src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecEwSh_VdwLJSh_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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "gromacs/legacyheaders/vec.h"
  47 #include "nrnb.h"
  48
  49 #include "kernelutil_sparc64_hpc_ace_double.h"
  50
  51 /*
  52  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_VF_sparc64_hpc_ace_double
  53  * Electrostatics interaction: Ewald
  54  * VdW interaction:            LennardJones
  55  * Geometry:                   Water3-Particle
  56  * Calculate force/pot:        PotentialAndForce
  57  */
  58 void
  59 nb_kernel_ElecEwSh_VdwLJSh_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       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 101     real             *ewtab;
 102     _fjsp_v2r8       itab_tmp;
 103     _fjsp_v2r8       dummy_mask,cutoff_mask;
 104     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 105     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 106     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 107
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 120     charge           = mdatoms->chargeA;
 121     nvdwtype         = fr->ntype;
 122     vdwparam         = fr->nbfp;
 123     vdwtype          = mdatoms->typeA;
 124
 125     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 126     ewtab            = fr->ic->tabq_coul_FDV0;
 127     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 128     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 129
 130     /* Setup water-specific parameters */
 131     inr              = nlist->iinr[0];
 132     iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+0]));
 133     iq1              = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+1]));
 134     iq2              = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+2]));
 135     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 136
 137     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 138     rcutoff_scalar   = fr->rcoulomb;
 139     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 140     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 141
 142     sh_vdw_invrcut6  = gmx_fjsp_set1_v2r8(fr->ic->sh_invrc6);
 143     rvdw             = gmx_fjsp_set1_v2r8(fr->rvdw);
 144
 145     /* Avoid stupid compiler warnings */
 146     jnrA = jnrB = 0;
 147     j_coord_offsetA = 0;
 148     j_coord_offsetB = 0;
 149
 150     outeriter        = 0;
 151     inneriter        = 0;
 152
 153     /* Start outer loop over neighborlists */
 154     for(iidx=0; iidx<nri; iidx++)
 155     {
 156         /* Load shift vector for this list */
 157         i_shift_offset   = DIM*shiftidx[iidx];
 158
 159         /* Load limits for loop over neighbors */
 160         j_index_start    = jindex[iidx];
 161         j_index_end      = jindex[iidx+1];
 162
 163         /* Get outer coordinate index */
 164         inr              = iinr[iidx];
 165         i_coord_offset   = DIM*inr;
 166
 167         /* Load i particle coords and add shift vector */
 168         gmx_fjsp_load_shift_and_3rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,
 169                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 170
 171         fix0             = _fjsp_setzero_v2r8();
 172         fiy0             = _fjsp_setzero_v2r8();
 173         fiz0             = _fjsp_setzero_v2r8();
 174         fix1             = _fjsp_setzero_v2r8();
 175         fiy1             = _fjsp_setzero_v2r8();
 176         fiz1             = _fjsp_setzero_v2r8();
 177         fix2             = _fjsp_setzero_v2r8();
 178         fiy2             = _fjsp_setzero_v2r8();
 179         fiz2             = _fjsp_setzero_v2r8();
 180
 181         /* Reset potential sums */
 182         velecsum         = _fjsp_setzero_v2r8();
 183         vvdwsum          = _fjsp_setzero_v2r8();
 184
 185         /* Start inner kernel loop */
 186         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 187         {
 188
 189             /* Get j neighbor index, and coordinate index */
 190             jnrA             = jjnr[jidx];
 191             jnrB             = jjnr[jidx+1];
 192             j_coord_offsetA  = DIM*jnrA;
 193             j_coord_offsetB  = DIM*jnrB;
 194
 195             /* load j atom coordinates */
 196             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 197                                               &jx0,&jy0,&jz0);
 198
 199             /* Calculate displacement vector */
 200             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 201             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 202             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 203             dx10             = _fjsp_sub_v2r8(ix1,jx0);
 204             dy10             = _fjsp_sub_v2r8(iy1,jy0);
 205             dz10             = _fjsp_sub_v2r8(iz1,jz0);
 206             dx20             = _fjsp_sub_v2r8(ix2,jx0);
 207             dy20             = _fjsp_sub_v2r8(iy2,jy0);
 208             dz20             = _fjsp_sub_v2r8(iz2,jz0);
 209
 210             /* Calculate squared distance and things based on it */
 211             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 212             rsq10            = gmx_fjsp_calc_rsq_v2r8(dx10,dy10,dz10);
 213             rsq20            = gmx_fjsp_calc_rsq_v2r8(dx20,dy20,dz20);
 214
 215             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 216             rinv10           = gmx_fjsp_invsqrt_v2r8(rsq10);
 217             rinv20           = gmx_fjsp_invsqrt_v2r8(rsq20);
 218
 219             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 220             rinvsq10         = _fjsp_mul_v2r8(rinv10,rinv10);
 221             rinvsq20         = _fjsp_mul_v2r8(rinv20,rinv20);
 222
 223             /* Load parameters for j particles */
 224             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 225             vdwjidx0A        = 2*vdwtype[jnrA+0];
 226             vdwjidx0B        = 2*vdwtype[jnrB+0];
 227
 228             fjx0             = _fjsp_setzero_v2r8();
 229             fjy0             = _fjsp_setzero_v2r8();
 230             fjz0             = _fjsp_setzero_v2r8();
 231
 232             /**************************
 233              * CALCULATE INTERACTIONS *
 234              **************************/
 235
 236             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 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             /* EWALD ELECTROSTATICS */
 247
 248             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 249             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 250             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 251             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 252             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 253
 254             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 255             ewtabD           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
 256             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 257             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 258             ewtabFn          = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
 259             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 260             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 261             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 262             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(_fjsp_sub_v2r8(rinv00,sh_ewald),velec));
 263             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 264
 265             /* LENNARD-JONES DISPERSION/REPULSION */
 266
 267             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 268             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 269             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 270             vvdw             = _fjsp_msub_v2r8(_fjsp_nmsub_v2r8(c12_00,_fjsp_mul_v2r8(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 271                                            _fjsp_mul_v2r8(_fjsp_nmsub_v2r8( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 272             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 273
 274             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 275
 276             /* Update potential sum for this i atom from the interaction with this j atom. */
 277             velec            = _fjsp_and_v2r8(velec,cutoff_mask);
 278             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 279             vvdw             = _fjsp_and_v2r8(vvdw,cutoff_mask);
 280             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 281
 282             fscal            = _fjsp_add_v2r8(felec,fvdw);
 283
 284             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 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
 297             /**************************
 298              * CALCULATE INTERACTIONS *
 299              **************************/
 300
 301             if (gmx_fjsp_any_lt_v2r8(rsq10,rcutoff2))
 302             {
 303
 304             r10              = _fjsp_mul_v2r8(rsq10,rinv10);
 305
 306             /* Compute parameters for interactions between i and j atoms */
 307             qq10             = _fjsp_mul_v2r8(iq1,jq0);
 308
 309             /* EWALD ELECTROSTATICS */
 310
 311             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 312             ewrt             = _fjsp_mul_v2r8(r10,ewtabscale);
 313             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 314             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 315             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 316
 317             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 318             ewtabD           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
 319             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 320             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 321             ewtabFn          = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
 322             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 323             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 324             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 325             velec            = _fjsp_mul_v2r8(qq10,_fjsp_sub_v2r8(_fjsp_sub_v2r8(rinv10,sh_ewald),velec));
 326             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq10,rinv10),_fjsp_sub_v2r8(rinvsq10,felec));
 327
 328             cutoff_mask      = _fjsp_cmplt_v2r8(rsq10,rcutoff2);
 329
 330             /* Update potential sum for this i atom from the interaction with this j atom. */
 331             velec            = _fjsp_and_v2r8(velec,cutoff_mask);
 332             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 333
 334             fscal            = felec;
 335
 336             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 337
 338             /* Update vectorial force */
 339             fix1             = _fjsp_madd_v2r8(dx10,fscal,fix1);
 340             fiy1             = _fjsp_madd_v2r8(dy10,fscal,fiy1);
 341             fiz1             = _fjsp_madd_v2r8(dz10,fscal,fiz1);
 342
 343             fjx0             = _fjsp_madd_v2r8(dx10,fscal,fjx0);
 344             fjy0             = _fjsp_madd_v2r8(dy10,fscal,fjy0);
 345             fjz0             = _fjsp_madd_v2r8(dz10,fscal,fjz0);
 346
 347             }
 348
 349             /**************************
 350              * CALCULATE INTERACTIONS *
 351              **************************/
 352
 353             if (gmx_fjsp_any_lt_v2r8(rsq20,rcutoff2))
 354             {
 355
 356             r20              = _fjsp_mul_v2r8(rsq20,rinv20);
 357
 358             /* Compute parameters for interactions between i and j atoms */
 359             qq20             = _fjsp_mul_v2r8(iq2,jq0);
 360
 361             /* EWALD ELECTROSTATICS */
 362
 363             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 364             ewrt             = _fjsp_mul_v2r8(r20,ewtabscale);
 365             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 366             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 367             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 368
 369             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 370             ewtabD           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
 371             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 372             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 373             ewtabFn          = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
 374             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 375             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 376             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 377             velec            = _fjsp_mul_v2r8(qq20,_fjsp_sub_v2r8(_fjsp_sub_v2r8(rinv20,sh_ewald),velec));
 378             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq20,rinv20),_fjsp_sub_v2r8(rinvsq20,felec));
 379
 380             cutoff_mask      = _fjsp_cmplt_v2r8(rsq20,rcutoff2);
 381
 382             /* Update potential sum for this i atom from the interaction with this j atom. */
 383             velec            = _fjsp_and_v2r8(velec,cutoff_mask);
 384             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 385
 386             fscal            = felec;
 387
 388             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 389
 390             /* Update vectorial force */
 391             fix2             = _fjsp_madd_v2r8(dx20,fscal,fix2);
 392             fiy2             = _fjsp_madd_v2r8(dy20,fscal,fiy2);
 393             fiz2             = _fjsp_madd_v2r8(dz20,fscal,fiz2);
 394
 395             fjx0             = _fjsp_madd_v2r8(dx20,fscal,fjx0);
 396             fjy0             = _fjsp_madd_v2r8(dy20,fscal,fjy0);
 397             fjz0             = _fjsp_madd_v2r8(dz20,fscal,fjz0);
 398
 399             }
 400
 401             gmx_fjsp_decrement_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 402
 403             /* Inner loop uses 168 flops */
 404         }
 405
 406         if(jidx<j_index_end)
 407         {
 408
 409             jnrA             = jjnr[jidx];
 410             j_coord_offsetA  = DIM*jnrA;
 411
 412             /* load j atom coordinates */
 413             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 414                                               &jx0,&jy0,&jz0);
 415
 416             /* Calculate displacement vector */
 417             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 418             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 419             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 420             dx10             = _fjsp_sub_v2r8(ix1,jx0);
 421             dy10             = _fjsp_sub_v2r8(iy1,jy0);
 422             dz10             = _fjsp_sub_v2r8(iz1,jz0);
 423             dx20             = _fjsp_sub_v2r8(ix2,jx0);
 424             dy20             = _fjsp_sub_v2r8(iy2,jy0);
 425             dz20             = _fjsp_sub_v2r8(iz2,jz0);
 426
 427             /* Calculate squared distance and things based on it */
 428             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 429             rsq10            = gmx_fjsp_calc_rsq_v2r8(dx10,dy10,dz10);
 430             rsq20            = gmx_fjsp_calc_rsq_v2r8(dx20,dy20,dz20);
 431
 432             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 433             rinv10           = gmx_fjsp_invsqrt_v2r8(rsq10);
 434             rinv20           = gmx_fjsp_invsqrt_v2r8(rsq20);
 435
 436             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 437             rinvsq10         = _fjsp_mul_v2r8(rinv10,rinv10);
 438             rinvsq20         = _fjsp_mul_v2r8(rinv20,rinv20);
 439
 440             /* Load parameters for j particles */
 441             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 442             vdwjidx0A        = 2*vdwtype[jnrA+0];
 443
 444             fjx0             = _fjsp_setzero_v2r8();
 445             fjy0             = _fjsp_setzero_v2r8();
 446             fjz0             = _fjsp_setzero_v2r8();
 447
 448             /**************************
 449              * CALCULATE INTERACTIONS *
 450              **************************/
 451
 452             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 453             {
 454
 455             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 456
 457             /* Compute parameters for interactions between i and j atoms */
 458             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 459             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 460                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 461
 462             /* EWALD ELECTROSTATICS */
 463
 464             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 465             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 466             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 467             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 468             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 469
 470             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 471             ewtabD           = _fjsp_setzero_v2r8();
 472             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 473             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 474             ewtabFn          = _fjsp_setzero_v2r8();
 475             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 476             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 477             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 478             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(_fjsp_sub_v2r8(rinv00,sh_ewald),velec));
 479             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 480
 481             /* LENNARD-JONES DISPERSION/REPULSION */
 482
 483             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 484             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 485             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 486             vvdw             = _fjsp_msub_v2r8(_fjsp_nmsub_v2r8(c12_00,_fjsp_mul_v2r8(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 487                                            _fjsp_mul_v2r8(_fjsp_nmsub_v2r8( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 488             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 489
 490             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 491
 492             /* Update potential sum for this i atom from the interaction with this j atom. */
 493             velec            = _fjsp_and_v2r8(velec,cutoff_mask);
 494             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 495             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 496             vvdw             = _fjsp_and_v2r8(vvdw,cutoff_mask);
 497             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 498             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 499
 500             fscal            = _fjsp_add_v2r8(felec,fvdw);
 501
 502             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 503
 504             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 505
 506             /* Update vectorial force */
 507             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 508             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 509             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 510
 511             fjx0             = _fjsp_madd_v2r8(dx00,fscal,fjx0);
 512             fjy0             = _fjsp_madd_v2r8(dy00,fscal,fjy0);
 513             fjz0             = _fjsp_madd_v2r8(dz00,fscal,fjz0);
 514
 515             }
 516
 517             /**************************
 518              * CALCULATE INTERACTIONS *
 519              **************************/
 520
 521             if (gmx_fjsp_any_lt_v2r8(rsq10,rcutoff2))
 522             {
 523
 524             r10              = _fjsp_mul_v2r8(rsq10,rinv10);
 525
 526             /* Compute parameters for interactions between i and j atoms */
 527             qq10             = _fjsp_mul_v2r8(iq1,jq0);
 528
 529             /* EWALD ELECTROSTATICS */
 530
 531             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 532             ewrt             = _fjsp_mul_v2r8(r10,ewtabscale);
 533             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 534             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 535             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 536
 537             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 538             ewtabD           = _fjsp_setzero_v2r8();
 539             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 540             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 541             ewtabFn          = _fjsp_setzero_v2r8();
 542             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 543             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 544             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 545             velec            = _fjsp_mul_v2r8(qq10,_fjsp_sub_v2r8(_fjsp_sub_v2r8(rinv10,sh_ewald),velec));
 546             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq10,rinv10),_fjsp_sub_v2r8(rinvsq10,felec));
 547
 548             cutoff_mask      = _fjsp_cmplt_v2r8(rsq10,rcutoff2);
 549
 550             /* Update potential sum for this i atom from the interaction with this j atom. */
 551             velec            = _fjsp_and_v2r8(velec,cutoff_mask);
 552             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 553             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 554
 555             fscal            = felec;
 556
 557             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 558
 559             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 560
 561             /* Update vectorial force */
 562             fix1             = _fjsp_madd_v2r8(dx10,fscal,fix1);
 563             fiy1             = _fjsp_madd_v2r8(dy10,fscal,fiy1);
 564             fiz1             = _fjsp_madd_v2r8(dz10,fscal,fiz1);
 565
 566             fjx0             = _fjsp_madd_v2r8(dx10,fscal,fjx0);
 567             fjy0             = _fjsp_madd_v2r8(dy10,fscal,fjy0);
 568             fjz0             = _fjsp_madd_v2r8(dz10,fscal,fjz0);
 569
 570             }
 571
 572             /**************************
 573              * CALCULATE INTERACTIONS *
 574              **************************/
 575
 576             if (gmx_fjsp_any_lt_v2r8(rsq20,rcutoff2))
 577             {
 578
 579             r20              = _fjsp_mul_v2r8(rsq20,rinv20);
 580
 581             /* Compute parameters for interactions between i and j atoms */
 582             qq20             = _fjsp_mul_v2r8(iq2,jq0);
 583
 584             /* EWALD ELECTROSTATICS */
 585
 586             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 587             ewrt             = _fjsp_mul_v2r8(r20,ewtabscale);
 588             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 589             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 590             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 591
 592             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 593             ewtabD           = _fjsp_setzero_v2r8();
 594             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 595             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 596             ewtabFn          = _fjsp_setzero_v2r8();
 597             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 598             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 599             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 600             velec            = _fjsp_mul_v2r8(qq20,_fjsp_sub_v2r8(_fjsp_sub_v2r8(rinv20,sh_ewald),velec));
 601             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq20,rinv20),_fjsp_sub_v2r8(rinvsq20,felec));
 602
 603             cutoff_mask      = _fjsp_cmplt_v2r8(rsq20,rcutoff2);
 604
 605             /* Update potential sum for this i atom from the interaction with this j atom. */
 606             velec            = _fjsp_and_v2r8(velec,cutoff_mask);
 607             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 608             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 609
 610             fscal            = felec;
 611
 612             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 613
 614             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 615
 616             /* Update vectorial force */
 617             fix2             = _fjsp_madd_v2r8(dx20,fscal,fix2);
 618             fiy2             = _fjsp_madd_v2r8(dy20,fscal,fiy2);
 619             fiz2             = _fjsp_madd_v2r8(dz20,fscal,fiz2);
 620
 621             fjx0             = _fjsp_madd_v2r8(dx20,fscal,fjx0);
 622             fjy0             = _fjsp_madd_v2r8(dy20,fscal,fjy0);
 623             fjz0             = _fjsp_madd_v2r8(dz20,fscal,fjz0);
 624
 625             }
 626
 627             gmx_fjsp_decrement_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 628
 629             /* Inner loop uses 168 flops */
 630         }
 631
 632         /* End of innermost loop */
 633
 634         gmx_fjsp_update_iforce_3atom_swizzle_v2r8(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 635                                               f+i_coord_offset,fshift+i_shift_offset);
 636
 637         ggid                        = gid[iidx];
 638         /* Update potential energies */
 639         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 640         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 641
 642         /* Increment number of inner iterations */
 643         inneriter                  += j_index_end - j_index_start;
 644
 645         /* Outer loop uses 20 flops */
 646     }
 647
 648     /* Increment number of outer iterations */
 649     outeriter        += nri;
 650
 651     /* Update outer/inner flops */
 652
 653     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*168);
 654 }
 655 /*
 656  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_sparc64_hpc_ace_double
 657  * Electrostatics interaction: Ewald
 658  * VdW interaction:            LennardJones
 659  * Geometry:                   Water3-Particle
 660  * Calculate force/pot:        Force
 661  */
 662 void
 663 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_sparc64_hpc_ace_double
 664                     (t_nblist                    * gmx_restrict       nlist,
 665                      rvec                        * gmx_restrict          xx,
 666                      rvec                        * gmx_restrict          ff,
 667                      t_forcerec                  * gmx_restrict          fr,
 668                      t_mdatoms                   * gmx_restrict     mdatoms,
 669                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 670                      t_nrnb                      * gmx_restrict        nrnb)
 671 {
 672     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 673      * just 0 for non-waters.
 674      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 675      * jnr indices corresponding to data put in the four positions in the SIMD register.
 676      */
 677     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 678     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 679     int              jnrA,jnrB;
 680     int              j_coord_offsetA,j_coord_offsetB;
 681     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 682     real             rcutoff_scalar;
 683     real             *shiftvec,*fshift,*x,*f;
 684     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 685     int              vdwioffset0;
 686     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 687     int              vdwioffset1;
 688     _fjsp_v2r8       ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 689     int              vdwioffset2;
 690     _fjsp_v2r8       ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 691     int              vdwjidx0A,vdwjidx0B;
 692     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 693     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 694     _fjsp_v2r8       dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 695     _fjsp_v2r8       dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 696     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 697     real             *charge;
 698     int              nvdwtype;
 699     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 700     int              *vdwtype;
 701     real             *vdwparam;
 702     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 703     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 704     _fjsp_v2r8       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 705     real             *ewtab;
 706     _fjsp_v2r8       itab_tmp;
 707     _fjsp_v2r8       dummy_mask,cutoff_mask;
 708     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 709     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 710     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 711
 712     x                = xx[0];
 713     f                = ff[0];
 714
 715     nri              = nlist->nri;
 716     iinr             = nlist->iinr;
 717     jindex           = nlist->jindex;
 718     jjnr             = nlist->jjnr;
 719     shiftidx         = nlist->shift;
 720     gid              = nlist->gid;
 721     shiftvec         = fr->shift_vec[0];
 722     fshift           = fr->fshift[0];
 723     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 724     charge           = mdatoms->chargeA;
 725     nvdwtype         = fr->ntype;
 726     vdwparam         = fr->nbfp;
 727     vdwtype          = mdatoms->typeA;
 728
 729     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 730     ewtab            = fr->ic->tabq_coul_F;
 731     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 732     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 733
 734     /* Setup water-specific parameters */
 735     inr              = nlist->iinr[0];
 736     iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+0]));
 737     iq1              = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+1]));
 738     iq2              = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+2]));
 739     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 740
 741     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 742     rcutoff_scalar   = fr->rcoulomb;
 743     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 744     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 745
 746     sh_vdw_invrcut6  = gmx_fjsp_set1_v2r8(fr->ic->sh_invrc6);
 747     rvdw             = gmx_fjsp_set1_v2r8(fr->rvdw);
 748
 749     /* Avoid stupid compiler warnings */
 750     jnrA = jnrB = 0;
 751     j_coord_offsetA = 0;
 752     j_coord_offsetB = 0;
 753
 754     outeriter        = 0;
 755     inneriter        = 0;
 756
 757     /* Start outer loop over neighborlists */
 758     for(iidx=0; iidx<nri; iidx++)
 759     {
 760         /* Load shift vector for this list */
 761         i_shift_offset   = DIM*shiftidx[iidx];
 762
 763         /* Load limits for loop over neighbors */
 764         j_index_start    = jindex[iidx];
 765         j_index_end      = jindex[iidx+1];
 766
 767         /* Get outer coordinate index */
 768         inr              = iinr[iidx];
 769         i_coord_offset   = DIM*inr;
 770
 771         /* Load i particle coords and add shift vector */
 772         gmx_fjsp_load_shift_and_3rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,
 773                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 774
 775         fix0             = _fjsp_setzero_v2r8();
 776         fiy0             = _fjsp_setzero_v2r8();
 777         fiz0             = _fjsp_setzero_v2r8();
 778         fix1             = _fjsp_setzero_v2r8();
 779         fiy1             = _fjsp_setzero_v2r8();
 780         fiz1             = _fjsp_setzero_v2r8();
 781         fix2             = _fjsp_setzero_v2r8();
 782         fiy2             = _fjsp_setzero_v2r8();
 783         fiz2             = _fjsp_setzero_v2r8();
 784
 785         /* Start inner kernel loop */
 786         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 787         {
 788
 789             /* Get j neighbor index, and coordinate index */
 790             jnrA             = jjnr[jidx];
 791             jnrB             = jjnr[jidx+1];
 792             j_coord_offsetA  = DIM*jnrA;
 793             j_coord_offsetB  = DIM*jnrB;
 794
 795             /* load j atom coordinates */
 796             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 797                                               &jx0,&jy0,&jz0);
 798
 799             /* Calculate displacement vector */
 800             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 801             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 802             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 803             dx10             = _fjsp_sub_v2r8(ix1,jx0);
 804             dy10             = _fjsp_sub_v2r8(iy1,jy0);
 805             dz10             = _fjsp_sub_v2r8(iz1,jz0);
 806             dx20             = _fjsp_sub_v2r8(ix2,jx0);
 807             dy20             = _fjsp_sub_v2r8(iy2,jy0);
 808             dz20             = _fjsp_sub_v2r8(iz2,jz0);
 809
 810             /* Calculate squared distance and things based on it */
 811             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 812             rsq10            = gmx_fjsp_calc_rsq_v2r8(dx10,dy10,dz10);
 813             rsq20            = gmx_fjsp_calc_rsq_v2r8(dx20,dy20,dz20);
 814
 815             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 816             rinv10           = gmx_fjsp_invsqrt_v2r8(rsq10);
 817             rinv20           = gmx_fjsp_invsqrt_v2r8(rsq20);
 818
 819             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 820             rinvsq10         = _fjsp_mul_v2r8(rinv10,rinv10);
 821             rinvsq20         = _fjsp_mul_v2r8(rinv20,rinv20);
 822
 823             /* Load parameters for j particles */
 824             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 825             vdwjidx0A        = 2*vdwtype[jnrA+0];
 826             vdwjidx0B        = 2*vdwtype[jnrB+0];
 827
 828             fjx0             = _fjsp_setzero_v2r8();
 829             fjy0             = _fjsp_setzero_v2r8();
 830             fjz0             = _fjsp_setzero_v2r8();
 831
 832             /**************************
 833              * CALCULATE INTERACTIONS *
 834              **************************/
 835
 836             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 837             {
 838
 839             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 840
 841             /* Compute parameters for interactions between i and j atoms */
 842             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 843             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 844                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 845
 846             /* EWALD ELECTROSTATICS */
 847
 848             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 849             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 850             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 851             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 852             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 853
 854             gmx_fjsp_load_2pair_swizzle_v2r8(ewtab+ewconv.i[0],ewtab+ewconv.i[1],
 855                                          &ewtabF,&ewtabFn);
 856             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 857             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 858
 859             /* LENNARD-JONES DISPERSION/REPULSION */
 860
 861             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 862             fvdw             = _fjsp_mul_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,c6_00),_fjsp_mul_v2r8(rinvsix,rinvsq00));
 863
 864             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 865
 866             fscal            = _fjsp_add_v2r8(felec,fvdw);
 867
 868             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 869
 870             /* Update vectorial force */
 871             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 872             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 873             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 874
 875             fjx0             = _fjsp_madd_v2r8(dx00,fscal,fjx0);
 876             fjy0             = _fjsp_madd_v2r8(dy00,fscal,fjy0);
 877             fjz0             = _fjsp_madd_v2r8(dz00,fscal,fjz0);
 878
 879             }
 880
 881             /**************************
 882              * CALCULATE INTERACTIONS *
 883              **************************/
 884
 885             if (gmx_fjsp_any_lt_v2r8(rsq10,rcutoff2))
 886             {
 887
 888             r10              = _fjsp_mul_v2r8(rsq10,rinv10);
 889
 890             /* Compute parameters for interactions between i and j atoms */
 891             qq10             = _fjsp_mul_v2r8(iq1,jq0);
 892
 893             /* EWALD ELECTROSTATICS */
 894
 895             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 896             ewrt             = _fjsp_mul_v2r8(r10,ewtabscale);
 897             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 898             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 899             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 900
 901             gmx_fjsp_load_2pair_swizzle_v2r8(ewtab+ewconv.i[0],ewtab+ewconv.i[1],
 902                                          &ewtabF,&ewtabFn);
 903             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 904             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq10,rinv10),_fjsp_sub_v2r8(rinvsq10,felec));
 905
 906             cutoff_mask      = _fjsp_cmplt_v2r8(rsq10,rcutoff2);
 907
 908             fscal            = felec;
 909
 910             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 911
 912             /* Update vectorial force */
 913             fix1             = _fjsp_madd_v2r8(dx10,fscal,fix1);
 914             fiy1             = _fjsp_madd_v2r8(dy10,fscal,fiy1);
 915             fiz1             = _fjsp_madd_v2r8(dz10,fscal,fiz1);
 916
 917             fjx0             = _fjsp_madd_v2r8(dx10,fscal,fjx0);
 918             fjy0             = _fjsp_madd_v2r8(dy10,fscal,fjy0);
 919             fjz0             = _fjsp_madd_v2r8(dz10,fscal,fjz0);
 920
 921             }
 922
 923             /**************************
 924              * CALCULATE INTERACTIONS *
 925              **************************/
 926
 927             if (gmx_fjsp_any_lt_v2r8(rsq20,rcutoff2))
 928             {
 929
 930             r20              = _fjsp_mul_v2r8(rsq20,rinv20);
 931
 932             /* Compute parameters for interactions between i and j atoms */
 933             qq20             = _fjsp_mul_v2r8(iq2,jq0);
 934
 935             /* EWALD ELECTROSTATICS */
 936
 937             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 938             ewrt             = _fjsp_mul_v2r8(r20,ewtabscale);
 939             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 940             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 941             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 942
 943             gmx_fjsp_load_2pair_swizzle_v2r8(ewtab+ewconv.i[0],ewtab+ewconv.i[1],
 944                                          &ewtabF,&ewtabFn);
 945             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 946             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq20,rinv20),_fjsp_sub_v2r8(rinvsq20,felec));
 947
 948             cutoff_mask      = _fjsp_cmplt_v2r8(rsq20,rcutoff2);
 949
 950             fscal            = felec;
 951
 952             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 953
 954             /* Update vectorial force */
 955             fix2             = _fjsp_madd_v2r8(dx20,fscal,fix2);
 956             fiy2             = _fjsp_madd_v2r8(dy20,fscal,fiy2);
 957             fiz2             = _fjsp_madd_v2r8(dz20,fscal,fiz2);
 958
 959             fjx0             = _fjsp_madd_v2r8(dx20,fscal,fjx0);
 960             fjy0             = _fjsp_madd_v2r8(dy20,fscal,fjy0);
 961             fjz0             = _fjsp_madd_v2r8(dz20,fscal,fjz0);
 962
 963             }
 964
 965             gmx_fjsp_decrement_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 966
 967             /* Inner loop uses 136 flops */
 968         }
 969
 970         if(jidx<j_index_end)
 971         {
 972
 973             jnrA             = jjnr[jidx];
 974             j_coord_offsetA  = DIM*jnrA;
 975
 976             /* load j atom coordinates */
 977             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 978                                               &jx0,&jy0,&jz0);
 979
 980             /* Calculate displacement vector */
 981             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 982             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 983             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 984             dx10             = _fjsp_sub_v2r8(ix1,jx0);
 985             dy10             = _fjsp_sub_v2r8(iy1,jy0);
 986             dz10             = _fjsp_sub_v2r8(iz1,jz0);
 987             dx20             = _fjsp_sub_v2r8(ix2,jx0);
 988             dy20             = _fjsp_sub_v2r8(iy2,jy0);
 989             dz20             = _fjsp_sub_v2r8(iz2,jz0);
 990
 991             /* Calculate squared distance and things based on it */
 992             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 993             rsq10            = gmx_fjsp_calc_rsq_v2r8(dx10,dy10,dz10);
 994             rsq20            = gmx_fjsp_calc_rsq_v2r8(dx20,dy20,dz20);
 995
 996             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 997             rinv10           = gmx_fjsp_invsqrt_v2r8(rsq10);
 998             rinv20           = gmx_fjsp_invsqrt_v2r8(rsq20);
 999
1000             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
1001             rinvsq10         = _fjsp_mul_v2r8(rinv10,rinv10);
1002             rinvsq20         = _fjsp_mul_v2r8(rinv20,rinv20);
1003
1004             /* Load parameters for j particles */
1005             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
1006             vdwjidx0A        = 2*vdwtype[jnrA+0];
1007
1008             fjx0             = _fjsp_setzero_v2r8();
1009             fjy0             = _fjsp_setzero_v2r8();
1010             fjz0             = _fjsp_setzero_v2r8();
1011
1012             /**************************
1013              * CALCULATE INTERACTIONS *
1014              **************************/
1015
1016             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
1017             {
1018
1019             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
1020
1021             /* Compute parameters for interactions between i and j atoms */
1022             qq00             = _fjsp_mul_v2r8(iq0,jq0);
1023             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
1024                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
1025
1026             /* EWALD ELECTROSTATICS */
1027
1028             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1029             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
1030             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
1031             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
1032             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
1033
1034             gmx_fjsp_load_1pair_swizzle_v2r8(ewtab+ewconv.i[0],&ewtabF,&ewtabFn);
1035             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
1036             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
1037
1038             /* LENNARD-JONES DISPERSION/REPULSION */
1039
1040             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
1041             fvdw             = _fjsp_mul_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,c6_00),_fjsp_mul_v2r8(rinvsix,rinvsq00));
1042
1043             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
1044
1045             fscal            = _fjsp_add_v2r8(felec,fvdw);
1046
1047             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
1048
1049             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
1050
1051             /* Update vectorial force */
1052             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
1053             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
1054             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
1055
1056             fjx0             = _fjsp_madd_v2r8(dx00,fscal,fjx0);
1057             fjy0             = _fjsp_madd_v2r8(dy00,fscal,fjy0);
1058             fjz0             = _fjsp_madd_v2r8(dz00,fscal,fjz0);
1059
1060             }
1061
1062             /**************************
1063              * CALCULATE INTERACTIONS *
1064              **************************/
1065
1066             if (gmx_fjsp_any_lt_v2r8(rsq10,rcutoff2))
1067             {
1068
1069             r10              = _fjsp_mul_v2r8(rsq10,rinv10);
1070
1071             /* Compute parameters for interactions between i and j atoms */
1072             qq10             = _fjsp_mul_v2r8(iq1,jq0);
1073
1074             /* EWALD ELECTROSTATICS */
1075
1076             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1077             ewrt             = _fjsp_mul_v2r8(r10,ewtabscale);
1078             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
1079             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
1080             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
1081
1082             gmx_fjsp_load_1pair_swizzle_v2r8(ewtab+ewconv.i[0],&ewtabF,&ewtabFn);
1083             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
1084             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq10,rinv10),_fjsp_sub_v2r8(rinvsq10,felec));
1085
1086             cutoff_mask      = _fjsp_cmplt_v2r8(rsq10,rcutoff2);
1087
1088             fscal            = felec;
1089
1090             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
1091
1092             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
1093
1094             /* Update vectorial force */
1095             fix1             = _fjsp_madd_v2r8(dx10,fscal,fix1);
1096             fiy1             = _fjsp_madd_v2r8(dy10,fscal,fiy1);
1097             fiz1             = _fjsp_madd_v2r8(dz10,fscal,fiz1);
1098
1099             fjx0             = _fjsp_madd_v2r8(dx10,fscal,fjx0);
1100             fjy0             = _fjsp_madd_v2r8(dy10,fscal,fjy0);
1101             fjz0             = _fjsp_madd_v2r8(dz10,fscal,fjz0);
1102
1103             }
1104
1105             /**************************
1106              * CALCULATE INTERACTIONS *
1107              **************************/
1108
1109             if (gmx_fjsp_any_lt_v2r8(rsq20,rcutoff2))
1110             {
1111
1112             r20              = _fjsp_mul_v2r8(rsq20,rinv20);
1113
1114             /* Compute parameters for interactions between i and j atoms */
1115             qq20             = _fjsp_mul_v2r8(iq2,jq0);
1116
1117             /* EWALD ELECTROSTATICS */
1118
1119             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1120             ewrt             = _fjsp_mul_v2r8(r20,ewtabscale);
1121             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
1122             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
1123             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
1124
1125             gmx_fjsp_load_1pair_swizzle_v2r8(ewtab+ewconv.i[0],&ewtabF,&ewtabFn);
1126             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
1127             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq20,rinv20),_fjsp_sub_v2r8(rinvsq20,felec));
1128
1129             cutoff_mask      = _fjsp_cmplt_v2r8(rsq20,rcutoff2);
1130
1131             fscal            = felec;
1132
1133             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
1134
1135             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
1136
1137             /* Update vectorial force */
1138             fix2             = _fjsp_madd_v2r8(dx20,fscal,fix2);
1139             fiy2             = _fjsp_madd_v2r8(dy20,fscal,fiy2);
1140             fiz2             = _fjsp_madd_v2r8(dz20,fscal,fiz2);
1141
1142             fjx0             = _fjsp_madd_v2r8(dx20,fscal,fjx0);
1143             fjy0             = _fjsp_madd_v2r8(dy20,fscal,fjy0);
1144             fjz0             = _fjsp_madd_v2r8(dz20,fscal,fjz0);
1145
1146             }
1147
1148             gmx_fjsp_decrement_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1149
1150             /* Inner loop uses 136 flops */
1151         }
1152
1153         /* End of innermost loop */
1154
1155         gmx_fjsp_update_iforce_3atom_swizzle_v2r8(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1156                                               f+i_coord_offset,fshift+i_shift_offset);
1157
1158         /* Increment number of inner iterations */
1159         inneriter                  += j_index_end - j_index_start;
1160
1161         /* Outer loop uses 18 flops */
1162     }
1163
1164     /* Increment number of outer iterations */
1165     outeriter        += nri;
1166
1167     /* Update outer/inner flops */
1168
1169     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*136);
1170 }