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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "kernelutil_sparc64_hpc_ace_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomP1P1_VF_sparc64_hpc_ace_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            None
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSw_VdwNone_GeomP1P1_VF_sparc64_hpc_ace_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      t_forcerec                  * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwjidx0A,vdwjidx0B;
  82     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  83     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  84     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
  85     real             *charge;
  86     _fjsp_v2r8       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  87     real             *ewtab;
  88     _fjsp_v2r8       rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  89     real             rswitch_scalar,d_scalar;
  90     _fjsp_v2r8       itab_tmp;
  91     _fjsp_v2r8       dummy_mask,cutoff_mask;
  92     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
  93     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
  94     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
  95
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109
 110     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 111     ewtab            = fr->ic->tabq_coul_FDV0;
 112     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 113     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 114
 115     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 116     rcutoff_scalar   = fr->rcoulomb;
 117     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 118     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 119
 120     rswitch_scalar   = fr->rcoulomb_switch;
 121     rswitch          = gmx_fjsp_set1_v2r8(rswitch_scalar);
 122     /* Setup switch parameters */
 123     d_scalar         = rcutoff_scalar-rswitch_scalar;
 124     d                = gmx_fjsp_set1_v2r8(d_scalar);
 125     swV3             = gmx_fjsp_set1_v2r8(-10.0/(d_scalar*d_scalar*d_scalar));
 126     swV4             = gmx_fjsp_set1_v2r8( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 127     swV5             = gmx_fjsp_set1_v2r8( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 128     swF2             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar));
 129     swF3             = gmx_fjsp_set1_v2r8( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 130     swF4             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 131
 132     /* Avoid stupid compiler warnings */
 133     jnrA = jnrB = 0;
 134     j_coord_offsetA = 0;
 135     j_coord_offsetB = 0;
 136
 137     outeriter        = 0;
 138     inneriter        = 0;
 139
 140     /* Start outer loop over neighborlists */
 141     for(iidx=0; iidx<nri; iidx++)
 142     {
 143         /* Load shift vector for this list */
 144         i_shift_offset   = DIM*shiftidx[iidx];
 145
 146         /* Load limits for loop over neighbors */
 147         j_index_start    = jindex[iidx];
 148         j_index_end      = jindex[iidx+1];
 149
 150         /* Get outer coordinate index */
 151         inr              = iinr[iidx];
 152         i_coord_offset   = DIM*inr;
 153
 154         /* Load i particle coords and add shift vector */
 155         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 156
 157         fix0             = _fjsp_setzero_v2r8();
 158         fiy0             = _fjsp_setzero_v2r8();
 159         fiz0             = _fjsp_setzero_v2r8();
 160
 161         /* Load parameters for i particles */
 162         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 163
 164         /* Reset potential sums */
 165         velecsum         = _fjsp_setzero_v2r8();
 166
 167         /* Start inner kernel loop */
 168         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 169         {
 170
 171             /* Get j neighbor index, and coordinate index */
 172             jnrA             = jjnr[jidx];
 173             jnrB             = jjnr[jidx+1];
 174             j_coord_offsetA  = DIM*jnrA;
 175             j_coord_offsetB  = DIM*jnrB;
 176
 177             /* load j atom coordinates */
 178             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 179                                               &jx0,&jy0,&jz0);
 180
 181             /* Calculate displacement vector */
 182             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 183             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 184             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 185
 186             /* Calculate squared distance and things based on it */
 187             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 188
 189             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 190
 191             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 192
 193             /* Load parameters for j particles */
 194             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+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             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 207
 208             /* EWALD ELECTROSTATICS */
 209
 210             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 211             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 212             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 213             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 214             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 215
 216             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 217             ewtabD           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
 218             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 219             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 220             ewtabFn          = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
 221             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 222             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 223             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 224             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 225             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 226
 227             d                = _fjsp_sub_v2r8(r00,rswitch);
 228             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 229             d2               = _fjsp_mul_v2r8(d,d);
 230             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 231
 232             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 233
 234             /* Evaluate switch function */
 235             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 236             felec            = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(velec,dsw)) );
 237             velec            = _fjsp_mul_v2r8(velec,sw);
 238             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 239
 240             /* Update potential sum for this i atom from the interaction with this j atom. */
 241             velec            = _fjsp_and_v2r8(velec,cutoff_mask);
 242             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 243
 244             fscal            = felec;
 245
 246             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 247
 248             /* Update vectorial force */
 249             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 250             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 251             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 252
 253             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 254
 255             }
 256
 257             /* Inner loop uses 68 flops */
 258         }
 259
 260         if(jidx<j_index_end)
 261         {
 262
 263             jnrA             = jjnr[jidx];
 264             j_coord_offsetA  = DIM*jnrA;
 265
 266             /* load j atom coordinates */
 267             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 268                                               &jx0,&jy0,&jz0);
 269
 270             /* Calculate displacement vector */
 271             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 272             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 273             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 274
 275             /* Calculate squared distance and things based on it */
 276             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 277
 278             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 279
 280             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 281
 282             /* Load parameters for j particles */
 283             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 284
 285             /**************************
 286              * CALCULATE INTERACTIONS *
 287              **************************/
 288
 289             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 290             {
 291
 292             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 293
 294             /* Compute parameters for interactions between i and j atoms */
 295             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 296
 297             /* EWALD ELECTROSTATICS */
 298
 299             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 300             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 301             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 302             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 303             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 304
 305             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 306             ewtabD           = _fjsp_setzero_v2r8();
 307             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 308             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 309             ewtabFn          = _fjsp_setzero_v2r8();
 310             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 311             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 312             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 313             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 314             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 315
 316             d                = _fjsp_sub_v2r8(r00,rswitch);
 317             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 318             d2               = _fjsp_mul_v2r8(d,d);
 319             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 320
 321             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 322
 323             /* Evaluate switch function */
 324             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 325             felec            = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(velec,dsw)) );
 326             velec            = _fjsp_mul_v2r8(velec,sw);
 327             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 328
 329             /* Update potential sum for this i atom from the interaction with this j atom. */
 330             velec            = _fjsp_and_v2r8(velec,cutoff_mask);
 331             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 332             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 333
 334             fscal            = felec;
 335
 336             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 337
 338             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 339
 340             /* Update vectorial force */
 341             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 342             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 343             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 344
 345             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 346
 347             }
 348
 349             /* Inner loop uses 68 flops */
 350         }
 351
 352         /* End of innermost loop */
 353
 354         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 355                                               f+i_coord_offset,fshift+i_shift_offset);
 356
 357         ggid                        = gid[iidx];
 358         /* Update potential energies */
 359         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 360
 361         /* Increment number of inner iterations */
 362         inneriter                  += j_index_end - j_index_start;
 363
 364         /* Outer loop uses 8 flops */
 365     }
 366
 367     /* Increment number of outer iterations */
 368     outeriter        += nri;
 369
 370     /* Update outer/inner flops */
 371
 372     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*68);
 373 }
 374 /*
 375  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomP1P1_F_sparc64_hpc_ace_double
 376  * Electrostatics interaction: Ewald
 377  * VdW interaction:            None
 378  * Geometry:                   Particle-Particle
 379  * Calculate force/pot:        Force
 380  */
 381 void
 382 nb_kernel_ElecEwSw_VdwNone_GeomP1P1_F_sparc64_hpc_ace_double
 383                     (t_nblist                    * gmx_restrict       nlist,
 384                      rvec                        * gmx_restrict          xx,
 385                      rvec                        * gmx_restrict          ff,
 386                      t_forcerec                  * gmx_restrict          fr,
 387                      t_mdatoms                   * gmx_restrict     mdatoms,
 388                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 389                      t_nrnb                      * gmx_restrict        nrnb)
 390 {
 391     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 392      * just 0 for non-waters.
 393      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 394      * jnr indices corresponding to data put in the four positions in the SIMD register.
 395      */
 396     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 397     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 398     int              jnrA,jnrB;
 399     int              j_coord_offsetA,j_coord_offsetB;
 400     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 401     real             rcutoff_scalar;
 402     real             *shiftvec,*fshift,*x,*f;
 403     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 404     int              vdwioffset0;
 405     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 406     int              vdwjidx0A,vdwjidx0B;
 407     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 408     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 409     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 410     real             *charge;
 411     _fjsp_v2r8       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 412     real             *ewtab;
 413     _fjsp_v2r8       rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 414     real             rswitch_scalar,d_scalar;
 415     _fjsp_v2r8       itab_tmp;
 416     _fjsp_v2r8       dummy_mask,cutoff_mask;
 417     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 418     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 419     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 420
 421     x                = xx[0];
 422     f                = ff[0];
 423
 424     nri              = nlist->nri;
 425     iinr             = nlist->iinr;
 426     jindex           = nlist->jindex;
 427     jjnr             = nlist->jjnr;
 428     shiftidx         = nlist->shift;
 429     gid              = nlist->gid;
 430     shiftvec         = fr->shift_vec[0];
 431     fshift           = fr->fshift[0];
 432     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 433     charge           = mdatoms->chargeA;
 434
 435     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 436     ewtab            = fr->ic->tabq_coul_FDV0;
 437     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 438     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 439
 440     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 441     rcutoff_scalar   = fr->rcoulomb;
 442     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 443     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 444
 445     rswitch_scalar   = fr->rcoulomb_switch;
 446     rswitch          = gmx_fjsp_set1_v2r8(rswitch_scalar);
 447     /* Setup switch parameters */
 448     d_scalar         = rcutoff_scalar-rswitch_scalar;
 449     d                = gmx_fjsp_set1_v2r8(d_scalar);
 450     swV3             = gmx_fjsp_set1_v2r8(-10.0/(d_scalar*d_scalar*d_scalar));
 451     swV4             = gmx_fjsp_set1_v2r8( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 452     swV5             = gmx_fjsp_set1_v2r8( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 453     swF2             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar));
 454     swF3             = gmx_fjsp_set1_v2r8( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 455     swF4             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 456
 457     /* Avoid stupid compiler warnings */
 458     jnrA = jnrB = 0;
 459     j_coord_offsetA = 0;
 460     j_coord_offsetB = 0;
 461
 462     outeriter        = 0;
 463     inneriter        = 0;
 464
 465     /* Start outer loop over neighborlists */
 466     for(iidx=0; iidx<nri; iidx++)
 467     {
 468         /* Load shift vector for this list */
 469         i_shift_offset   = DIM*shiftidx[iidx];
 470
 471         /* Load limits for loop over neighbors */
 472         j_index_start    = jindex[iidx];
 473         j_index_end      = jindex[iidx+1];
 474
 475         /* Get outer coordinate index */
 476         inr              = iinr[iidx];
 477         i_coord_offset   = DIM*inr;
 478
 479         /* Load i particle coords and add shift vector */
 480         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 481
 482         fix0             = _fjsp_setzero_v2r8();
 483         fiy0             = _fjsp_setzero_v2r8();
 484         fiz0             = _fjsp_setzero_v2r8();
 485
 486         /* Load parameters for i particles */
 487         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 488
 489         /* Start inner kernel loop */
 490         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 491         {
 492
 493             /* Get j neighbor index, and coordinate index */
 494             jnrA             = jjnr[jidx];
 495             jnrB             = jjnr[jidx+1];
 496             j_coord_offsetA  = DIM*jnrA;
 497             j_coord_offsetB  = DIM*jnrB;
 498
 499             /* load j atom coordinates */
 500             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 501                                               &jx0,&jy0,&jz0);
 502
 503             /* Calculate displacement vector */
 504             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 505             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 506             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 507
 508             /* Calculate squared distance and things based on it */
 509             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 510
 511             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 512
 513             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 514
 515             /* Load parameters for j particles */
 516             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 517
 518             /**************************
 519              * CALCULATE INTERACTIONS *
 520              **************************/
 521
 522             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 523             {
 524
 525             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 526
 527             /* Compute parameters for interactions between i and j atoms */
 528             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 529
 530             /* EWALD ELECTROSTATICS */
 531
 532             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 533             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 534             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 535             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 536             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 537
 538             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 539             ewtabD           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
 540             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 541             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 542             ewtabFn          = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
 543             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 544             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 545             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 546             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 547             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 548
 549             d                = _fjsp_sub_v2r8(r00,rswitch);
 550             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 551             d2               = _fjsp_mul_v2r8(d,d);
 552             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 553
 554             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 555
 556             /* Evaluate switch function */
 557             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 558             felec            = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(velec,dsw)) );
 559             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 560
 561             fscal            = felec;
 562
 563             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 564
 565             /* Update vectorial force */
 566             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 567             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 568             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 569
 570             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 571
 572             }
 573
 574             /* Inner loop uses 65 flops */
 575         }
 576
 577         if(jidx<j_index_end)
 578         {
 579
 580             jnrA             = jjnr[jidx];
 581             j_coord_offsetA  = DIM*jnrA;
 582
 583             /* load j atom coordinates */
 584             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 585                                               &jx0,&jy0,&jz0);
 586
 587             /* Calculate displacement vector */
 588             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 589             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 590             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 591
 592             /* Calculate squared distance and things based on it */
 593             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 594
 595             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 596
 597             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 598
 599             /* Load parameters for j particles */
 600             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 601
 602             /**************************
 603              * CALCULATE INTERACTIONS *
 604              **************************/
 605
 606             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 607             {
 608
 609             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 610
 611             /* Compute parameters for interactions between i and j atoms */
 612             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 613
 614             /* EWALD ELECTROSTATICS */
 615
 616             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 617             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 618             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 619             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 620             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 621
 622             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 623             ewtabD           = _fjsp_setzero_v2r8();
 624             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 625             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 626             ewtabFn          = _fjsp_setzero_v2r8();
 627             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 628             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 629             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 630             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 631             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 632
 633             d                = _fjsp_sub_v2r8(r00,rswitch);
 634             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 635             d2               = _fjsp_mul_v2r8(d,d);
 636             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 637
 638             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 639
 640             /* Evaluate switch function */
 641             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 642             felec            = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(velec,dsw)) );
 643             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 644
 645             fscal            = felec;
 646
 647             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 648
 649             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 650
 651             /* Update vectorial force */
 652             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 653             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 654             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 655
 656             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 657
 658             }
 659
 660             /* Inner loop uses 65 flops */
 661         }
 662
 663         /* End of innermost loop */
 664
 665         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 666                                               f+i_coord_offset,fshift+i_shift_offset);
 667
 668         /* Increment number of inner iterations */
 669         inneriter                  += j_index_end - j_index_start;
 670
 671         /* Outer loop uses 7 flops */
 672     }
 673
 674     /* Increment number of outer iterations */
 675     outeriter        += nri;
 676
 677     /* Update outer/inner flops */
 678
 679     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*65);
 680 }