src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecGB_VdwLJ_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 "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "kernelutil_sparc64_hpc_ace_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwLJ_GeomP1P1_VF_sparc64_hpc_ace_double
  51  * Electrostatics interaction: GeneralizedBorn
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecGB_VdwLJ_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       vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,twogbeps,dvdatmp;
  87     _fjsp_v2r8       minushalf = gmx_fjsp_set1_v2r8(-0.5);
  88     real             *invsqrta,*dvda,*gbtab;
  89     int              nvdwtype;
  90     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  91     int              *vdwtype;
  92     real             *vdwparam;
  93     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
  94     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
  95     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  96     real             *vftab;
  97     _fjsp_v2r8       itab_tmp;
  98     _fjsp_v2r8       dummy_mask,cutoff_mask;
  99     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 100     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 101     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 102
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 115     charge           = mdatoms->chargeA;
 116     nvdwtype         = fr->ntype;
 117     vdwparam         = fr->nbfp;
 118     vdwtype          = mdatoms->typeA;
 119
 120     invsqrta         = fr->invsqrta;
 121     dvda             = fr->dvda;
 122     gbtabscale       = gmx_fjsp_set1_v2r8(fr->gbtab.scale);
 123     gbtab            = fr->gbtab.data;
 124     gbinvepsdiff     = gmx_fjsp_set1_v2r8((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 125
 126     /* Avoid stupid compiler warnings */
 127     jnrA = jnrB = 0;
 128     j_coord_offsetA = 0;
 129     j_coord_offsetB = 0;
 130
 131     outeriter        = 0;
 132     inneriter        = 0;
 133
 134     /* Start outer loop over neighborlists */
 135     for(iidx=0; iidx<nri; iidx++)
 136     {
 137         /* Load shift vector for this list */
 138         i_shift_offset   = DIM*shiftidx[iidx];
 139
 140         /* Load limits for loop over neighbors */
 141         j_index_start    = jindex[iidx];
 142         j_index_end      = jindex[iidx+1];
 143
 144         /* Get outer coordinate index */
 145         inr              = iinr[iidx];
 146         i_coord_offset   = DIM*inr;
 147
 148         /* Load i particle coords and add shift vector */
 149         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 150
 151         fix0             = _fjsp_setzero_v2r8();
 152         fiy0             = _fjsp_setzero_v2r8();
 153         fiz0             = _fjsp_setzero_v2r8();
 154
 155         /* Load parameters for i particles */
 156         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 157         isai0            = gmx_fjsp_load1_v2r8(invsqrta+inr+0);
 158         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 159
 160         /* Reset potential sums */
 161         velecsum         = _fjsp_setzero_v2r8();
 162         vgbsum           = _fjsp_setzero_v2r8();
 163         vvdwsum          = _fjsp_setzero_v2r8();
 164         dvdasum          = _fjsp_setzero_v2r8();
 165
 166         /* Start inner kernel loop */
 167         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 168         {
 169
 170             /* Get j neighbor index, and coordinate index */
 171             jnrA             = jjnr[jidx];
 172             jnrB             = jjnr[jidx+1];
 173             j_coord_offsetA  = DIM*jnrA;
 174             j_coord_offsetB  = DIM*jnrB;
 175
 176             /* load j atom coordinates */
 177             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 178                                               &jx0,&jy0,&jz0);
 179
 180             /* Calculate displacement vector */
 181             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 182             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 183             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 184
 185             /* Calculate squared distance and things based on it */
 186             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 187
 188             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 189
 190             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 191
 192             /* Load parameters for j particles */
 193             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 194             isaj0            = gmx_fjsp_load_2real_swizzle_v2r8(invsqrta+jnrA+0,invsqrta+jnrB+0);
 195             vdwjidx0A        = 2*vdwtype[jnrA+0];
 196             vdwjidx0B        = 2*vdwtype[jnrB+0];
 197
 198             /**************************
 199              * CALCULATE INTERACTIONS *
 200              **************************/
 201
 202             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 203
 204             /* Compute parameters for interactions between i and j atoms */
 205             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 206             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 207                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 208
 209             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 210             isaprod          = _fjsp_mul_v2r8(isai0,isaj0);
 211             gbqqfactor       = _fjsp_neg_v2r8(_fjsp_mul_v2r8(qq00,_fjsp_mul_v2r8(isaprod,gbinvepsdiff)));
 212             gbscale          = _fjsp_mul_v2r8(isaprod,gbtabscale);
 213
 214             /* Calculate generalized born table index - this is a separate table from the normal one,
 215              * but we use the same procedure by multiplying r with scale and truncating to integer.
 216              */
 217             rt               = _fjsp_mul_v2r8(r00,gbscale);
 218             itab_tmp         = _fjsp_dtox_v2r8(rt);
 219             gbeps            = _fjsp_sub_v2r8(rt,_fjsp_xtod_v2r8(itab_tmp));
 220             _fjsp_store_v2r8(&gbconv.simd,itab_tmp);
 221
 222             Y                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] );
 223             F                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] );
 224             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 225             G                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] +2);
 226             H                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] +2);
 227             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 228             Fp               = _fjsp_madd_v2r8(gbeps,_fjsp_madd_v2r8(gbeps,H,G),F);
 229             VV               = _fjsp_madd_v2r8(gbeps,Fp,Y);
 230             vgb              = _fjsp_mul_v2r8(gbqqfactor,VV);
 231
 232             twogbeps         = _fjsp_add_v2r8(gbeps,gbeps);
 233             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twogbeps,H,G),gbeps,Fp);
 234             fgb              = _fjsp_mul_v2r8(gbqqfactor,_fjsp_mul_v2r8(FF,gbscale));
 235             dvdatmp          = _fjsp_mul_v2r8(minushalf,_fjsp_madd_v2r8(fgb,r00,vgb));
 236             dvdasum          = _fjsp_add_v2r8(dvdasum,dvdatmp);
 237             gmx_fjsp_increment_2real_swizzle_v2r8(dvda+jnrA,dvda+jnrB,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj0,isaj0)));
 238             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 239             felec            = _fjsp_mul_v2r8(_fjsp_msub_v2r8(velec,rinv00,fgb),rinv00);
 240
 241             /* LENNARD-JONES DISPERSION/REPULSION */
 242
 243             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 244             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 245             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 246             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 247             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 248
 249             /* Update potential sum for this i atom from the interaction with this j atom. */
 250             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 251             vgbsum           = _fjsp_add_v2r8(vgbsum,vgb);
 252             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 253
 254             fscal            = _fjsp_add_v2r8(felec,fvdw);
 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             /* Inner loop uses 74 flops */
 264         }
 265
 266         if(jidx<j_index_end)
 267         {
 268
 269             jnrA             = jjnr[jidx];
 270             j_coord_offsetA  = DIM*jnrA;
 271
 272             /* load j atom coordinates */
 273             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 274                                               &jx0,&jy0,&jz0);
 275
 276             /* Calculate displacement vector */
 277             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 278             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 279             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 280
 281             /* Calculate squared distance and things based on it */
 282             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 283
 284             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 285
 286             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 287
 288             /* Load parameters for j particles */
 289             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 290             isaj0            = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),invsqrta+jnrA+0);
 291             vdwjidx0A        = 2*vdwtype[jnrA+0];
 292
 293             /**************************
 294              * CALCULATE INTERACTIONS *
 295              **************************/
 296
 297             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 298
 299             /* Compute parameters for interactions between i and j atoms */
 300             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 301             gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 302
 303             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 304             isaprod          = _fjsp_mul_v2r8(isai0,isaj0);
 305             gbqqfactor       = _fjsp_neg_v2r8(_fjsp_mul_v2r8(qq00,_fjsp_mul_v2r8(isaprod,gbinvepsdiff)));
 306             gbscale          = _fjsp_mul_v2r8(isaprod,gbtabscale);
 307
 308             /* Calculate generalized born table index - this is a separate table from the normal one,
 309              * but we use the same procedure by multiplying r with scale and truncating to integer.
 310              */
 311             rt               = _fjsp_mul_v2r8(r00,gbscale);
 312             itab_tmp         = _fjsp_dtox_v2r8(rt);
 313             gbeps            = _fjsp_sub_v2r8(rt,_fjsp_xtod_v2r8(itab_tmp));
 314             _fjsp_store_v2r8(&gbconv.simd,itab_tmp);
 315
 316             Y                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] );
 317             F                = _fjsp_setzero_v2r8();
 318             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 319             G                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] +2);
 320             H                = _fjsp_setzero_v2r8();
 321             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 322             Fp               = _fjsp_madd_v2r8(gbeps,_fjsp_madd_v2r8(gbeps,H,G),F);
 323             VV               = _fjsp_madd_v2r8(gbeps,Fp,Y);
 324             vgb              = _fjsp_mul_v2r8(gbqqfactor,VV);
 325
 326             twogbeps         = _fjsp_add_v2r8(gbeps,gbeps);
 327             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twogbeps,H,G),gbeps,Fp);
 328             fgb              = _fjsp_mul_v2r8(gbqqfactor,_fjsp_mul_v2r8(FF,gbscale));
 329             dvdatmp          = _fjsp_mul_v2r8(minushalf,_fjsp_madd_v2r8(fgb,r00,vgb));
 330             dvdasum          = _fjsp_add_v2r8(dvdasum,dvdatmp);
 331             gmx_fjsp_increment_1real_v2r8(dvda+jnrA,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj0,isaj0)));
 332             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 333             felec            = _fjsp_mul_v2r8(_fjsp_msub_v2r8(velec,rinv00,fgb),rinv00);
 334
 335             /* LENNARD-JONES DISPERSION/REPULSION */
 336
 337             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 338             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 339             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 340             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 341             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 342
 343             /* Update potential sum for this i atom from the interaction with this j atom. */
 344             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 345             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 346             vgb              = _fjsp_unpacklo_v2r8(vgb,_fjsp_setzero_v2r8());
 347             vgbsum           = _fjsp_add_v2r8(vgbsum,vgb);
 348             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 349             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 350
 351             fscal            = _fjsp_add_v2r8(felec,fvdw);
 352
 353             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 354
 355             /* Update vectorial force */
 356             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 357             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 358             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 359
 360             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 361
 362             /* Inner loop uses 74 flops */
 363         }
 364
 365         /* End of innermost loop */
 366
 367         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 368                                               f+i_coord_offset,fshift+i_shift_offset);
 369
 370         ggid                        = gid[iidx];
 371         /* Update potential energies */
 372         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 373         gmx_fjsp_update_1pot_v2r8(vgbsum,kernel_data->energygrp_polarization+ggid);
 374         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 375         dvdasum = _fjsp_mul_v2r8(dvdasum, _fjsp_mul_v2r8(isai0,isai0));
 376         gmx_fjsp_update_1pot_v2r8(dvdasum,dvda+inr);
 377
 378         /* Increment number of inner iterations */
 379         inneriter                  += j_index_end - j_index_start;
 380
 381         /* Outer loop uses 10 flops */
 382     }
 383
 384     /* Increment number of outer iterations */
 385     outeriter        += nri;
 386
 387     /* Update outer/inner flops */
 388
 389     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*10 + inneriter*74);
 390 }
 391 /*
 392  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwLJ_GeomP1P1_F_sparc64_hpc_ace_double
 393  * Electrostatics interaction: GeneralizedBorn
 394  * VdW interaction:            LennardJones
 395  * Geometry:                   Particle-Particle
 396  * Calculate force/pot:        Force
 397  */
 398 void
 399 nb_kernel_ElecGB_VdwLJ_GeomP1P1_F_sparc64_hpc_ace_double
 400                     (t_nblist                    * gmx_restrict       nlist,
 401                      rvec                        * gmx_restrict          xx,
 402                      rvec                        * gmx_restrict          ff,
 403                      t_forcerec                  * gmx_restrict          fr,
 404                      t_mdatoms                   * gmx_restrict     mdatoms,
 405                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 406                      t_nrnb                      * gmx_restrict        nrnb)
 407 {
 408     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 409      * just 0 for non-waters.
 410      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 411      * jnr indices corresponding to data put in the four positions in the SIMD register.
 412      */
 413     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 414     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 415     int              jnrA,jnrB;
 416     int              j_coord_offsetA,j_coord_offsetB;
 417     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 418     real             rcutoff_scalar;
 419     real             *shiftvec,*fshift,*x,*f;
 420     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 421     int              vdwioffset0;
 422     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 423     int              vdwjidx0A,vdwjidx0B;
 424     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 425     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 426     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 427     real             *charge;
 428     _fjsp_v2r8       vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,twogbeps,dvdatmp;
 429     _fjsp_v2r8       minushalf = gmx_fjsp_set1_v2r8(-0.5);
 430     real             *invsqrta,*dvda,*gbtab;
 431     int              nvdwtype;
 432     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 433     int              *vdwtype;
 434     real             *vdwparam;
 435     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 436     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 437     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 438     real             *vftab;
 439     _fjsp_v2r8       itab_tmp;
 440     _fjsp_v2r8       dummy_mask,cutoff_mask;
 441     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 442     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 443     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 444
 445     x                = xx[0];
 446     f                = ff[0];
 447
 448     nri              = nlist->nri;
 449     iinr             = nlist->iinr;
 450     jindex           = nlist->jindex;
 451     jjnr             = nlist->jjnr;
 452     shiftidx         = nlist->shift;
 453     gid              = nlist->gid;
 454     shiftvec         = fr->shift_vec[0];
 455     fshift           = fr->fshift[0];
 456     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 457     charge           = mdatoms->chargeA;
 458     nvdwtype         = fr->ntype;
 459     vdwparam         = fr->nbfp;
 460     vdwtype          = mdatoms->typeA;
 461
 462     invsqrta         = fr->invsqrta;
 463     dvda             = fr->dvda;
 464     gbtabscale       = gmx_fjsp_set1_v2r8(fr->gbtab.scale);
 465     gbtab            = fr->gbtab.data;
 466     gbinvepsdiff     = gmx_fjsp_set1_v2r8((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 467
 468     /* Avoid stupid compiler warnings */
 469     jnrA = jnrB = 0;
 470     j_coord_offsetA = 0;
 471     j_coord_offsetB = 0;
 472
 473     outeriter        = 0;
 474     inneriter        = 0;
 475
 476     /* Start outer loop over neighborlists */
 477     for(iidx=0; iidx<nri; iidx++)
 478     {
 479         /* Load shift vector for this list */
 480         i_shift_offset   = DIM*shiftidx[iidx];
 481
 482         /* Load limits for loop over neighbors */
 483         j_index_start    = jindex[iidx];
 484         j_index_end      = jindex[iidx+1];
 485
 486         /* Get outer coordinate index */
 487         inr              = iinr[iidx];
 488         i_coord_offset   = DIM*inr;
 489
 490         /* Load i particle coords and add shift vector */
 491         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 492
 493         fix0             = _fjsp_setzero_v2r8();
 494         fiy0             = _fjsp_setzero_v2r8();
 495         fiz0             = _fjsp_setzero_v2r8();
 496
 497         /* Load parameters for i particles */
 498         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 499         isai0            = gmx_fjsp_load1_v2r8(invsqrta+inr+0);
 500         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 501
 502         dvdasum          = _fjsp_setzero_v2r8();
 503
 504         /* Start inner kernel loop */
 505         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 506         {
 507
 508             /* Get j neighbor index, and coordinate index */
 509             jnrA             = jjnr[jidx];
 510             jnrB             = jjnr[jidx+1];
 511             j_coord_offsetA  = DIM*jnrA;
 512             j_coord_offsetB  = DIM*jnrB;
 513
 514             /* load j atom coordinates */
 515             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 516                                               &jx0,&jy0,&jz0);
 517
 518             /* Calculate displacement vector */
 519             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 520             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 521             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 522
 523             /* Calculate squared distance and things based on it */
 524             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 525
 526             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 527
 528             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 529
 530             /* Load parameters for j particles */
 531             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 532             isaj0            = gmx_fjsp_load_2real_swizzle_v2r8(invsqrta+jnrA+0,invsqrta+jnrB+0);
 533             vdwjidx0A        = 2*vdwtype[jnrA+0];
 534             vdwjidx0B        = 2*vdwtype[jnrB+0];
 535
 536             /**************************
 537              * CALCULATE INTERACTIONS *
 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             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 548             isaprod          = _fjsp_mul_v2r8(isai0,isaj0);
 549             gbqqfactor       = _fjsp_neg_v2r8(_fjsp_mul_v2r8(qq00,_fjsp_mul_v2r8(isaprod,gbinvepsdiff)));
 550             gbscale          = _fjsp_mul_v2r8(isaprod,gbtabscale);
 551
 552             /* Calculate generalized born table index - this is a separate table from the normal one,
 553              * but we use the same procedure by multiplying r with scale and truncating to integer.
 554              */
 555             rt               = _fjsp_mul_v2r8(r00,gbscale);
 556             itab_tmp         = _fjsp_dtox_v2r8(rt);
 557             gbeps            = _fjsp_sub_v2r8(rt,_fjsp_xtod_v2r8(itab_tmp));
 558             _fjsp_store_v2r8(&gbconv.simd,itab_tmp);
 559
 560             Y                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] );
 561             F                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] );
 562             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 563             G                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] +2);
 564             H                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] +2);
 565             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 566             Fp               = _fjsp_madd_v2r8(gbeps,_fjsp_madd_v2r8(gbeps,H,G),F);
 567             VV               = _fjsp_madd_v2r8(gbeps,Fp,Y);
 568             vgb              = _fjsp_mul_v2r8(gbqqfactor,VV);
 569
 570             twogbeps         = _fjsp_add_v2r8(gbeps,gbeps);
 571             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twogbeps,H,G),gbeps,Fp);
 572             fgb              = _fjsp_mul_v2r8(gbqqfactor,_fjsp_mul_v2r8(FF,gbscale));
 573             dvdatmp          = _fjsp_mul_v2r8(minushalf,_fjsp_madd_v2r8(fgb,r00,vgb));
 574             dvdasum          = _fjsp_add_v2r8(dvdasum,dvdatmp);
 575             gmx_fjsp_increment_2real_swizzle_v2r8(dvda+jnrA,dvda+jnrB,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj0,isaj0)));
 576             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 577             felec            = _fjsp_mul_v2r8(_fjsp_msub_v2r8(velec,rinv00,fgb),rinv00);
 578
 579             /* LENNARD-JONES DISPERSION/REPULSION */
 580
 581             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 582             fvdw             = _fjsp_mul_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,c6_00),_fjsp_mul_v2r8(rinvsix,rinvsq00));
 583
 584             fscal            = _fjsp_add_v2r8(felec,fvdw);
 585
 586             /* Update vectorial force */
 587             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 588             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 589             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 590
 591             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 592
 593             /* Inner loop uses 67 flops */
 594         }
 595
 596         if(jidx<j_index_end)
 597         {
 598
 599             jnrA             = jjnr[jidx];
 600             j_coord_offsetA  = DIM*jnrA;
 601
 602             /* load j atom coordinates */
 603             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 604                                               &jx0,&jy0,&jz0);
 605
 606             /* Calculate displacement vector */
 607             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 608             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 609             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 610
 611             /* Calculate squared distance and things based on it */
 612             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 613
 614             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 615
 616             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 617
 618             /* Load parameters for j particles */
 619             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 620             isaj0            = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),invsqrta+jnrA+0);
 621             vdwjidx0A        = 2*vdwtype[jnrA+0];
 622
 623             /**************************
 624              * CALCULATE INTERACTIONS *
 625              **************************/
 626
 627             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 628
 629             /* Compute parameters for interactions between i and j atoms */
 630             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 631             gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 632
 633             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 634             isaprod          = _fjsp_mul_v2r8(isai0,isaj0);
 635             gbqqfactor       = _fjsp_neg_v2r8(_fjsp_mul_v2r8(qq00,_fjsp_mul_v2r8(isaprod,gbinvepsdiff)));
 636             gbscale          = _fjsp_mul_v2r8(isaprod,gbtabscale);
 637
 638             /* Calculate generalized born table index - this is a separate table from the normal one,
 639              * but we use the same procedure by multiplying r with scale and truncating to integer.
 640              */
 641             rt               = _fjsp_mul_v2r8(r00,gbscale);
 642             itab_tmp         = _fjsp_dtox_v2r8(rt);
 643             gbeps            = _fjsp_sub_v2r8(rt,_fjsp_xtod_v2r8(itab_tmp));
 644             _fjsp_store_v2r8(&gbconv.simd,itab_tmp);
 645
 646             Y                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] );
 647             F                = _fjsp_setzero_v2r8();
 648             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 649             G                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] +2);
 650             H                = _fjsp_setzero_v2r8();
 651             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 652             Fp               = _fjsp_madd_v2r8(gbeps,_fjsp_madd_v2r8(gbeps,H,G),F);
 653             VV               = _fjsp_madd_v2r8(gbeps,Fp,Y);
 654             vgb              = _fjsp_mul_v2r8(gbqqfactor,VV);
 655
 656             twogbeps         = _fjsp_add_v2r8(gbeps,gbeps);
 657             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twogbeps,H,G),gbeps,Fp);
 658             fgb              = _fjsp_mul_v2r8(gbqqfactor,_fjsp_mul_v2r8(FF,gbscale));
 659             dvdatmp          = _fjsp_mul_v2r8(minushalf,_fjsp_madd_v2r8(fgb,r00,vgb));
 660             dvdasum          = _fjsp_add_v2r8(dvdasum,dvdatmp);
 661             gmx_fjsp_increment_1real_v2r8(dvda+jnrA,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj0,isaj0)));
 662             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 663             felec            = _fjsp_mul_v2r8(_fjsp_msub_v2r8(velec,rinv00,fgb),rinv00);
 664
 665             /* LENNARD-JONES DISPERSION/REPULSION */
 666
 667             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 668             fvdw             = _fjsp_mul_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,c6_00),_fjsp_mul_v2r8(rinvsix,rinvsq00));
 669
 670             fscal            = _fjsp_add_v2r8(felec,fvdw);
 671
 672             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 673
 674             /* Update vectorial force */
 675             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 676             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 677             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 678
 679             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 680
 681             /* Inner loop uses 67 flops */
 682         }
 683
 684         /* End of innermost loop */
 685
 686         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 687                                               f+i_coord_offset,fshift+i_shift_offset);
 688
 689         dvdasum = _fjsp_mul_v2r8(dvdasum, _fjsp_mul_v2r8(isai0,isai0));
 690         gmx_fjsp_update_1pot_v2r8(dvdasum,dvda+inr);
 691
 692         /* Increment number of inner iterations */
 693         inneriter                  += j_index_end - j_index_start;
 694
 695         /* Outer loop uses 7 flops */
 696     }
 697
 698     /* Increment number of outer iterations */
 699     outeriter        += nri;
 700
 701     /* Update outer/inner flops */
 702
 703     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*67);
 704 }