src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecNone_VdwLJSw_GeomP1P1_sparc64_hpc_ace_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sparc64_hpc_ace_double kernel generator.
  37  */
  38 #include "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_ElecNone_VdwLJSw_GeomP1P1_VF_sparc64_hpc_ace_double
  53  * Electrostatics interaction: None
  54  * VdW interaction:            LennardJones
  55  * Geometry:                   Particle-Particle
  56  * Calculate force/pot:        PotentialAndForce
  57  */
  58 void
  59 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_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              vdwjidx0A,vdwjidx0B;
  84     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  85     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  86     int              nvdwtype;
  87     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  88     int              *vdwtype;
  89     real             *vdwparam;
  90     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
  91     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
  92     _fjsp_v2r8       rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  93     real             rswitch_scalar,d_scalar;
  94     _fjsp_v2r8       itab_tmp;
  95     _fjsp_v2r8       dummy_mask,cutoff_mask;
  96     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
  97     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
  98     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
  99
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     nvdwtype         = fr->ntype;
 112     vdwparam         = fr->nbfp;
 113     vdwtype          = mdatoms->typeA;
 114
 115     rcutoff_scalar   = fr->rvdw;
 116     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 117     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 118
 119     rswitch_scalar   = fr->rvdw_switch;
 120     rswitch          = gmx_fjsp_set1_v2r8(rswitch_scalar);
 121     /* Setup switch parameters */
 122     d_scalar         = rcutoff_scalar-rswitch_scalar;
 123     d                = gmx_fjsp_set1_v2r8(d_scalar);
 124     swV3             = gmx_fjsp_set1_v2r8(-10.0/(d_scalar*d_scalar*d_scalar));
 125     swV4             = gmx_fjsp_set1_v2r8( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 126     swV5             = gmx_fjsp_set1_v2r8( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 127     swF2             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar));
 128     swF3             = gmx_fjsp_set1_v2r8( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 129     swF4             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 130
 131     /* Avoid stupid compiler warnings */
 132     jnrA = jnrB = 0;
 133     j_coord_offsetA = 0;
 134     j_coord_offsetB = 0;
 135
 136     outeriter        = 0;
 137     inneriter        = 0;
 138
 139     /* Start outer loop over neighborlists */
 140     for(iidx=0; iidx<nri; iidx++)
 141     {
 142         /* Load shift vector for this list */
 143         i_shift_offset   = DIM*shiftidx[iidx];
 144
 145         /* Load limits for loop over neighbors */
 146         j_index_start    = jindex[iidx];
 147         j_index_end      = jindex[iidx+1];
 148
 149         /* Get outer coordinate index */
 150         inr              = iinr[iidx];
 151         i_coord_offset   = DIM*inr;
 152
 153         /* Load i particle coords and add shift vector */
 154         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 155
 156         fix0             = _fjsp_setzero_v2r8();
 157         fiy0             = _fjsp_setzero_v2r8();
 158         fiz0             = _fjsp_setzero_v2r8();
 159
 160         /* Load parameters for i particles */
 161         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 162
 163         /* Reset potential sums */
 164         vvdwsum          = _fjsp_setzero_v2r8();
 165
 166         /* Start inner kernel loop */
 167         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 168         {
 169
 170             /* Get j neighbor index, and coordinate index */
 171             jnrA             = jjnr[jidx];
 172             jnrB             = jjnr[jidx+1];
 173             j_coord_offsetA  = DIM*jnrA;
 174             j_coord_offsetB  = DIM*jnrB;
 175
 176             /* load j atom coordinates */
 177             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 178                                               &jx0,&jy0,&jz0);
 179
 180             /* Calculate displacement vector */
 181             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 182             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 183             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 184
 185             /* Calculate squared distance and things based on it */
 186             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 187
 188             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 189
 190             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 191
 192             /* Load parameters for j particles */
 193             vdwjidx0A        = 2*vdwtype[jnrA+0];
 194             vdwjidx0B        = 2*vdwtype[jnrB+0];
 195
 196             /**************************
 197              * CALCULATE INTERACTIONS *
 198              **************************/
 199
 200             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 201             {
 202
 203             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 204
 205             /* Compute parameters for interactions between i and j atoms */
 206             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 207                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 208
 209             /* LENNARD-JONES DISPERSION/REPULSION */
 210
 211             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 212             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 213             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 214             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 215             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 216
 217             d                = _fjsp_sub_v2r8(r00,rswitch);
 218             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 219             d2               = _fjsp_mul_v2r8(d,d);
 220             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 221
 222             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 223
 224             /* Evaluate switch function */
 225             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 226             fvdw             = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
 227             vvdw             = _fjsp_mul_v2r8(vvdw,sw);
 228             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 229
 230             /* Update potential sum for this i atom from the interaction with this j atom. */
 231             vvdw             = _fjsp_and_v2r8(vvdw,cutoff_mask);
 232             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 233
 234             fscal            = fvdw;
 235
 236             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 237
 238             /* Update vectorial force */
 239             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 240             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 241             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 242
 243             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 244
 245             }
 246
 247             /* Inner loop uses 62 flops */
 248         }
 249
 250         if(jidx<j_index_end)
 251         {
 252
 253             jnrA             = jjnr[jidx];
 254             j_coord_offsetA  = DIM*jnrA;
 255
 256             /* load j atom coordinates */
 257             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 258                                               &jx0,&jy0,&jz0);
 259
 260             /* Calculate displacement vector */
 261             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 262             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 263             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 264
 265             /* Calculate squared distance and things based on it */
 266             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 267
 268             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 269
 270             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 271
 272             /* Load parameters for j particles */
 273             vdwjidx0A        = 2*vdwtype[jnrA+0];
 274
 275             /**************************
 276              * CALCULATE INTERACTIONS *
 277              **************************/
 278
 279             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 280             {
 281
 282             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 283
 284             /* Compute parameters for interactions between i and j atoms */
 285             gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 286
 287             /* LENNARD-JONES DISPERSION/REPULSION */
 288
 289             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 290             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 291             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 292             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 293             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 294
 295             d                = _fjsp_sub_v2r8(r00,rswitch);
 296             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 297             d2               = _fjsp_mul_v2r8(d,d);
 298             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 299
 300             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 301
 302             /* Evaluate switch function */
 303             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 304             fvdw             = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
 305             vvdw             = _fjsp_mul_v2r8(vvdw,sw);
 306             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 307
 308             /* Update potential sum for this i atom from the interaction with this j atom. */
 309             vvdw             = _fjsp_and_v2r8(vvdw,cutoff_mask);
 310             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 311             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 312
 313             fscal            = fvdw;
 314
 315             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 316
 317             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 318
 319             /* Update vectorial force */
 320             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 321             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 322             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 323
 324             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 325
 326             }
 327
 328             /* Inner loop uses 62 flops */
 329         }
 330
 331         /* End of innermost loop */
 332
 333         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 334                                               f+i_coord_offset,fshift+i_shift_offset);
 335
 336         ggid                        = gid[iidx];
 337         /* Update potential energies */
 338         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 339
 340         /* Increment number of inner iterations */
 341         inneriter                  += j_index_end - j_index_start;
 342
 343         /* Outer loop uses 7 flops */
 344     }
 345
 346     /* Increment number of outer iterations */
 347     outeriter        += nri;
 348
 349     /* Update outer/inner flops */
 350
 351     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*62);
 352 }
 353 /*
 354  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_sparc64_hpc_ace_double
 355  * Electrostatics interaction: None
 356  * VdW interaction:            LennardJones
 357  * Geometry:                   Particle-Particle
 358  * Calculate force/pot:        Force
 359  */
 360 void
 361 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_sparc64_hpc_ace_double
 362                     (t_nblist                    * gmx_restrict       nlist,
 363                      rvec                        * gmx_restrict          xx,
 364                      rvec                        * gmx_restrict          ff,
 365                      t_forcerec                  * gmx_restrict          fr,
 366                      t_mdatoms                   * gmx_restrict     mdatoms,
 367                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 368                      t_nrnb                      * gmx_restrict        nrnb)
 369 {
 370     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 371      * just 0 for non-waters.
 372      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 373      * jnr indices corresponding to data put in the four positions in the SIMD register.
 374      */
 375     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 376     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 377     int              jnrA,jnrB;
 378     int              j_coord_offsetA,j_coord_offsetB;
 379     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 380     real             rcutoff_scalar;
 381     real             *shiftvec,*fshift,*x,*f;
 382     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 383     int              vdwioffset0;
 384     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 385     int              vdwjidx0A,vdwjidx0B;
 386     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 387     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 388     int              nvdwtype;
 389     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 390     int              *vdwtype;
 391     real             *vdwparam;
 392     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 393     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 394     _fjsp_v2r8       rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 395     real             rswitch_scalar,d_scalar;
 396     _fjsp_v2r8       itab_tmp;
 397     _fjsp_v2r8       dummy_mask,cutoff_mask;
 398     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 399     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 400     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 401
 402     x                = xx[0];
 403     f                = ff[0];
 404
 405     nri              = nlist->nri;
 406     iinr             = nlist->iinr;
 407     jindex           = nlist->jindex;
 408     jjnr             = nlist->jjnr;
 409     shiftidx         = nlist->shift;
 410     gid              = nlist->gid;
 411     shiftvec         = fr->shift_vec[0];
 412     fshift           = fr->fshift[0];
 413     nvdwtype         = fr->ntype;
 414     vdwparam         = fr->nbfp;
 415     vdwtype          = mdatoms->typeA;
 416
 417     rcutoff_scalar   = fr->rvdw;
 418     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 419     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 420
 421     rswitch_scalar   = fr->rvdw_switch;
 422     rswitch          = gmx_fjsp_set1_v2r8(rswitch_scalar);
 423     /* Setup switch parameters */
 424     d_scalar         = rcutoff_scalar-rswitch_scalar;
 425     d                = gmx_fjsp_set1_v2r8(d_scalar);
 426     swV3             = gmx_fjsp_set1_v2r8(-10.0/(d_scalar*d_scalar*d_scalar));
 427     swV4             = gmx_fjsp_set1_v2r8( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 428     swV5             = gmx_fjsp_set1_v2r8( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 429     swF2             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar));
 430     swF3             = gmx_fjsp_set1_v2r8( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 431     swF4             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 432
 433     /* Avoid stupid compiler warnings */
 434     jnrA = jnrB = 0;
 435     j_coord_offsetA = 0;
 436     j_coord_offsetB = 0;
 437
 438     outeriter        = 0;
 439     inneriter        = 0;
 440
 441     /* Start outer loop over neighborlists */
 442     for(iidx=0; iidx<nri; iidx++)
 443     {
 444         /* Load shift vector for this list */
 445         i_shift_offset   = DIM*shiftidx[iidx];
 446
 447         /* Load limits for loop over neighbors */
 448         j_index_start    = jindex[iidx];
 449         j_index_end      = jindex[iidx+1];
 450
 451         /* Get outer coordinate index */
 452         inr              = iinr[iidx];
 453         i_coord_offset   = DIM*inr;
 454
 455         /* Load i particle coords and add shift vector */
 456         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 457
 458         fix0             = _fjsp_setzero_v2r8();
 459         fiy0             = _fjsp_setzero_v2r8();
 460         fiz0             = _fjsp_setzero_v2r8();
 461
 462         /* Load parameters for i particles */
 463         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 464
 465         /* Start inner kernel loop */
 466         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 467         {
 468
 469             /* Get j neighbor index, and coordinate index */
 470             jnrA             = jjnr[jidx];
 471             jnrB             = jjnr[jidx+1];
 472             j_coord_offsetA  = DIM*jnrA;
 473             j_coord_offsetB  = DIM*jnrB;
 474
 475             /* load j atom coordinates */
 476             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 477                                               &jx0,&jy0,&jz0);
 478
 479             /* Calculate displacement vector */
 480             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 481             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 482             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 483
 484             /* Calculate squared distance and things based on it */
 485             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 486
 487             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 488
 489             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 490
 491             /* Load parameters for j particles */
 492             vdwjidx0A        = 2*vdwtype[jnrA+0];
 493             vdwjidx0B        = 2*vdwtype[jnrB+0];
 494
 495             /**************************
 496              * CALCULATE INTERACTIONS *
 497              **************************/
 498
 499             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 500             {
 501
 502             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 503
 504             /* Compute parameters for interactions between i and j atoms */
 505             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 506                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 507
 508             /* LENNARD-JONES DISPERSION/REPULSION */
 509
 510             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 511             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 512             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 513             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 514             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 515
 516             d                = _fjsp_sub_v2r8(r00,rswitch);
 517             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 518             d2               = _fjsp_mul_v2r8(d,d);
 519             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 520
 521             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 522
 523             /* Evaluate switch function */
 524             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 525             fvdw             = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
 526             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 527
 528             fscal            = fvdw;
 529
 530             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 531
 532             /* Update vectorial force */
 533             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 534             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 535             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 536
 537             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 538
 539             }
 540
 541             /* Inner loop uses 59 flops */
 542         }
 543
 544         if(jidx<j_index_end)
 545         {
 546
 547             jnrA             = jjnr[jidx];
 548             j_coord_offsetA  = DIM*jnrA;
 549
 550             /* load j atom coordinates */
 551             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 552                                               &jx0,&jy0,&jz0);
 553
 554             /* Calculate displacement vector */
 555             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 556             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 557             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 558
 559             /* Calculate squared distance and things based on it */
 560             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 561
 562             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 563
 564             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 565
 566             /* Load parameters for j particles */
 567             vdwjidx0A        = 2*vdwtype[jnrA+0];
 568
 569             /**************************
 570              * CALCULATE INTERACTIONS *
 571              **************************/
 572
 573             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 574             {
 575
 576             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 577
 578             /* Compute parameters for interactions between i and j atoms */
 579             gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 580
 581             /* LENNARD-JONES DISPERSION/REPULSION */
 582
 583             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 584             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 585             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 586             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 587             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 588
 589             d                = _fjsp_sub_v2r8(r00,rswitch);
 590             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 591             d2               = _fjsp_mul_v2r8(d,d);
 592             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 593
 594             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 595
 596             /* Evaluate switch function */
 597             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 598             fvdw             = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
 599             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 600
 601             fscal            = fvdw;
 602
 603             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 604
 605             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 606
 607             /* Update vectorial force */
 608             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 609             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 610             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 611
 612             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 613
 614             }
 615
 616             /* Inner loop uses 59 flops */
 617         }
 618
 619         /* End of innermost loop */
 620
 621         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 622                                               f+i_coord_offset,fshift+i_shift_offset);
 623
 624         /* Increment number of inner iterations */
 625         inneriter                  += j_index_end - j_index_start;
 626
 627         /* Outer loop uses 6 flops */
 628     }
 629
 630     /* Increment number of outer iterations */
 631     outeriter        += nri;
 632
 633     /* Update outer/inner flops */
 634
 635     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*59);
 636 }