src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecRFCut_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 #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/math/vec.h"
  47 #include "nrnb.h"
  48
  49 #include "kernelutil_sparc64_hpc_ace_double.h"
  50
  51 /*
  52  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_VF_sparc64_hpc_ace_double
  53  * Electrostatics interaction: ReactionField
  54  * VdW interaction:            LennardJones
  55  * Geometry:                   Particle-Particle
  56  * Calculate force/pot:        PotentialAndForce
  57  */
  58 void
  59 nb_kernel_ElecRFCut_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     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
  87     real             *charge;
  88     int              nvdwtype;
  89     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  90     int              *vdwtype;
  91     real             *vdwparam;
  92     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
  93     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
  94     _fjsp_v2r8       rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  95     real             rswitch_scalar,d_scalar;
  96     _fjsp_v2r8       itab_tmp;
  97     _fjsp_v2r8       dummy_mask,cutoff_mask;
  98     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
  99     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 100     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 101
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115     krf              = gmx_fjsp_set1_v2r8(fr->ic->k_rf);
 116     krf2             = gmx_fjsp_set1_v2r8(fr->ic->k_rf*2.0);
 117     crf              = gmx_fjsp_set1_v2r8(fr->ic->c_rf);
 118     nvdwtype         = fr->ntype;
 119     vdwparam         = fr->nbfp;
 120     vdwtype          = mdatoms->typeA;
 121
 122     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 123     rcutoff_scalar   = fr->rcoulomb;
 124     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 125     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 126
 127     rswitch_scalar   = fr->rvdw_switch;
 128     rswitch          = gmx_fjsp_set1_v2r8(rswitch_scalar);
 129     /* Setup switch parameters */
 130     d_scalar         = rcutoff_scalar-rswitch_scalar;
 131     d                = gmx_fjsp_set1_v2r8(d_scalar);
 132     swV3             = gmx_fjsp_set1_v2r8(-10.0/(d_scalar*d_scalar*d_scalar));
 133     swV4             = gmx_fjsp_set1_v2r8( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 134     swV5             = gmx_fjsp_set1_v2r8( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 135     swF2             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar));
 136     swF3             = gmx_fjsp_set1_v2r8( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 137     swF4             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 138
 139     /* Avoid stupid compiler warnings */
 140     jnrA = jnrB = 0;
 141     j_coord_offsetA = 0;
 142     j_coord_offsetB = 0;
 143
 144     outeriter        = 0;
 145     inneriter        = 0;
 146
 147     /* Start outer loop over neighborlists */
 148     for(iidx=0; iidx<nri; iidx++)
 149     {
 150         /* Load shift vector for this list */
 151         i_shift_offset   = DIM*shiftidx[iidx];
 152
 153         /* Load limits for loop over neighbors */
 154         j_index_start    = jindex[iidx];
 155         j_index_end      = jindex[iidx+1];
 156
 157         /* Get outer coordinate index */
 158         inr              = iinr[iidx];
 159         i_coord_offset   = DIM*inr;
 160
 161         /* Load i particle coords and add shift vector */
 162         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 163
 164         fix0             = _fjsp_setzero_v2r8();
 165         fiy0             = _fjsp_setzero_v2r8();
 166         fiz0             = _fjsp_setzero_v2r8();
 167
 168         /* Load parameters for i particles */
 169         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 170         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 171
 172         /* Reset potential sums */
 173         velecsum         = _fjsp_setzero_v2r8();
 174         vvdwsum          = _fjsp_setzero_v2r8();
 175
 176         /* Start inner kernel loop */
 177         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 178         {
 179
 180             /* Get j neighbor index, and coordinate index */
 181             jnrA             = jjnr[jidx];
 182             jnrB             = jjnr[jidx+1];
 183             j_coord_offsetA  = DIM*jnrA;
 184             j_coord_offsetB  = DIM*jnrB;
 185
 186             /* load j atom coordinates */
 187             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 188                                               &jx0,&jy0,&jz0);
 189
 190             /* Calculate displacement vector */
 191             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 192             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 193             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 194
 195             /* Calculate squared distance and things based on it */
 196             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 197
 198             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 199
 200             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 201
 202             /* Load parameters for j particles */
 203             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 204             vdwjidx0A        = 2*vdwtype[jnrA+0];
 205             vdwjidx0B        = 2*vdwtype[jnrB+0];
 206
 207             /**************************
 208              * CALCULATE INTERACTIONS *
 209              **************************/
 210
 211             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 212             {
 213
 214             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 215
 216             /* Compute parameters for interactions between i and j atoms */
 217             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 218             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 219                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 220
 221             /* REACTION-FIELD ELECTROSTATICS */
 222             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(_fjsp_madd_v2r8(krf,rsq00,rinv00),crf));
 223             felec            = _fjsp_mul_v2r8(qq00,_fjsp_msub_v2r8(rinv00,rinvsq00,krf2));
 224
 225             /* LENNARD-JONES DISPERSION/REPULSION */
 226
 227             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 228             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 229             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 230             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 231             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 232
 233             d                = _fjsp_sub_v2r8(r00,rswitch);
 234             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 235             d2               = _fjsp_mul_v2r8(d,d);
 236             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 237
 238             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 239
 240             /* Evaluate switch function */
 241             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 242             fvdw             = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
 243             vvdw             = _fjsp_mul_v2r8(vvdw,sw);
 244             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 245
 246             /* Update potential sum for this i atom from the interaction with this j atom. */
 247             velec            = _fjsp_and_v2r8(velec,cutoff_mask);
 248             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 249             vvdw             = _fjsp_and_v2r8(vvdw,cutoff_mask);
 250             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 251
 252             fscal            = _fjsp_add_v2r8(felec,fvdw);
 253
 254             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 255
 256             /* Update vectorial force */
 257             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 258             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 259             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 260
 261             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 262
 263             }
 264
 265             /* Inner loop uses 73 flops */
 266         }
 267
 268         if(jidx<j_index_end)
 269         {
 270
 271             jnrA             = jjnr[jidx];
 272             j_coord_offsetA  = DIM*jnrA;
 273
 274             /* load j atom coordinates */
 275             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 276                                               &jx0,&jy0,&jz0);
 277
 278             /* Calculate displacement vector */
 279             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 280             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 281             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 282
 283             /* Calculate squared distance and things based on it */
 284             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 285
 286             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 287
 288             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 289
 290             /* Load parameters for j particles */
 291             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 292             vdwjidx0A        = 2*vdwtype[jnrA+0];
 293
 294             /**************************
 295              * CALCULATE INTERACTIONS *
 296              **************************/
 297
 298             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 299             {
 300
 301             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 302
 303             /* Compute parameters for interactions between i and j atoms */
 304             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 305             gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 306
 307             /* REACTION-FIELD ELECTROSTATICS */
 308             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(_fjsp_madd_v2r8(krf,rsq00,rinv00),crf));
 309             felec            = _fjsp_mul_v2r8(qq00,_fjsp_msub_v2r8(rinv00,rinvsq00,krf2));
 310
 311             /* LENNARD-JONES DISPERSION/REPULSION */
 312
 313             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 314             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 315             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 316             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 317             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 318
 319             d                = _fjsp_sub_v2r8(r00,rswitch);
 320             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 321             d2               = _fjsp_mul_v2r8(d,d);
 322             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 323
 324             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 325
 326             /* Evaluate switch function */
 327             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 328             fvdw             = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
 329             vvdw             = _fjsp_mul_v2r8(vvdw,sw);
 330             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 331
 332             /* Update potential sum for this i atom from the interaction with this j atom. */
 333             velec            = _fjsp_and_v2r8(velec,cutoff_mask);
 334             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 335             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 336             vvdw             = _fjsp_and_v2r8(vvdw,cutoff_mask);
 337             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 338             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 339
 340             fscal            = _fjsp_add_v2r8(felec,fvdw);
 341
 342             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 343
 344             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 345
 346             /* Update vectorial force */
 347             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 348             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 349             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 350
 351             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 352
 353             }
 354
 355             /* Inner loop uses 73 flops */
 356         }
 357
 358         /* End of innermost loop */
 359
 360         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 361                                               f+i_coord_offset,fshift+i_shift_offset);
 362
 363         ggid                        = gid[iidx];
 364         /* Update potential energies */
 365         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 366         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 367
 368         /* Increment number of inner iterations */
 369         inneriter                  += j_index_end - j_index_start;
 370
 371         /* Outer loop uses 9 flops */
 372     }
 373
 374     /* Increment number of outer iterations */
 375     outeriter        += nri;
 376
 377     /* Update outer/inner flops */
 378
 379     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*73);
 380 }
 381 /*
 382  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_F_sparc64_hpc_ace_double
 383  * Electrostatics interaction: ReactionField
 384  * VdW interaction:            LennardJones
 385  * Geometry:                   Particle-Particle
 386  * Calculate force/pot:        Force
 387  */
 388 void
 389 nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_F_sparc64_hpc_ace_double
 390                     (t_nblist                    * gmx_restrict       nlist,
 391                      rvec                        * gmx_restrict          xx,
 392                      rvec                        * gmx_restrict          ff,
 393                      t_forcerec                  * gmx_restrict          fr,
 394                      t_mdatoms                   * gmx_restrict     mdatoms,
 395                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 396                      t_nrnb                      * gmx_restrict        nrnb)
 397 {
 398     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 399      * just 0 for non-waters.
 400      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 401      * jnr indices corresponding to data put in the four positions in the SIMD register.
 402      */
 403     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 404     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 405     int              jnrA,jnrB;
 406     int              j_coord_offsetA,j_coord_offsetB;
 407     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 408     real             rcutoff_scalar;
 409     real             *shiftvec,*fshift,*x,*f;
 410     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 411     int              vdwioffset0;
 412     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 413     int              vdwjidx0A,vdwjidx0B;
 414     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 415     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 416     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 417     real             *charge;
 418     int              nvdwtype;
 419     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 420     int              *vdwtype;
 421     real             *vdwparam;
 422     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 423     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 424     _fjsp_v2r8       rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 425     real             rswitch_scalar,d_scalar;
 426     _fjsp_v2r8       itab_tmp;
 427     _fjsp_v2r8       dummy_mask,cutoff_mask;
 428     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 429     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 430     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 431
 432     x                = xx[0];
 433     f                = ff[0];
 434
 435     nri              = nlist->nri;
 436     iinr             = nlist->iinr;
 437     jindex           = nlist->jindex;
 438     jjnr             = nlist->jjnr;
 439     shiftidx         = nlist->shift;
 440     gid              = nlist->gid;
 441     shiftvec         = fr->shift_vec[0];
 442     fshift           = fr->fshift[0];
 443     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 444     charge           = mdatoms->chargeA;
 445     krf              = gmx_fjsp_set1_v2r8(fr->ic->k_rf);
 446     krf2             = gmx_fjsp_set1_v2r8(fr->ic->k_rf*2.0);
 447     crf              = gmx_fjsp_set1_v2r8(fr->ic->c_rf);
 448     nvdwtype         = fr->ntype;
 449     vdwparam         = fr->nbfp;
 450     vdwtype          = mdatoms->typeA;
 451
 452     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 453     rcutoff_scalar   = fr->rcoulomb;
 454     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 455     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 456
 457     rswitch_scalar   = fr->rvdw_switch;
 458     rswitch          = gmx_fjsp_set1_v2r8(rswitch_scalar);
 459     /* Setup switch parameters */
 460     d_scalar         = rcutoff_scalar-rswitch_scalar;
 461     d                = gmx_fjsp_set1_v2r8(d_scalar);
 462     swV3             = gmx_fjsp_set1_v2r8(-10.0/(d_scalar*d_scalar*d_scalar));
 463     swV4             = gmx_fjsp_set1_v2r8( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 464     swV5             = gmx_fjsp_set1_v2r8( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 465     swF2             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar));
 466     swF3             = gmx_fjsp_set1_v2r8( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 467     swF4             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 468
 469     /* Avoid stupid compiler warnings */
 470     jnrA = jnrB = 0;
 471     j_coord_offsetA = 0;
 472     j_coord_offsetB = 0;
 473
 474     outeriter        = 0;
 475     inneriter        = 0;
 476
 477     /* Start outer loop over neighborlists */
 478     for(iidx=0; iidx<nri; iidx++)
 479     {
 480         /* Load shift vector for this list */
 481         i_shift_offset   = DIM*shiftidx[iidx];
 482
 483         /* Load limits for loop over neighbors */
 484         j_index_start    = jindex[iidx];
 485         j_index_end      = jindex[iidx+1];
 486
 487         /* Get outer coordinate index */
 488         inr              = iinr[iidx];
 489         i_coord_offset   = DIM*inr;
 490
 491         /* Load i particle coords and add shift vector */
 492         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 493
 494         fix0             = _fjsp_setzero_v2r8();
 495         fiy0             = _fjsp_setzero_v2r8();
 496         fiz0             = _fjsp_setzero_v2r8();
 497
 498         /* Load parameters for i particles */
 499         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 500         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 501
 502         /* Start inner kernel loop */
 503         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 504         {
 505
 506             /* Get j neighbor index, and coordinate index */
 507             jnrA             = jjnr[jidx];
 508             jnrB             = jjnr[jidx+1];
 509             j_coord_offsetA  = DIM*jnrA;
 510             j_coord_offsetB  = DIM*jnrB;
 511
 512             /* load j atom coordinates */
 513             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 514                                               &jx0,&jy0,&jz0);
 515
 516             /* Calculate displacement vector */
 517             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 518             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 519             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 520
 521             /* Calculate squared distance and things based on it */
 522             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 523
 524             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 525
 526             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 527
 528             /* Load parameters for j particles */
 529             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 530             vdwjidx0A        = 2*vdwtype[jnrA+0];
 531             vdwjidx0B        = 2*vdwtype[jnrB+0];
 532
 533             /**************************
 534              * CALCULATE INTERACTIONS *
 535              **************************/
 536
 537             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 538             {
 539
 540             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 541
 542             /* Compute parameters for interactions between i and j atoms */
 543             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 544             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 545                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 546
 547             /* REACTION-FIELD ELECTROSTATICS */
 548             felec            = _fjsp_mul_v2r8(qq00,_fjsp_msub_v2r8(rinv00,rinvsq00,krf2));
 549
 550             /* LENNARD-JONES DISPERSION/REPULSION */
 551
 552             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 553             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 554             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 555             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 556             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 557
 558             d                = _fjsp_sub_v2r8(r00,rswitch);
 559             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 560             d2               = _fjsp_mul_v2r8(d,d);
 561             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 562
 563             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 564
 565             /* Evaluate switch function */
 566             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 567             fvdw             = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
 568             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 569
 570             fscal            = _fjsp_add_v2r8(felec,fvdw);
 571
 572             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 573
 574             /* Update vectorial force */
 575             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 576             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 577             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 578
 579             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 580
 581             }
 582
 583             /* Inner loop uses 64 flops */
 584         }
 585
 586         if(jidx<j_index_end)
 587         {
 588
 589             jnrA             = jjnr[jidx];
 590             j_coord_offsetA  = DIM*jnrA;
 591
 592             /* load j atom coordinates */
 593             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 594                                               &jx0,&jy0,&jz0);
 595
 596             /* Calculate displacement vector */
 597             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 598             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 599             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 600
 601             /* Calculate squared distance and things based on it */
 602             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 603
 604             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 605
 606             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 607
 608             /* Load parameters for j particles */
 609             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 610             vdwjidx0A        = 2*vdwtype[jnrA+0];
 611
 612             /**************************
 613              * CALCULATE INTERACTIONS *
 614              **************************/
 615
 616             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 617             {
 618
 619             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 620
 621             /* Compute parameters for interactions between i and j atoms */
 622             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 623             gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 624
 625             /* REACTION-FIELD ELECTROSTATICS */
 626             felec            = _fjsp_mul_v2r8(qq00,_fjsp_msub_v2r8(rinv00,rinvsq00,krf2));
 627
 628             /* LENNARD-JONES DISPERSION/REPULSION */
 629
 630             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 631             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 632             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 633             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 634             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 635
 636             d                = _fjsp_sub_v2r8(r00,rswitch);
 637             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 638             d2               = _fjsp_mul_v2r8(d,d);
 639             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 640
 641             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 642
 643             /* Evaluate switch function */
 644             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 645             fvdw             = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
 646             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 647
 648             fscal            = _fjsp_add_v2r8(felec,fvdw);
 649
 650             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 651
 652             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 653
 654             /* Update vectorial force */
 655             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 656             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 657             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 658
 659             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 660
 661             }
 662
 663             /* Inner loop uses 64 flops */
 664         }
 665
 666         /* End of innermost loop */
 667
 668         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 669                                               f+i_coord_offset,fshift+i_shift_offset);
 670
 671         /* Increment number of inner iterations */
 672         inneriter                  += j_index_end - j_index_start;
 673
 674         /* Outer loop uses 7 flops */
 675     }
 676
 677     /* Increment number of outer iterations */
 678     outeriter        += nri;
 679
 680     /* Update outer/inner flops */
 681
 682     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*64);
 683 }