src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecEwSw_VdwNone_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_ElecEwSw_VdwNone_GeomP1P1_VF_sparc64_hpc_ace_double
  53  * Electrostatics interaction: Ewald
  54  * VdW interaction:            None
  55  * Geometry:                   Particle-Particle
  56  * Calculate force/pot:        PotentialAndForce
  57  */
  58 void
  59 nb_kernel_ElecEwSw_VdwNone_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     _fjsp_v2r8       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  89     real             *ewtab;
  90     _fjsp_v2r8       rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  91     real             rswitch_scalar,d_scalar;
  92     _fjsp_v2r8       itab_tmp;
  93     _fjsp_v2r8       dummy_mask,cutoff_mask;
  94     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
  95     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
  96     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
  97
  98     x                = xx[0];
  99     f                = ff[0];
 100
 101     nri              = nlist->nri;
 102     iinr             = nlist->iinr;
 103     jindex           = nlist->jindex;
 104     jjnr             = nlist->jjnr;
 105     shiftidx         = nlist->shift;
 106     gid              = nlist->gid;
 107     shiftvec         = fr->shift_vec[0];
 108     fshift           = fr->fshift[0];
 109     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 110     charge           = mdatoms->chargeA;
 111
 112     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 113     ewtab            = fr->ic->tabq_coul_FDV0;
 114     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 115     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 116
 117     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 118     rcutoff_scalar   = fr->rcoulomb;
 119     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 120     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 121
 122     rswitch_scalar   = fr->rcoulomb_switch;
 123     rswitch          = gmx_fjsp_set1_v2r8(rswitch_scalar);
 124     /* Setup switch parameters */
 125     d_scalar         = rcutoff_scalar-rswitch_scalar;
 126     d                = gmx_fjsp_set1_v2r8(d_scalar);
 127     swV3             = gmx_fjsp_set1_v2r8(-10.0/(d_scalar*d_scalar*d_scalar));
 128     swV4             = gmx_fjsp_set1_v2r8( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 129     swV5             = gmx_fjsp_set1_v2r8( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 130     swF2             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar));
 131     swF3             = gmx_fjsp_set1_v2r8( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 132     swF4             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 133
 134     /* Avoid stupid compiler warnings */
 135     jnrA = jnrB = 0;
 136     j_coord_offsetA = 0;
 137     j_coord_offsetB = 0;
 138
 139     outeriter        = 0;
 140     inneriter        = 0;
 141
 142     /* Start outer loop over neighborlists */
 143     for(iidx=0; iidx<nri; iidx++)
 144     {
 145         /* Load shift vector for this list */
 146         i_shift_offset   = DIM*shiftidx[iidx];
 147
 148         /* Load limits for loop over neighbors */
 149         j_index_start    = jindex[iidx];
 150         j_index_end      = jindex[iidx+1];
 151
 152         /* Get outer coordinate index */
 153         inr              = iinr[iidx];
 154         i_coord_offset   = DIM*inr;
 155
 156         /* Load i particle coords and add shift vector */
 157         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 158
 159         fix0             = _fjsp_setzero_v2r8();
 160         fiy0             = _fjsp_setzero_v2r8();
 161         fiz0             = _fjsp_setzero_v2r8();
 162
 163         /* Load parameters for i particles */
 164         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 165
 166         /* Reset potential sums */
 167         velecsum         = _fjsp_setzero_v2r8();
 168
 169         /* Start inner kernel loop */
 170         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 171         {
 172
 173             /* Get j neighbor index, and coordinate index */
 174             jnrA             = jjnr[jidx];
 175             jnrB             = jjnr[jidx+1];
 176             j_coord_offsetA  = DIM*jnrA;
 177             j_coord_offsetB  = DIM*jnrB;
 178
 179             /* load j atom coordinates */
 180             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 181                                               &jx0,&jy0,&jz0);
 182
 183             /* Calculate displacement vector */
 184             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 185             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 186             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 187
 188             /* Calculate squared distance and things based on it */
 189             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 190
 191             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 192
 193             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 194
 195             /* Load parameters for j particles */
 196             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 197
 198             /**************************
 199              * CALCULATE INTERACTIONS *
 200              **************************/
 201
 202             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 203             {
 204
 205             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 206
 207             /* Compute parameters for interactions between i and j atoms */
 208             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 209
 210             /* EWALD ELECTROSTATICS */
 211
 212             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 213             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 214             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 215             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 216             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 217
 218             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 219             ewtabD           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
 220             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 221             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 222             ewtabFn          = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
 223             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 224             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 225             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 226             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 227             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 228
 229             d                = _fjsp_sub_v2r8(r00,rswitch);
 230             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 231             d2               = _fjsp_mul_v2r8(d,d);
 232             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 233
 234             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 235
 236             /* Evaluate switch function */
 237             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 238             felec            = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(velec,dsw)) );
 239             velec            = _fjsp_mul_v2r8(velec,sw);
 240             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 241
 242             /* Update potential sum for this i atom from the interaction with this j atom. */
 243             velec            = _fjsp_and_v2r8(velec,cutoff_mask);
 244             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 245
 246             fscal            = felec;
 247
 248             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 249
 250             /* Update vectorial force */
 251             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 252             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 253             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 254
 255             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 256
 257             }
 258
 259             /* Inner loop uses 68 flops */
 260         }
 261
 262         if(jidx<j_index_end)
 263         {
 264
 265             jnrA             = jjnr[jidx];
 266             j_coord_offsetA  = DIM*jnrA;
 267
 268             /* load j atom coordinates */
 269             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 270                                               &jx0,&jy0,&jz0);
 271
 272             /* Calculate displacement vector */
 273             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 274             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 275             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 276
 277             /* Calculate squared distance and things based on it */
 278             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 279
 280             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 281
 282             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 283
 284             /* Load parameters for j particles */
 285             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 286
 287             /**************************
 288              * CALCULATE INTERACTIONS *
 289              **************************/
 290
 291             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 292             {
 293
 294             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 295
 296             /* Compute parameters for interactions between i and j atoms */
 297             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 298
 299             /* EWALD ELECTROSTATICS */
 300
 301             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 302             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 303             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 304             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 305             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 306
 307             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 308             ewtabD           = _fjsp_setzero_v2r8();
 309             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 310             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 311             ewtabFn          = _fjsp_setzero_v2r8();
 312             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 313             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 314             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 315             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 316             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 317
 318             d                = _fjsp_sub_v2r8(r00,rswitch);
 319             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 320             d2               = _fjsp_mul_v2r8(d,d);
 321             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 322
 323             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 324
 325             /* Evaluate switch function */
 326             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 327             felec            = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(velec,dsw)) );
 328             velec            = _fjsp_mul_v2r8(velec,sw);
 329             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 330
 331             /* Update potential sum for this i atom from the interaction with this j atom. */
 332             velec            = _fjsp_and_v2r8(velec,cutoff_mask);
 333             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 334             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 335
 336             fscal            = felec;
 337
 338             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 339
 340             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 341
 342             /* Update vectorial force */
 343             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 344             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 345             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 346
 347             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 348
 349             }
 350
 351             /* Inner loop uses 68 flops */
 352         }
 353
 354         /* End of innermost loop */
 355
 356         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 357                                               f+i_coord_offset,fshift+i_shift_offset);
 358
 359         ggid                        = gid[iidx];
 360         /* Update potential energies */
 361         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 362
 363         /* Increment number of inner iterations */
 364         inneriter                  += j_index_end - j_index_start;
 365
 366         /* Outer loop uses 8 flops */
 367     }
 368
 369     /* Increment number of outer iterations */
 370     outeriter        += nri;
 371
 372     /* Update outer/inner flops */
 373
 374     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*68);
 375 }
 376 /*
 377  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomP1P1_F_sparc64_hpc_ace_double
 378  * Electrostatics interaction: Ewald
 379  * VdW interaction:            None
 380  * Geometry:                   Particle-Particle
 381  * Calculate force/pot:        Force
 382  */
 383 void
 384 nb_kernel_ElecEwSw_VdwNone_GeomP1P1_F_sparc64_hpc_ace_double
 385                     (t_nblist                    * gmx_restrict       nlist,
 386                      rvec                        * gmx_restrict          xx,
 387                      rvec                        * gmx_restrict          ff,
 388                      t_forcerec                  * gmx_restrict          fr,
 389                      t_mdatoms                   * gmx_restrict     mdatoms,
 390                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 391                      t_nrnb                      * gmx_restrict        nrnb)
 392 {
 393     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 394      * just 0 for non-waters.
 395      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 396      * jnr indices corresponding to data put in the four positions in the SIMD register.
 397      */
 398     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 399     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 400     int              jnrA,jnrB;
 401     int              j_coord_offsetA,j_coord_offsetB;
 402     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 403     real             rcutoff_scalar;
 404     real             *shiftvec,*fshift,*x,*f;
 405     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 406     int              vdwioffset0;
 407     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 408     int              vdwjidx0A,vdwjidx0B;
 409     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 410     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 411     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 412     real             *charge;
 413     _fjsp_v2r8       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 414     real             *ewtab;
 415     _fjsp_v2r8       rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 416     real             rswitch_scalar,d_scalar;
 417     _fjsp_v2r8       itab_tmp;
 418     _fjsp_v2r8       dummy_mask,cutoff_mask;
 419     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 420     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 421     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 422
 423     x                = xx[0];
 424     f                = ff[0];
 425
 426     nri              = nlist->nri;
 427     iinr             = nlist->iinr;
 428     jindex           = nlist->jindex;
 429     jjnr             = nlist->jjnr;
 430     shiftidx         = nlist->shift;
 431     gid              = nlist->gid;
 432     shiftvec         = fr->shift_vec[0];
 433     fshift           = fr->fshift[0];
 434     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 435     charge           = mdatoms->chargeA;
 436
 437     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 438     ewtab            = fr->ic->tabq_coul_FDV0;
 439     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 440     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 441
 442     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 443     rcutoff_scalar   = fr->rcoulomb;
 444     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 445     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 446
 447     rswitch_scalar   = fr->rcoulomb_switch;
 448     rswitch          = gmx_fjsp_set1_v2r8(rswitch_scalar);
 449     /* Setup switch parameters */
 450     d_scalar         = rcutoff_scalar-rswitch_scalar;
 451     d                = gmx_fjsp_set1_v2r8(d_scalar);
 452     swV3             = gmx_fjsp_set1_v2r8(-10.0/(d_scalar*d_scalar*d_scalar));
 453     swV4             = gmx_fjsp_set1_v2r8( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 454     swV5             = gmx_fjsp_set1_v2r8( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 455     swF2             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar));
 456     swF3             = gmx_fjsp_set1_v2r8( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 457     swF4             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 458
 459     /* Avoid stupid compiler warnings */
 460     jnrA = jnrB = 0;
 461     j_coord_offsetA = 0;
 462     j_coord_offsetB = 0;
 463
 464     outeriter        = 0;
 465     inneriter        = 0;
 466
 467     /* Start outer loop over neighborlists */
 468     for(iidx=0; iidx<nri; iidx++)
 469     {
 470         /* Load shift vector for this list */
 471         i_shift_offset   = DIM*shiftidx[iidx];
 472
 473         /* Load limits for loop over neighbors */
 474         j_index_start    = jindex[iidx];
 475         j_index_end      = jindex[iidx+1];
 476
 477         /* Get outer coordinate index */
 478         inr              = iinr[iidx];
 479         i_coord_offset   = DIM*inr;
 480
 481         /* Load i particle coords and add shift vector */
 482         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 483
 484         fix0             = _fjsp_setzero_v2r8();
 485         fiy0             = _fjsp_setzero_v2r8();
 486         fiz0             = _fjsp_setzero_v2r8();
 487
 488         /* Load parameters for i particles */
 489         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 490
 491         /* Start inner kernel loop */
 492         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 493         {
 494
 495             /* Get j neighbor index, and coordinate index */
 496             jnrA             = jjnr[jidx];
 497             jnrB             = jjnr[jidx+1];
 498             j_coord_offsetA  = DIM*jnrA;
 499             j_coord_offsetB  = DIM*jnrB;
 500
 501             /* load j atom coordinates */
 502             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 503                                               &jx0,&jy0,&jz0);
 504
 505             /* Calculate displacement vector */
 506             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 507             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 508             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 509
 510             /* Calculate squared distance and things based on it */
 511             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 512
 513             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 514
 515             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 516
 517             /* Load parameters for j particles */
 518             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 519
 520             /**************************
 521              * CALCULATE INTERACTIONS *
 522              **************************/
 523
 524             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 525             {
 526
 527             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 528
 529             /* Compute parameters for interactions between i and j atoms */
 530             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 531
 532             /* EWALD ELECTROSTATICS */
 533
 534             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 535             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 536             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 537             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 538             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 539
 540             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 541             ewtabD           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
 542             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 543             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 544             ewtabFn          = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
 545             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 546             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 547             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 548             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 549             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 550
 551             d                = _fjsp_sub_v2r8(r00,rswitch);
 552             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 553             d2               = _fjsp_mul_v2r8(d,d);
 554             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 555
 556             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 557
 558             /* Evaluate switch function */
 559             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 560             felec            = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(velec,dsw)) );
 561             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 562
 563             fscal            = felec;
 564
 565             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 566
 567             /* Update vectorial force */
 568             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 569             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 570             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 571
 572             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 573
 574             }
 575
 576             /* Inner loop uses 65 flops */
 577         }
 578
 579         if(jidx<j_index_end)
 580         {
 581
 582             jnrA             = jjnr[jidx];
 583             j_coord_offsetA  = DIM*jnrA;
 584
 585             /* load j atom coordinates */
 586             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 587                                               &jx0,&jy0,&jz0);
 588
 589             /* Calculate displacement vector */
 590             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 591             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 592             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 593
 594             /* Calculate squared distance and things based on it */
 595             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 596
 597             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 598
 599             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 600
 601             /* Load parameters for j particles */
 602             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 603
 604             /**************************
 605              * CALCULATE INTERACTIONS *
 606              **************************/
 607
 608             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 609             {
 610
 611             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 612
 613             /* Compute parameters for interactions between i and j atoms */
 614             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 615
 616             /* EWALD ELECTROSTATICS */
 617
 618             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 619             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 620             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 621             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 622             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 623
 624             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 625             ewtabD           = _fjsp_setzero_v2r8();
 626             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 627             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 628             ewtabFn          = _fjsp_setzero_v2r8();
 629             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 630             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 631             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 632             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 633             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 634
 635             d                = _fjsp_sub_v2r8(r00,rswitch);
 636             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 637             d2               = _fjsp_mul_v2r8(d,d);
 638             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 639
 640             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 641
 642             /* Evaluate switch function */
 643             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 644             felec            = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(velec,dsw)) );
 645             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 646
 647             fscal            = felec;
 648
 649             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 650
 651             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 652
 653             /* Update vectorial force */
 654             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 655             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 656             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 657
 658             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 659
 660             }
 661
 662             /* Inner loop uses 65 flops */
 663         }
 664
 665         /* End of innermost loop */
 666
 667         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 668                                               f+i_coord_offset,fshift+i_shift_offset);
 669
 670         /* Increment number of inner iterations */
 671         inneriter                  += j_index_end - j_index_start;
 672
 673         /* Outer loop uses 7 flops */
 674     }
 675
 676     /* Increment number of outer iterations */
 677     outeriter        += nri;
 678
 679     /* Update outer/inner flops */
 680
 681     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*65);
 682 }