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 "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_ElecGB_VdwLJ_GeomP1P1_VF_sparc64_hpc_ace_double
  53  * Electrostatics interaction: GeneralizedBorn
  54  * VdW interaction:            LennardJones
  55  * Geometry:                   Particle-Particle
  56  * Calculate force/pot:        PotentialAndForce
  57  */
  58 void
  59 nb_kernel_ElecGB_VdwLJ_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       vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,twogbeps,dvdatmp;
  89     _fjsp_v2r8       minushalf = gmx_fjsp_set1_v2r8(-0.5);
  90     real             *invsqrta,*dvda,*gbtab;
  91     int              nvdwtype;
  92     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  93     int              *vdwtype;
  94     real             *vdwparam;
  95     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
  96     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
  97     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  98     real             *vftab;
  99     _fjsp_v2r8       itab_tmp;
 100     _fjsp_v2r8       dummy_mask,cutoff_mask;
 101     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 102     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 103     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 104
 105     x                = xx[0];
 106     f                = ff[0];
 107
 108     nri              = nlist->nri;
 109     iinr             = nlist->iinr;
 110     jindex           = nlist->jindex;
 111     jjnr             = nlist->jjnr;
 112     shiftidx         = nlist->shift;
 113     gid              = nlist->gid;
 114     shiftvec         = fr->shift_vec[0];
 115     fshift           = fr->fshift[0];
 116     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 117     charge           = mdatoms->chargeA;
 118     nvdwtype         = fr->ntype;
 119     vdwparam         = fr->nbfp;
 120     vdwtype          = mdatoms->typeA;
 121
 122     invsqrta         = fr->invsqrta;
 123     dvda             = fr->dvda;
 124     gbtabscale       = gmx_fjsp_set1_v2r8(fr->gbtab.scale);
 125     gbtab            = fr->gbtab.data;
 126     gbinvepsdiff     = gmx_fjsp_set1_v2r8((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 127
 128     /* Avoid stupid compiler warnings */
 129     jnrA = jnrB = 0;
 130     j_coord_offsetA = 0;
 131     j_coord_offsetB = 0;
 132
 133     outeriter        = 0;
 134     inneriter        = 0;
 135
 136     /* Start outer loop over neighborlists */
 137     for(iidx=0; iidx<nri; iidx++)
 138     {
 139         /* Load shift vector for this list */
 140         i_shift_offset   = DIM*shiftidx[iidx];
 141
 142         /* Load limits for loop over neighbors */
 143         j_index_start    = jindex[iidx];
 144         j_index_end      = jindex[iidx+1];
 145
 146         /* Get outer coordinate index */
 147         inr              = iinr[iidx];
 148         i_coord_offset   = DIM*inr;
 149
 150         /* Load i particle coords and add shift vector */
 151         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 152
 153         fix0             = _fjsp_setzero_v2r8();
 154         fiy0             = _fjsp_setzero_v2r8();
 155         fiz0             = _fjsp_setzero_v2r8();
 156
 157         /* Load parameters for i particles */
 158         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 159         isai0            = gmx_fjsp_load1_v2r8(invsqrta+inr+0);
 160         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 161
 162         /* Reset potential sums */
 163         velecsum         = _fjsp_setzero_v2r8();
 164         vgbsum           = _fjsp_setzero_v2r8();
 165         vvdwsum          = _fjsp_setzero_v2r8();
 166         dvdasum          = _fjsp_setzero_v2r8();
 167
 168         /* Start inner kernel loop */
 169         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 170         {
 171
 172             /* Get j neighbor index, and coordinate index */
 173             jnrA             = jjnr[jidx];
 174             jnrB             = jjnr[jidx+1];
 175             j_coord_offsetA  = DIM*jnrA;
 176             j_coord_offsetB  = DIM*jnrB;
 177
 178             /* load j atom coordinates */
 179             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 180                                               &jx0,&jy0,&jz0);
 181
 182             /* Calculate displacement vector */
 183             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 184             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 185             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 186
 187             /* Calculate squared distance and things based on it */
 188             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 189
 190             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 191
 192             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 193
 194             /* Load parameters for j particles */
 195             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 196             isaj0            = gmx_fjsp_load_2real_swizzle_v2r8(invsqrta+jnrA+0,invsqrta+jnrB+0);
 197             vdwjidx0A        = 2*vdwtype[jnrA+0];
 198             vdwjidx0B        = 2*vdwtype[jnrB+0];
 199
 200             /**************************
 201              * CALCULATE INTERACTIONS *
 202              **************************/
 203
 204             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 205
 206             /* Compute parameters for interactions between i and j atoms */
 207             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 208             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 209                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 210
 211             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 212             isaprod          = _fjsp_mul_v2r8(isai0,isaj0);
 213             gbqqfactor       = _fjsp_neg_v2r8(_fjsp_mul_v2r8(qq00,_fjsp_mul_v2r8(isaprod,gbinvepsdiff)));
 214             gbscale          = _fjsp_mul_v2r8(isaprod,gbtabscale);
 215
 216             /* Calculate generalized born table index - this is a separate table from the normal one,
 217              * but we use the same procedure by multiplying r with scale and truncating to integer.
 218              */
 219             rt               = _fjsp_mul_v2r8(r00,gbscale);
 220             itab_tmp         = _fjsp_dtox_v2r8(rt);
 221             gbeps            = _fjsp_sub_v2r8(rt,_fjsp_xtod_v2r8(itab_tmp));
 222             _fjsp_store_v2r8(&gbconv.simd,itab_tmp);
 223
 224             Y                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] );
 225             F                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] );
 226             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 227             G                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] +2);
 228             H                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] +2);
 229             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 230             Fp               = _fjsp_madd_v2r8(gbeps,_fjsp_madd_v2r8(gbeps,H,G),F);
 231             VV               = _fjsp_madd_v2r8(gbeps,Fp,Y);
 232             vgb              = _fjsp_mul_v2r8(gbqqfactor,VV);
 233
 234             twogbeps         = _fjsp_add_v2r8(gbeps,gbeps);
 235             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twogbeps,H,G),gbeps,Fp);
 236             fgb              = _fjsp_mul_v2r8(gbqqfactor,_fjsp_mul_v2r8(FF,gbscale));
 237             dvdatmp          = _fjsp_mul_v2r8(minushalf,_fjsp_madd_v2r8(fgb,r00,vgb));
 238             dvdasum          = _fjsp_add_v2r8(dvdasum,dvdatmp);
 239             gmx_fjsp_increment_2real_swizzle_v2r8(dvda+jnrA,dvda+jnrB,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj0,isaj0)));
 240             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 241             felec            = _fjsp_mul_v2r8(_fjsp_msub_v2r8(velec,rinv00,fgb),rinv00);
 242
 243             /* LENNARD-JONES DISPERSION/REPULSION */
 244
 245             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 246             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 247             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 248             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 249             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 250
 251             /* Update potential sum for this i atom from the interaction with this j atom. */
 252             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 253             vgbsum           = _fjsp_add_v2r8(vgbsum,vgb);
 254             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 255
 256             fscal            = _fjsp_add_v2r8(felec,fvdw);
 257
 258             /* Update vectorial force */
 259             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 260             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 261             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 262
 263             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 264
 265             /* Inner loop uses 74 flops */
 266         }
 267
 268         if(jidx<j_index_end)
 269         {
 270
 271             jnrA             = jjnr[jidx];
 272             j_coord_offsetA  = DIM*jnrA;
 273
 274             /* load j atom coordinates */
 275             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 276                                               &jx0,&jy0,&jz0);
 277
 278             /* Calculate displacement vector */
 279             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 280             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 281             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 282
 283             /* Calculate squared distance and things based on it */
 284             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 285
 286             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 287
 288             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 289
 290             /* Load parameters for j particles */
 291             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 292             isaj0            = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),invsqrta+jnrA+0);
 293             vdwjidx0A        = 2*vdwtype[jnrA+0];
 294
 295             /**************************
 296              * CALCULATE INTERACTIONS *
 297              **************************/
 298
 299             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 300
 301             /* Compute parameters for interactions between i and j atoms */
 302             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 303             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 304                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 305
 306             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 307             isaprod          = _fjsp_mul_v2r8(isai0,isaj0);
 308             gbqqfactor       = _fjsp_neg_v2r8(_fjsp_mul_v2r8(qq00,_fjsp_mul_v2r8(isaprod,gbinvepsdiff)));
 309             gbscale          = _fjsp_mul_v2r8(isaprod,gbtabscale);
 310
 311             /* Calculate generalized born table index - this is a separate table from the normal one,
 312              * but we use the same procedure by multiplying r with scale and truncating to integer.
 313              */
 314             rt               = _fjsp_mul_v2r8(r00,gbscale);
 315             itab_tmp         = _fjsp_dtox_v2r8(rt);
 316             gbeps            = _fjsp_sub_v2r8(rt,_fjsp_xtod_v2r8(itab_tmp));
 317             _fjsp_store_v2r8(&gbconv.simd,itab_tmp);
 318
 319             Y                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] );
 320             F                = _fjsp_setzero_v2r8();
 321             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 322             G                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] +2);
 323             H                = _fjsp_setzero_v2r8();
 324             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 325             Fp               = _fjsp_madd_v2r8(gbeps,_fjsp_madd_v2r8(gbeps,H,G),F);
 326             VV               = _fjsp_madd_v2r8(gbeps,Fp,Y);
 327             vgb              = _fjsp_mul_v2r8(gbqqfactor,VV);
 328
 329             twogbeps         = _fjsp_add_v2r8(gbeps,gbeps);
 330             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twogbeps,H,G),gbeps,Fp);
 331             fgb              = _fjsp_mul_v2r8(gbqqfactor,_fjsp_mul_v2r8(FF,gbscale));
 332             dvdatmp          = _fjsp_mul_v2r8(minushalf,_fjsp_madd_v2r8(fgb,r00,vgb));
 333             dvdasum          = _fjsp_add_v2r8(dvdasum,dvdatmp);
 334             gmx_fjsp_increment_1real_v2r8(dvda+jnrA,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj0,isaj0)));
 335             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 336             felec            = _fjsp_mul_v2r8(_fjsp_msub_v2r8(velec,rinv00,fgb),rinv00);
 337
 338             /* LENNARD-JONES DISPERSION/REPULSION */
 339
 340             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 341             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 342             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 343             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 344             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 345
 346             /* Update potential sum for this i atom from the interaction with this j atom. */
 347             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 348             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 349             vgb              = _fjsp_unpacklo_v2r8(vgb,_fjsp_setzero_v2r8());
 350             vgbsum           = _fjsp_add_v2r8(vgbsum,vgb);
 351             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 352             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 353
 354             fscal            = _fjsp_add_v2r8(felec,fvdw);
 355
 356             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 357
 358             /* Update vectorial force */
 359             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 360             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 361             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 362
 363             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 364
 365             /* Inner loop uses 74 flops */
 366         }
 367
 368         /* End of innermost loop */
 369
 370         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 371                                               f+i_coord_offset,fshift+i_shift_offset);
 372
 373         ggid                        = gid[iidx];
 374         /* Update potential energies */
 375         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 376         gmx_fjsp_update_1pot_v2r8(vgbsum,kernel_data->energygrp_polarization+ggid);
 377         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 378         dvdasum = _fjsp_mul_v2r8(dvdasum, _fjsp_mul_v2r8(isai0,isai0));
 379         gmx_fjsp_update_1pot_v2r8(dvdasum,dvda+inr);
 380
 381         /* Increment number of inner iterations */
 382         inneriter                  += j_index_end - j_index_start;
 383
 384         /* Outer loop uses 10 flops */
 385     }
 386
 387     /* Increment number of outer iterations */
 388     outeriter        += nri;
 389
 390     /* Update outer/inner flops */
 391
 392     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*10 + inneriter*74);
 393 }
 394 /*
 395  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwLJ_GeomP1P1_F_sparc64_hpc_ace_double
 396  * Electrostatics interaction: GeneralizedBorn
 397  * VdW interaction:            LennardJones
 398  * Geometry:                   Particle-Particle
 399  * Calculate force/pot:        Force
 400  */
 401 void
 402 nb_kernel_ElecGB_VdwLJ_GeomP1P1_F_sparc64_hpc_ace_double
 403                     (t_nblist                    * gmx_restrict       nlist,
 404                      rvec                        * gmx_restrict          xx,
 405                      rvec                        * gmx_restrict          ff,
 406                      t_forcerec                  * gmx_restrict          fr,
 407                      t_mdatoms                   * gmx_restrict     mdatoms,
 408                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 409                      t_nrnb                      * gmx_restrict        nrnb)
 410 {
 411     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 412      * just 0 for non-waters.
 413      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 414      * jnr indices corresponding to data put in the four positions in the SIMD register.
 415      */
 416     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 417     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 418     int              jnrA,jnrB;
 419     int              j_coord_offsetA,j_coord_offsetB;
 420     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 421     real             rcutoff_scalar;
 422     real             *shiftvec,*fshift,*x,*f;
 423     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 424     int              vdwioffset0;
 425     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 426     int              vdwjidx0A,vdwjidx0B;
 427     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 428     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 429     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 430     real             *charge;
 431     _fjsp_v2r8       vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,twogbeps,dvdatmp;
 432     _fjsp_v2r8       minushalf = gmx_fjsp_set1_v2r8(-0.5);
 433     real             *invsqrta,*dvda,*gbtab;
 434     int              nvdwtype;
 435     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 436     int              *vdwtype;
 437     real             *vdwparam;
 438     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 439     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 440     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 441     real             *vftab;
 442     _fjsp_v2r8       itab_tmp;
 443     _fjsp_v2r8       dummy_mask,cutoff_mask;
 444     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 445     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 446     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 447
 448     x                = xx[0];
 449     f                = ff[0];
 450
 451     nri              = nlist->nri;
 452     iinr             = nlist->iinr;
 453     jindex           = nlist->jindex;
 454     jjnr             = nlist->jjnr;
 455     shiftidx         = nlist->shift;
 456     gid              = nlist->gid;
 457     shiftvec         = fr->shift_vec[0];
 458     fshift           = fr->fshift[0];
 459     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 460     charge           = mdatoms->chargeA;
 461     nvdwtype         = fr->ntype;
 462     vdwparam         = fr->nbfp;
 463     vdwtype          = mdatoms->typeA;
 464
 465     invsqrta         = fr->invsqrta;
 466     dvda             = fr->dvda;
 467     gbtabscale       = gmx_fjsp_set1_v2r8(fr->gbtab.scale);
 468     gbtab            = fr->gbtab.data;
 469     gbinvepsdiff     = gmx_fjsp_set1_v2r8((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 470
 471     /* Avoid stupid compiler warnings */
 472     jnrA = jnrB = 0;
 473     j_coord_offsetA = 0;
 474     j_coord_offsetB = 0;
 475
 476     outeriter        = 0;
 477     inneriter        = 0;
 478
 479     /* Start outer loop over neighborlists */
 480     for(iidx=0; iidx<nri; iidx++)
 481     {
 482         /* Load shift vector for this list */
 483         i_shift_offset   = DIM*shiftidx[iidx];
 484
 485         /* Load limits for loop over neighbors */
 486         j_index_start    = jindex[iidx];
 487         j_index_end      = jindex[iidx+1];
 488
 489         /* Get outer coordinate index */
 490         inr              = iinr[iidx];
 491         i_coord_offset   = DIM*inr;
 492
 493         /* Load i particle coords and add shift vector */
 494         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 495
 496         fix0             = _fjsp_setzero_v2r8();
 497         fiy0             = _fjsp_setzero_v2r8();
 498         fiz0             = _fjsp_setzero_v2r8();
 499
 500         /* Load parameters for i particles */
 501         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 502         isai0            = gmx_fjsp_load1_v2r8(invsqrta+inr+0);
 503         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 504
 505         dvdasum          = _fjsp_setzero_v2r8();
 506
 507         /* Start inner kernel loop */
 508         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 509         {
 510
 511             /* Get j neighbor index, and coordinate index */
 512             jnrA             = jjnr[jidx];
 513             jnrB             = jjnr[jidx+1];
 514             j_coord_offsetA  = DIM*jnrA;
 515             j_coord_offsetB  = DIM*jnrB;
 516
 517             /* load j atom coordinates */
 518             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 519                                               &jx0,&jy0,&jz0);
 520
 521             /* Calculate displacement vector */
 522             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 523             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 524             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 525
 526             /* Calculate squared distance and things based on it */
 527             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 528
 529             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 530
 531             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 532
 533             /* Load parameters for j particles */
 534             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 535             isaj0            = gmx_fjsp_load_2real_swizzle_v2r8(invsqrta+jnrA+0,invsqrta+jnrB+0);
 536             vdwjidx0A        = 2*vdwtype[jnrA+0];
 537             vdwjidx0B        = 2*vdwtype[jnrB+0];
 538
 539             /**************************
 540              * CALCULATE INTERACTIONS *
 541              **************************/
 542
 543             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 544
 545             /* Compute parameters for interactions between i and j atoms */
 546             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 547             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 548                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 549
 550             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 551             isaprod          = _fjsp_mul_v2r8(isai0,isaj0);
 552             gbqqfactor       = _fjsp_neg_v2r8(_fjsp_mul_v2r8(qq00,_fjsp_mul_v2r8(isaprod,gbinvepsdiff)));
 553             gbscale          = _fjsp_mul_v2r8(isaprod,gbtabscale);
 554
 555             /* Calculate generalized born table index - this is a separate table from the normal one,
 556              * but we use the same procedure by multiplying r with scale and truncating to integer.
 557              */
 558             rt               = _fjsp_mul_v2r8(r00,gbscale);
 559             itab_tmp         = _fjsp_dtox_v2r8(rt);
 560             gbeps            = _fjsp_sub_v2r8(rt,_fjsp_xtod_v2r8(itab_tmp));
 561             _fjsp_store_v2r8(&gbconv.simd,itab_tmp);
 562
 563             Y                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] );
 564             F                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] );
 565             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 566             G                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] +2);
 567             H                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] +2);
 568             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 569             Fp               = _fjsp_madd_v2r8(gbeps,_fjsp_madd_v2r8(gbeps,H,G),F);
 570             VV               = _fjsp_madd_v2r8(gbeps,Fp,Y);
 571             vgb              = _fjsp_mul_v2r8(gbqqfactor,VV);
 572
 573             twogbeps         = _fjsp_add_v2r8(gbeps,gbeps);
 574             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twogbeps,H,G),gbeps,Fp);
 575             fgb              = _fjsp_mul_v2r8(gbqqfactor,_fjsp_mul_v2r8(FF,gbscale));
 576             dvdatmp          = _fjsp_mul_v2r8(minushalf,_fjsp_madd_v2r8(fgb,r00,vgb));
 577             dvdasum          = _fjsp_add_v2r8(dvdasum,dvdatmp);
 578             gmx_fjsp_increment_2real_swizzle_v2r8(dvda+jnrA,dvda+jnrB,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj0,isaj0)));
 579             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 580             felec            = _fjsp_mul_v2r8(_fjsp_msub_v2r8(velec,rinv00,fgb),rinv00);
 581
 582             /* LENNARD-JONES DISPERSION/REPULSION */
 583
 584             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 585             fvdw             = _fjsp_mul_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,c6_00),_fjsp_mul_v2r8(rinvsix,rinvsq00));
 586
 587             fscal            = _fjsp_add_v2r8(felec,fvdw);
 588
 589             /* Update vectorial force */
 590             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 591             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 592             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 593
 594             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 595
 596             /* Inner loop uses 67 flops */
 597         }
 598
 599         if(jidx<j_index_end)
 600         {
 601
 602             jnrA             = jjnr[jidx];
 603             j_coord_offsetA  = DIM*jnrA;
 604
 605             /* load j atom coordinates */
 606             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 607                                               &jx0,&jy0,&jz0);
 608
 609             /* Calculate displacement vector */
 610             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 611             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 612             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 613
 614             /* Calculate squared distance and things based on it */
 615             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 616
 617             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 618
 619             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 620
 621             /* Load parameters for j particles */
 622             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 623             isaj0            = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),invsqrta+jnrA+0);
 624             vdwjidx0A        = 2*vdwtype[jnrA+0];
 625
 626             /**************************
 627              * CALCULATE INTERACTIONS *
 628              **************************/
 629
 630             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 631
 632             /* Compute parameters for interactions between i and j atoms */
 633             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 634             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 635                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 636
 637             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 638             isaprod          = _fjsp_mul_v2r8(isai0,isaj0);
 639             gbqqfactor       = _fjsp_neg_v2r8(_fjsp_mul_v2r8(qq00,_fjsp_mul_v2r8(isaprod,gbinvepsdiff)));
 640             gbscale          = _fjsp_mul_v2r8(isaprod,gbtabscale);
 641
 642             /* Calculate generalized born table index - this is a separate table from the normal one,
 643              * but we use the same procedure by multiplying r with scale and truncating to integer.
 644              */
 645             rt               = _fjsp_mul_v2r8(r00,gbscale);
 646             itab_tmp         = _fjsp_dtox_v2r8(rt);
 647             gbeps            = _fjsp_sub_v2r8(rt,_fjsp_xtod_v2r8(itab_tmp));
 648             _fjsp_store_v2r8(&gbconv.simd,itab_tmp);
 649
 650             Y                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] );
 651             F                = _fjsp_setzero_v2r8();
 652             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 653             G                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] +2);
 654             H                = _fjsp_setzero_v2r8();
 655             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 656             Fp               = _fjsp_madd_v2r8(gbeps,_fjsp_madd_v2r8(gbeps,H,G),F);
 657             VV               = _fjsp_madd_v2r8(gbeps,Fp,Y);
 658             vgb              = _fjsp_mul_v2r8(gbqqfactor,VV);
 659
 660             twogbeps         = _fjsp_add_v2r8(gbeps,gbeps);
 661             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twogbeps,H,G),gbeps,Fp);
 662             fgb              = _fjsp_mul_v2r8(gbqqfactor,_fjsp_mul_v2r8(FF,gbscale));
 663             dvdatmp          = _fjsp_mul_v2r8(minushalf,_fjsp_madd_v2r8(fgb,r00,vgb));
 664             dvdasum          = _fjsp_add_v2r8(dvdasum,dvdatmp);
 665             gmx_fjsp_increment_1real_v2r8(dvda+jnrA,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj0,isaj0)));
 666             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 667             felec            = _fjsp_mul_v2r8(_fjsp_msub_v2r8(velec,rinv00,fgb),rinv00);
 668
 669             /* LENNARD-JONES DISPERSION/REPULSION */
 670
 671             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 672             fvdw             = _fjsp_mul_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,c6_00),_fjsp_mul_v2r8(rinvsix,rinvsq00));
 673
 674             fscal            = _fjsp_add_v2r8(felec,fvdw);
 675
 676             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 677
 678             /* Update vectorial force */
 679             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 680             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 681             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 682
 683             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 684
 685             /* Inner loop uses 67 flops */
 686         }
 687
 688         /* End of innermost loop */
 689
 690         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 691                                               f+i_coord_offset,fshift+i_shift_offset);
 692
 693         dvdasum = _fjsp_mul_v2r8(dvdasum, _fjsp_mul_v2r8(isai0,isai0));
 694         gmx_fjsp_update_1pot_v2r8(dvdasum,dvda+inr);
 695
 696         /* Increment number of inner iterations */
 697         inneriter                  += j_index_end - j_index_start;
 698
 699         /* Outer loop uses 7 flops */
 700     }
 701
 702     /* Increment number of outer iterations */
 703     outeriter        += nri;
 704
 705     /* Update outer/inner flops */
 706
 707     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*67);
 708 }