 172                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 173
 174         fix0             = _fjsp_setzero_v2r8();
 175         fiy0             = _fjsp_setzero_v2r8();
 176         fiz0             = _fjsp_setzero_v2r8();
 177         fix1             = _fjsp_setzero_v2r8();
 178         fiy1             = _fjsp_setzero_v2r8();
 179         fiz1             = _fjsp_setzero_v2r8();
 180         fix2             = _fjsp_setzero_v2r8();
 181         fiy2             = _fjsp_setzero_v2r8();
 182         fiz2             = _fjsp_setzero_v2r8();
 183
 184         /* Reset potential sums */
 185         velecsum         = _fjsp_setzero_v2r8();
 186         vvdwsum          = _fjsp_setzero_v2r8();
 187
 188         /* Start inner kernel loop */
 189         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 190         {
 191
 192             /* Get j neighbor index, and coordinate index */
 193             jnrA             = jjnr[jidx];
 194             jnrB             = jjnr[jidx+1];
 195             j_coord_offsetA  = DIM*jnrA;
 196             j_coord_offsetB  = DIM*jnrB;
 197
 198             /* load j atom coordinates */
 199             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 200                                               &jx0,&jy0,&jz0);
 201
 202             /* Calculate displacement vector */
 203             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 204             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 205             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 206             dx10             = _fjsp_sub_v2r8(ix1,jx0);
 207             dy10             = _fjsp_sub_v2r8(iy1,jy0);
 208             dz10             = _fjsp_sub_v2r8(iz1,jz0);
 209             dx20             = _fjsp_sub_v2r8(ix2,jx0);
 210             dy20             = _fjsp_sub_v2r8(iy2,jy0);
 211             dz20             = _fjsp_sub_v2r8(iz2,jz0);
 212
 213             /* Calculate squared distance and things based on it */
 214             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 215             rsq10            = gmx_fjsp_calc_rsq_v2r8(dx10,dy10,dz10);
 216             rsq20            = gmx_fjsp_calc_rsq_v2r8(dx20,dy20,dz20);
 217
 218             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 219             rinv10           = gmx_fjsp_invsqrt_v2r8(rsq10);
 220             rinv20           = gmx_fjsp_invsqrt_v2r8(rsq20);
 221
 222             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 223             rinvsq10         = _fjsp_mul_v2r8(rinv10,rinv10);
 224             rinvsq20         = _fjsp_mul_v2r8(rinv20,rinv20);
 225
 226             /* Load parameters for j particles */
 227             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 228             vdwjidx0A        = 2*vdwtype[jnrA+0];
 229             vdwjidx0B        = 2*vdwtype[jnrB+0];
 230
 231             fjx0             = _fjsp_setzero_v2r8();
 232             fjy0             = _fjsp_setzero_v2r8();
 233             fjz0             = _fjsp_setzero_v2r8();
 234
 235             /**************************
 236              * CALCULATE INTERACTIONS *
 237              **************************/
 238
 239             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 240
 241             /* Compute parameters for interactions between i and j atoms */
 242             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 243             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 244                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 245
 246             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 247                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 248
 249             /* EWALD ELECTROSTATICS */
 250
 251             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 252             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 253             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 254             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 255             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 256
 257             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 258             ewtabD           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
 259             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 260             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 261             ewtabFn          = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
 262             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 263             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 264             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 265             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 266             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 267
 268             /* Analytical LJ-PME */
 269             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 270             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 271             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 272             exponent         = gmx_simd_exp_d(ewcljrsq);
 273             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 274             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 275             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 276             vvdw6            = _fjsp_mul_v2r8(_fjsp_madd_v2r8(c6grid_00,_fjsp_sub_v2r8(poly,one),c6_00),rinvsix);
 277             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 278             vvdw             = _fjsp_msub_v2r8(vvdw12,one_twelfth,_fjsp_mul_v2r8(vvdw6,one_sixth));
 279             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 280             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);
 281
 282             /* Update potential sum for this i atom from the interaction with this j atom. */
 283             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 284             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 285
 286             fscal            = _fjsp_add_v2r8(felec,fvdw);
 287
 288             /* Update vectorial force */
 289             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 290             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 291             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 292
 293             fjx0             = _fjsp_madd_v2r8(dx00,fscal,fjx0);
 294             fjy0             = _fjsp_madd_v2r8(dy00,fscal,fjy0);
 295             fjz0             = _fjsp_madd_v2r8(dz00,fscal,fjz0);
 296
 297             /**************************
 298              * CALCULATE INTERACTIONS *
 299              **************************/
 300
 301             r10              = _fjsp_mul_v2r8(rsq10,rinv10);
 302
 303             /* Compute parameters for interactions between i and j atoms */
 304             qq10             = _fjsp_mul_v2r8(iq1,jq0);
 305
 306             /* EWALD ELECTROSTATICS */
 307
 308             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 309             ewrt             = _fjsp_mul_v2r8(r10,ewtabscale);
 310             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 311             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 312             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 313
 314             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 315             ewtabD           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
 316             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 317             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 318             ewtabFn          = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
 319             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 320             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 321             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 322             velec            = _fjsp_mul_v2r8(qq10,_fjsp_sub_v2r8(rinv10,velec));
 323             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq10,rinv10),_fjsp_sub_v2r8(rinvsq10,felec));
 324
 325             /* Update potential sum for this i atom from the interaction with this j atom. */
 326             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 327
 328             fscal            = felec;
 329
 330             /* Update vectorial force */
 331             fix1             = _fjsp_madd_v2r8(dx10,fscal,fix1);
 332             fiy1             = _fjsp_madd_v2r8(dy10,fscal,fiy1);
 333             fiz1             = _fjsp_madd_v2r8(dz10,fscal,fiz1);
 334
 335             fjx0             = _fjsp_madd_v2r8(dx10,fscal,fjx0);
 336             fjy0             = _fjsp_madd_v2r8(dy10,fscal,fjy0);
 337             fjz0             = _fjsp_madd_v2r8(dz10,fscal,fjz0);
 338
 339             /**************************
 340              * CALCULATE INTERACTIONS *
 341              **************************/
 342
 343             r20              = _fjsp_mul_v2r8(rsq20,rinv20);
 344
 345             /* Compute parameters for interactions between i and j atoms */
 346             qq20             = _fjsp_mul_v2r8(iq2,jq0);
 347
 348             /* EWALD ELECTROSTATICS */
 349
 350             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 351             ewrt             = _fjsp_mul_v2r8(r20,ewtabscale);
 352             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 353             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 354             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 355
 356             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 357             ewtabD           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
 358             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 359             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 360             ewtabFn          = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
 361             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 362             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 363             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 364             velec            = _fjsp_mul_v2r8(qq20,_fjsp_sub_v2r8(rinv20,velec));
 365             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq20,rinv20),_fjsp_sub_v2r8(rinvsq20,felec));
 366
 367             /* Update potential sum for this i atom from the interaction with this j atom. */
 368             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 369
 370             fscal            = felec;
 371
 372             /* Update vectorial force */
 373             fix2             = _fjsp_madd_v2r8(dx20,fscal,fix2);
 374             fiy2             = _fjsp_madd_v2r8(dy20,fscal,fiy2);
 375             fiz2             = _fjsp_madd_v2r8(dz20,fscal,fiz2);
 376
 377             fjx0             = _fjsp_madd_v2r8(dx20,fscal,fjx0);
 378             fjy0             = _fjsp_madd_v2r8(dy20,fscal,fjy0);
 379             fjz0             = _fjsp_madd_v2r8(dz20,fscal,fjz0);
 380
 381             gmx_fjsp_decrement_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 382
 383             /* Inner loop uses 159 flops */
 384         }
 385
 386         if(jidx<j_index_end)
 387         {
 388
 389             jnrA             = jjnr[jidx];
 390             j_coord_offsetA  = DIM*jnrA;
 391
 392             /* load j atom coordinates */
 393             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 394                                               &jx0,&jy0,&jz0);
 395
 396             /* Calculate displacement vector */
 397             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 398             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 399             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 400             dx10             = _fjsp_sub_v2r8(ix1,jx0);
 401             dy10             = _fjsp_sub_v2r8(iy1,jy0);
 402             dz10             = _fjsp_sub_v2r8(iz1,jz0);
 403             dx20             = _fjsp_sub_v2r8(ix2,jx0);
 404             dy20             = _fjsp_sub_v2r8(iy2,jy0);
 405             dz20             = _fjsp_sub_v2r8(iz2,jz0);
 406
 407             /* Calculate squared distance and things based on it */
 408             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 409             rsq10            = gmx_fjsp_calc_rsq_v2r8(dx10,dy10,dz10);
 410             rsq20            = gmx_fjsp_calc_rsq_v2r8(dx20,dy20,dz20);
 411
 412             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 413             rinv10           = gmx_fjsp_invsqrt_v2r8(rsq10);
 414             rinv20           = gmx_fjsp_invsqrt_v2r8(rsq20);
 415
 416             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 417             rinvsq10         = _fjsp_mul_v2r8(rinv10,rinv10);
 418             rinvsq20         = _fjsp_mul_v2r8(rinv20,rinv20);
 419
 420             /* Load parameters for j particles */
 421             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 422             vdwjidx0A        = 2*vdwtype[jnrA+0];
 423
 424             fjx0             = _fjsp_setzero_v2r8();
 425             fjy0             = _fjsp_setzero_v2r8();
 426             fjz0             = _fjsp_setzero_v2r8();
 427
 428             /**************************
 429              * CALCULATE INTERACTIONS *
 430              **************************/
 431
 432             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 433
 434             /* Compute parameters for interactions between i and j atoms */
 435             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 436             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 437                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 438
 439             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 440                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 441
 442             /* EWALD ELECTROSTATICS */
 443
 444             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 445             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 446             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 447             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 448             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 449
 450             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 451             ewtabD           = _fjsp_setzero_v2r8();
 452             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 453             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 454             ewtabFn          = _fjsp_setzero_v2r8();
 455             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 456             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 457             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 458             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 459             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 460
 461             /* Analytical LJ-PME */
 462             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 463             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 464             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 465             exponent         = gmx_simd_exp_d(ewcljrsq);
 466             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 467             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 468             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 469             vvdw6            = _fjsp_mul_v2r8(_fjsp_madd_v2r8(c6grid_00,_fjsp_sub_v2r8(poly,one),c6_00),rinvsix);
 470             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 471             vvdw             = _fjsp_msub_v2r8(vvdw12,one_twelfth,_fjsp_mul_v2r8(vvdw6,one_sixth));
 472             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 473             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);
 474
 475             /* Update potential sum for this i atom from the interaction with this j atom. */
 476             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 477             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 478             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 479             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 480
 481             fscal            = _fjsp_add_v2r8(felec,fvdw);
 482
 483             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 484
 485             /* Update vectorial force */
 486             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 487             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 488             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 489
 490             fjx0             = _fjsp_madd_v2r8(dx00,fscal,fjx0);
 491             fjy0             = _fjsp_madd_v2r8(dy00,fscal,fjy0);
 492             fjz0             = _fjsp_madd_v2r8(dz00,fscal,fjz0);
 493
 494             /**************************
 495              * CALCULATE INTERACTIONS *
 496              **************************/
 497
 498             r10              = _fjsp_mul_v2r8(rsq10,rinv10);
 499
 500             /* Compute parameters for interactions between i and j atoms */
 501             qq10             = _fjsp_mul_v2r8(iq1,jq0);
 502
 503             /* EWALD ELECTROSTATICS */
 504
 505             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 506             ewrt             = _fjsp_mul_v2r8(r10,ewtabscale);
 507             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 508             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 509             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 510
 511             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 512             ewtabD           = _fjsp_setzero_v2r8();
 513             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 514             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 515             ewtabFn          = _fjsp_setzero_v2r8();
 516             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 517             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 518             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 519             velec            = _fjsp_mul_v2r8(qq10,_fjsp_sub_v2r8(rinv10,velec));
 520             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq10,rinv10),_fjsp_sub_v2r8(rinvsq10,felec));
 521
 522             /* Update potential sum for this i atom from the interaction with this j atom. */
 523             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 524             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 525
 526             fscal            = felec;
 527
 528             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 529
 530             /* Update vectorial force */
 531             fix1             = _fjsp_madd_v2r8(dx10,fscal,fix1);
 532             fiy1             = _fjsp_madd_v2r8(dy10,fscal,fiy1);
 533             fiz1             = _fjsp_madd_v2r8(dz10,fscal,fiz1);
 534
 535             fjx0             = _fjsp_madd_v2r8(dx10,fscal,fjx0);
 536             fjy0             = _fjsp_madd_v2r8(dy10,fscal,fjy0);
 537             fjz0             = _fjsp_madd_v2r8(dz10,fscal,fjz0);
 538
 539             /**************************
 540              * CALCULATE INTERACTIONS *
 541              **************************/
 542
 543             r20              = _fjsp_mul_v2r8(rsq20,rinv20);
 544
 545             /* Compute parameters for interactions between i and j atoms */
 546             qq20             = _fjsp_mul_v2r8(iq2,jq0);
 547
 548             /* EWALD ELECTROSTATICS */
 549
 550             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 551             ewrt             = _fjsp_mul_v2r8(r20,ewtabscale);
 552             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 553             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 554             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 555
 556             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 557             ewtabD           = _fjsp_setzero_v2r8();
 558             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 559             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 560             ewtabFn          = _fjsp_setzero_v2r8();
 561             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 562             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 563             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 564             velec            = _fjsp_mul_v2r8(qq20,_fjsp_sub_v2r8(rinv20,velec));
 565             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq20,rinv20),_fjsp_sub_v2r8(rinvsq20,felec));
 566
 567             /* Update potential sum for this i atom from the interaction with this j atom. */
 568             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 569             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 570
 571             fscal            = felec;
 572
 573             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 574
 575             /* Update vectorial force */
 576             fix2             = _fjsp_madd_v2r8(dx20,fscal,fix2);
 577             fiy2             = _fjsp_madd_v2r8(dy20,fscal,fiy2);
 578             fiz2             = _fjsp_madd_v2r8(dz20,fscal,fiz2);
 579
 580             fjx0             = _fjsp_madd_v2r8(dx20,fscal,fjx0);
 581             fjy0             = _fjsp_madd_v2r8(dy20,fscal,fjy0);
 582             fjz0             = _fjsp_madd_v2r8(dz20,fscal,fjz0);
 583
 584             gmx_fjsp_decrement_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 585
 586             /* Inner loop uses 159 flops */
 587         }
 588
 589         /* End of innermost loop */
 590
 591         gmx_fjsp_update_iforce_3atom_swizzle_v2r8(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 592                                               f+i_coord_offset,fshift+i_shift_offset);
 593
 594         ggid                        = gid[iidx];
 595         /* Update potential energies */
 596         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 597         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 598
 599         /* Increment number of inner iterations */
 600         inneriter                  += j_index_end - j_index_start;
 601
 602         /* Outer loop uses 20 flops */
 603     }
 604
 605     /* Increment number of outer iterations */
 606     outeriter        += nri;
 607
 608     /* Update outer/inner flops */
 609
 610     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*159);
 611 }
 612 /*
 613  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_sparc64_hpc_ace_double
 614  * Electrostatics interaction: Ewald
 615  * VdW interaction:            LJEwald
 616  * Geometry:                   Water3-Particle
 617  * Calculate force/pot:        Force
 618  */
 619 void
 620 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_sparc64_hpc_ace_double
 621                     (t_nblist                    * gmx_restrict       nlist,
 622                      rvec                        * gmx_restrict          xx,
 623                      rvec                        * gmx_restrict          ff,
 624                      t_forcerec                  * gmx_restrict          fr,
 625                      t_mdatoms                   * gmx_restrict     mdatoms,
 626                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 627                      t_nrnb                      * gmx_restrict        nrnb)
 628 {
 629     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 630      * just 0 for non-waters.
 631      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 632      * jnr indices corresponding to data put in the four positions in the SIMD register.
 633      */
 634     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 635     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 636     int              jnrA,jnrB;
 637     int              j_coord_offsetA,j_coord_offsetB;
 638     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 639     real             rcutoff_scalar;
 640     real             *shiftvec,*fshift,*x,*f;
 641     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 642     int              vdwioffset0;
 643     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 644     int              vdwioffset1;
 645     _fjsp_v2r8       ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 646     int              vdwioffset2;
 647     _fjsp_v2r8       ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 648     int              vdwjidx0A,vdwjidx0B;
 649     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 650     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 651     _fjsp_v2r8       dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 652     _fjsp_v2r8       dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 653     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 654     real             *charge;
 655     int              nvdwtype;
 656     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 657     int              *vdwtype;
 658     real             *vdwparam;
 659     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 660     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 661     _fjsp_v2r8           c6grid_00;
 662     _fjsp_v2r8           c6grid_10;
 663     _fjsp_v2r8           c6grid_20;
 664     real                 *vdwgridparam;
 665     _fjsp_v2r8           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 666     _fjsp_v2r8           one_half = gmx_fjsp_set1_v2r8(0.5);
 667     _fjsp_v2r8           minus_one = gmx_fjsp_set1_v2r8(-1.0);
 668     _fjsp_v2r8       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 669     real             *ewtab;
 670     _fjsp_v2r8       itab_tmp;
 671     _fjsp_v2r8       dummy_mask,cutoff_mask;
 672     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 673     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 674     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 675
 676     x                = xx[0];
 677     f                = ff[0];
 678
 679     nri              = nlist->nri;
 680     iinr             = nlist->iinr;
 681     jindex           = nlist->jindex;
 682     jjnr             = nlist->jjnr;
 683     shiftidx         = nlist->shift;
 684     gid              = nlist->gid;
 685     shiftvec         = fr->shift_vec[0];
 686     fshift           = fr->fshift[0];
 687     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 688     charge           = mdatoms->chargeA;
 689     nvdwtype         = fr->ntype;
 690     vdwparam         = fr->nbfp;
 691     vdwtype          = mdatoms->typeA;
 692     vdwgridparam     = fr->ljpme_c6grid;
 693     sh_lj_ewald      = gmx_fjsp_set1_v2r8(fr->ic->sh_lj_ewald);
 694     ewclj            = gmx_fjsp_set1_v2r8(fr->ewaldcoeff_lj);
 695     ewclj2           = _fjsp_mul_v2r8(minus_one,_fjsp_mul_v2r8(ewclj,ewclj));
 696
 697     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 698     ewtab            = fr->ic->tabq_coul_F;
 699     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 700     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 701
 702     /* Setup water-specific parameters */
 703     inr              = nlist->iinr[0];
 704     iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+0]));
 705     iq1              = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+1]));
 706     iq2              = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+2]));
 707     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 708
 709     /* Avoid stupid compiler warnings */
 710     jnrA = jnrB = 0;
 711     j_coord_offsetA = 0;
 712     j_coord_offsetB = 0;
 713
 714     outeriter        = 0;
 715     inneriter        = 0;
 716
 717     /* Start outer loop over neighborlists */
 718     for(iidx=0; iidx<nri; iidx++)
 719     {
 720         /* Load shift vector for this list */
 721         i_shift_offset   = DIM*shiftidx[iidx];
 722
 723         /* Load limits for loop over neighbors */
 724         j_index_start    = jindex[iidx];
 725         j_index_end      = jindex[iidx+1];
 726
 727         /* Get outer coordinate index */
 728         inr              = iinr[iidx];
 729         i_coord_offset   = DIM*inr;
 730
 731         /* Load i particle coords and add shift vector */
 732         gmx_fjsp_load_shift_and_3rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,
 733                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 734
 735         fix0             = _fjsp_setzero_v2r8();
 736         fiy0             = _fjsp_setzero_v2r8();
 737         fiz0             = _fjsp_setzero_v2r8();
 738         fix1             = _fjsp_setzero_v2r8();
 739         fiy1             = _fjsp_setzero_v2r8();
 740         fiz1             = _fjsp_setzero_v2r8();
 741         fix2             = _fjsp_setzero_v2r8();
 742         fiy2             = _fjsp_setzero_v2r8();
 743         fiz2             = _fjsp_setzero_v2r8();
 744
 745         /* Start inner kernel loop */
 746         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 747         {
 748
 749             /* Get j neighbor index, and coordinate index */
 750             jnrA             = jjnr[jidx];
 751             jnrB             = jjnr[jidx+1];
 752             j_coord_offsetA  = DIM*jnrA;
 753             j_coord_offsetB  = DIM*jnrB;
 754
 755             /* load j atom coordinates */
 756             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 757                                               &jx0,&jy0,&jz0);
 758
 759             /* Calculate displacement vector */
 760             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 761             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 762             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 763             dx10             = _fjsp_sub_v2r8(ix1,jx0);
 764             dy10             = _fjsp_sub_v2r8(iy1,jy0);
 765             dz10             = _fjsp_sub_v2r8(iz1,jz0);
 766             dx20             = _fjsp_sub_v2r8(ix2,jx0);
 767             dy20             = _fjsp_sub_v2r8(iy2,jy0);
 768             dz20             = _fjsp_sub_v2r8(iz2,jz0);
 769
 770             /* Calculate squared distance and things based on it */
 771             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 772             rsq10            = gmx_fjsp_calc_rsq_v2r8(dx10,dy10,dz10);
 773             rsq20            = gmx_fjsp_calc_rsq_v2r8(dx20,dy20,dz20);
 774
 775             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 776             rinv10           = gmx_fjsp_invsqrt_v2r8(rsq10);
 777             rinv20           = gmx_fjsp_invsqrt_v2r8(rsq20);
 778
 779             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 780             rinvsq10         = _fjsp_mul_v2r8(rinv10,rinv10);
 781             rinvsq20         = _fjsp_mul_v2r8(rinv20,rinv20);
 782
 783             /* Load parameters for j particles */
 784             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 785             vdwjidx0A        = 2*vdwtype[jnrA+0];
 786             vdwjidx0B        = 2*vdwtype[jnrB+0];
 787
 788             fjx0             = _fjsp_setzero_v2r8();
 789             fjy0             = _fjsp_setzero_v2r8();
 790             fjz0             = _fjsp_setzero_v2r8();
 791
 792             /**************************
 793              * CALCULATE INTERACTIONS *
 794              **************************/
 795
 796             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 797
 798             /* Compute parameters for interactions between i and j atoms */
 799             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 800             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 801                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 802
 803             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 804                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 805
 806             /* EWALD ELECTROSTATICS */
 807
 808             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 809             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 810             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 811             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 812             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 813
 814             gmx_fjsp_load_2pair_swizzle_v2r8(ewtab+ewconv.i[0],ewtab+ewconv.i[1],
 815                                          &ewtabF,&ewtabFn);
 816             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 817             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 818
 819             /* Analytical LJ-PME */
 820             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 821             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 822             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 823             exponent         = gmx_simd_exp_d(ewcljrsq);
 824             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 825             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 826             /* f6A = 6 * C6grid * (1 - poly) */
 827             f6A              = _fjsp_mul_v2r8(c6grid_00,_fjsp_sub_v2r8(one,poly));
 828             /* f6B = C6grid * exponent * beta^6 */
 829             f6B              = _fjsp_mul_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6));
 830             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 831             fvdw              = _fjsp_mul_v2r8(_fjsp_madd_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,_fjsp_sub_v2r8(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 832
 833             fscal            = _fjsp_add_v2r8(felec,fvdw);
 834
 835             /* Update vectorial force */
 836             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 837             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 838             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 839
 840             fjx0             = _fjsp_madd_v2r8(dx00,fscal,fjx0);
 841             fjy0             = _fjsp_madd_v2r8(dy00,fscal,fjy0);
 842             fjz0             = _fjsp_madd_v2r8(dz00,fscal,fjz0);
 843
 844             /**************************
 845              * CALCULATE INTERACTIONS *
 846              **************************/
 847
 848             r10              = _fjsp_mul_v2r8(rsq10,rinv10);
 849
 850             /* Compute parameters for interactions between i and j atoms */
 851             qq10             = _fjsp_mul_v2r8(iq1,jq0);
 852
 853             /* EWALD ELECTROSTATICS */
 854
 855             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 856             ewrt             = _fjsp_mul_v2r8(r10,ewtabscale);
 857             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 858             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 859             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 860
 861             gmx_fjsp_load_2pair_swizzle_v2r8(ewtab+ewconv.i[0],ewtab+ewconv.i[1],
 862                                          &ewtabF,&ewtabFn);
 863             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 864             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq10,rinv10),_fjsp_sub_v2r8(rinvsq10,felec));
 865
 866             fscal            = felec;
 867
 868             /* Update vectorial force */
 869             fix1             = _fjsp_madd_v2r8(dx10,fscal,fix1);
 870             fiy1             = _fjsp_madd_v2r8(dy10,fscal,fiy1);
 871             fiz1             = _fjsp_madd_v2r8(dz10,fscal,fiz1);
 872
 873             fjx0             = _fjsp_madd_v2r8(dx10,fscal,fjx0);
 874             fjy0             = _fjsp_madd_v2r8(dy10,fscal,fjy0);
 875             fjz0             = _fjsp_madd_v2r8(dz10,fscal,fjz0);
 876
 877             /**************************
 878              * CALCULATE INTERACTIONS *
 879              **************************/
 880
 881             r20              = _fjsp_mul_v2r8(rsq20,rinv20);
 882
 883             /* Compute parameters for interactions between i and j atoms */
 884             qq20             = _fjsp_mul_v2r8(iq2,jq0);
 885
 886             /* EWALD ELECTROSTATICS */
 887
 888             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 889             ewrt             = _fjsp_mul_v2r8(r20,ewtabscale);
 890             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 891             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 892             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 893
 894             gmx_fjsp_load_2pair_swizzle_v2r8(ewtab+ewconv.i[0],ewtab+ewconv.i[1],
 895                                          &ewtabF,&ewtabFn);
 896             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 897             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq20,rinv20),_fjsp_sub_v2r8(rinvsq20,felec));
 898
 899             fscal            = felec;
 900
 901             /* Update vectorial force */
 902             fix2             = _fjsp_madd_v2r8(dx20,fscal,fix2);
 903             fiy2             = _fjsp_madd_v2r8(dy20,fscal,fiy2);
 904             fiz2             = _fjsp_madd_v2r8(dz20,fscal,fiz2);
 905
 906             fjx0             = _fjsp_madd_v2r8(dx20,fscal,fjx0);
 907             fjy0             = _fjsp_madd_v2r8(dy20,fscal,fjy0);
 908             fjz0             = _fjsp_madd_v2r8(dz20,fscal,fjz0);
 909
 910             gmx_fjsp_decrement_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 911
 912             /* Inner loop uses 142 flops */
 913         }
 914
 915         if(jidx<j_index_end)
 916         {
 917
 918             jnrA             = jjnr[jidx];
 919             j_coord_offsetA  = DIM*jnrA;
 920
 921             /* load j atom coordinates */
 922             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 923                                               &jx0,&jy0,&jz0);
 924
 925             /* Calculate displacement vector */
 926             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 927             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 928             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 929             dx10             = _fjsp_sub_v2r8(ix1,jx0);
 930             dy10             = _fjsp_sub_v2r8(iy1,jy0);
 931             dz10             = _fjsp_sub_v2r8(iz1,jz0);
 932             dx20             = _fjsp_sub_v2r8(ix2,jx0);
 933             dy20             = _fjsp_sub_v2r8(iy2,jy0);
 934             dz20             = _fjsp_sub_v2r8(iz2,jz0);
 935
 936             /* Calculate squared distance and things based on it */
 937             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 938             rsq10            = gmx_fjsp_calc_rsq_v2r8(dx10,dy10,dz10);
 939             rsq20            = gmx_fjsp_calc_rsq_v2r8(dx20,dy20,dz20);
 940
 941             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 942             rinv10           = gmx_fjsp_invsqrt_v2r8(rsq10);
 943             rinv20           = gmx_fjsp_invsqrt_v2r8(rsq20);
 944
 945             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 946             rinvsq10         = _fjsp_mul_v2r8(rinv10,rinv10);
 947             rinvsq20         = _fjsp_mul_v2r8(rinv20,rinv20);
 948
 949             /* Load parameters for j particles */
 950             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 951             vdwjidx0A        = 2*vdwtype[jnrA+0];
 952
 953             fjx0             = _fjsp_setzero_v2r8();
 954             fjy0             = _fjsp_setzero_v2r8();
 955             fjz0             = _fjsp_setzero_v2r8();
 956
 957             /**************************
 958              * CALCULATE INTERACTIONS *
 959              **************************/
 960
 961             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 962
 963             /* Compute parameters for interactions between i and j atoms */
 964             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 965             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 966                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 967
 968             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 969                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 970
 971             /* EWALD ELECTROSTATICS */
 972
 973             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 974             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 975             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 976             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 977             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 978
 979             gmx_fjsp_load_1pair_swizzle_v2r8(ewtab+ewconv.i[0],&ewtabF,&ewtabFn);
 980             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 981             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 982
 983             /* Analytical LJ-PME */
 984             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 985             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 986             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 987             exponent         = gmx_simd_exp_d(ewcljrsq);
 988             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 989             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 990             /* f6A = 6 * C6grid * (1 - poly) */
 991             f6A              = _fjsp_mul_v2r8(c6grid_00,_fjsp_sub_v2r8(one,poly));
 992             /* f6B = C6grid * exponent * beta^6 */
 993             f6B              = _fjsp_mul_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6));
 994             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 995             fvdw              = _fjsp_mul_v2r8(_fjsp_madd_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,_fjsp_sub_v2r8(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 996
 997             fscal            = _fjsp_add_v2r8(felec,fvdw);
 998
 999             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
1000
1001             /* Update vectorial force */
1002             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
1003             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
1004             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
1005
1006             fjx0             = _fjsp_madd_v2r8(dx00,fscal,fjx0);
1007             fjy0             = _fjsp_madd_v2r8(dy00,fscal,fjy0);
1008             fjz0             = _fjsp_madd_v2r8(dz00,fscal,fjz0);
1009
1010             /**************************
1011              * CALCULATE INTERACTIONS *
1012              **************************/
1013
1014             r10              = _fjsp_mul_v2r8(rsq10,rinv10);
1015
1016             /* Compute parameters for interactions between i and j atoms */
1017             qq10             = _fjsp_mul_v2r8(iq1,jq0);
1018
1019             /* EWALD ELECTROSTATICS */
1020
1021             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1022             ewrt             = _fjsp_mul_v2r8(r10,ewtabscale);
1023             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
1024             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
1025             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
1026
1027             gmx_fjsp_load_1pair_swizzle_v2r8(ewtab+ewconv.i[0],&ewtabF,&ewtabFn);
1028             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
1029             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq10,rinv10),_fjsp_sub_v2r8(rinvsq10,felec));
1030
1031             fscal            = felec;
1032
1033             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
1034
1035             /* Update vectorial force */
1036             fix1             = _fjsp_madd_v2r8(dx10,fscal,fix1);
1037             fiy1             = _fjsp_madd_v2r8(dy10,fscal,fiy1);
1038             fiz1             = _fjsp_madd_v2r8(dz10,fscal,fiz1);
1039
1040             fjx0             = _fjsp_madd_v2r8(dx10,fscal,fjx0);
1041             fjy0             = _fjsp_madd_v2r8(dy10,fscal,fjy0);
1042             fjz0             = _fjsp_madd_v2r8(dz10,fscal,fjz0);
1043
1044             /**************************
1045              * CALCULATE INTERACTIONS *
1046              **************************/
1047
1048             r20              = _fjsp_mul_v2r8(rsq20,rinv20);
1049
1050             /* Compute parameters for interactions between i and j atoms */
1051             qq20             = _fjsp_mul_v2r8(iq2,jq0);
1052
1053             /* EWALD ELECTROSTATICS */
1054
1055             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1056             ewrt             = _fjsp_mul_v2r8(r20,ewtabscale);
1057             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
1058             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
1059             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
1060
1061             gmx_fjsp_load_1pair_swizzle_v2r8(ewtab+ewconv.i[0],&ewtabF,&ewtabFn);
1062             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
1063             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq20,rinv20),_fjsp_sub_v2r8(rinvsq20,felec));
1064
1065             fscal            = felec;
1066
1067             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
1068
1069             /* Update vectorial force */
1070             fix2             = _fjsp_madd_v2r8(dx20,fscal,fix2);
1071             fiy2             = _fjsp_madd_v2r8(dy20,fscal,fiy2);
1072             fiz2             = _fjsp_madd_v2r8(dz20,fscal,fiz2);
1073
1074             fjx0             = _fjsp_madd_v2r8(dx20,fscal,fjx0);
1075             fjy0             = _fjsp_madd_v2r8(dy20,fscal,fjy0);
1076             fjz0             = _fjsp_madd_v2r8(dz20,fscal,fjz0);
1077
1078             gmx_fjsp_decrement_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1079
1080             /* Inner loop uses 142 flops */
1081         }
1082
1083         /* End of innermost loop */
1084
1085         gmx_fjsp_update_iforce_3atom_swizzle_v2r8(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1086                                               f+i_coord_offset,fshift+i_shift_offset);
1087
1088         /* Increment number of inner iterations */
1089         inneriter                  += j_index_end - j_index_start;
1090
1091         /* Outer loop uses 18 flops */
1092     }
1093
1094     /* Increment number of outer iterations */
1095     outeriter        += nri;
1096
1097     /* Update outer/inner flops */
1098
1099     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*142);
1100 }