src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecCSTab_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 "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_ElecCSTab_VdwLJ_GeomP1P1_VF_sparc64_hpc_ace_double
  51  * Electrostatics interaction: CubicSplineTable
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCSTab_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     int              nvdwtype;
  87     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  88     int              *vdwtype;
  89     real             *vdwparam;
  90     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
  91     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
  92     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  93     real             *vftab;
  94     _fjsp_v2r8       itab_tmp;
  95     _fjsp_v2r8       dummy_mask,cutoff_mask;
  96     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
  97     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
  98     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
  99
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 112     charge           = mdatoms->chargeA;
 113     nvdwtype         = fr->ntype;
 114     vdwparam         = fr->nbfp;
 115     vdwtype          = mdatoms->typeA;
 116
 117     vftab            = kernel_data->table_elec->data;
 118     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_elec->scale);
 119
 120     /* Avoid stupid compiler warnings */
 121     jnrA = jnrB = 0;
 122     j_coord_offsetA = 0;
 123     j_coord_offsetB = 0;
 124
 125     outeriter        = 0;
 126     inneriter        = 0;
 127
 128     /* Start outer loop over neighborlists */
 129     for(iidx=0; iidx<nri; iidx++)
 130     {
 131         /* Load shift vector for this list */
 132         i_shift_offset   = DIM*shiftidx[iidx];
 133
 134         /* Load limits for loop over neighbors */
 135         j_index_start    = jindex[iidx];
 136         j_index_end      = jindex[iidx+1];
 137
 138         /* Get outer coordinate index */
 139         inr              = iinr[iidx];
 140         i_coord_offset   = DIM*inr;
 141
 142         /* Load i particle coords and add shift vector */
 143         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 144
 145         fix0             = _fjsp_setzero_v2r8();
 146         fiy0             = _fjsp_setzero_v2r8();
 147         fiz0             = _fjsp_setzero_v2r8();
 148
 149         /* Load parameters for i particles */
 150         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 151         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 152
 153         /* Reset potential sums */
 154         velecsum         = _fjsp_setzero_v2r8();
 155         vvdwsum          = _fjsp_setzero_v2r8();
 156
 157         /* Start inner kernel loop */
 158         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 159         {
 160
 161             /* Get j neighbor index, and coordinate index */
 162             jnrA             = jjnr[jidx];
 163             jnrB             = jjnr[jidx+1];
 164             j_coord_offsetA  = DIM*jnrA;
 165             j_coord_offsetB  = DIM*jnrB;
 166
 167             /* load j atom coordinates */
 168             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 169                                               &jx0,&jy0,&jz0);
 170
 171             /* Calculate displacement vector */
 172             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 173             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 174             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 175
 176             /* Calculate squared distance and things based on it */
 177             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 178
 179             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 180
 181             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 182
 183             /* Load parameters for j particles */
 184             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 185             vdwjidx0A        = 2*vdwtype[jnrA+0];
 186             vdwjidx0B        = 2*vdwtype[jnrB+0];
 187
 188             /**************************
 189              * CALCULATE INTERACTIONS *
 190              **************************/
 191
 192             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 193
 194             /* Compute parameters for interactions between i and j atoms */
 195             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 196             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 197                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 198
 199             /* Calculate table index by multiplying r with table scale and truncate to integer */
 200             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 201             itab_tmp         = _fjsp_dtox_v2r8(rt);
 202             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 203             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 204             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 205
 206             vfconv.i[0]     *= 4;
 207             vfconv.i[1]     *= 4;
 208
 209             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 210             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 211             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 212             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 213             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] +2);
 214             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] +2);
 215             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 216             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(vfeps,H,G),F);
 217             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 218             velec            = _fjsp_mul_v2r8(qq00,VV);
 219             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twovfeps,H,G),vfeps,Fp);
 220             felec            = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,FF),_fjsp_mul_v2r8(vftabscale,rinv00)));
 221
 222             /* LENNARD-JONES DISPERSION/REPULSION */
 223
 224             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 225             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 226             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 227             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 228             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 229
 230             /* Update potential sum for this i atom from the interaction with this j atom. */
 231             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 232             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 233
 234             fscal            = _fjsp_add_v2r8(felec,fvdw);
 235
 236             /* Update vectorial force */
 237             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 238             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 239             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 240
 241             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 242
 243             /* Inner loop uses 59 flops */
 244         }
 245
 246         if(jidx<j_index_end)
 247         {
 248
 249             jnrA             = jjnr[jidx];
 250             j_coord_offsetA  = DIM*jnrA;
 251
 252             /* load j atom coordinates */
 253             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 254                                               &jx0,&jy0,&jz0);
 255
 256             /* Calculate displacement vector */
 257             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 258             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 259             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 260
 261             /* Calculate squared distance and things based on it */
 262             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 263
 264             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 265
 266             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 267
 268             /* Load parameters for j particles */
 269             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 270             vdwjidx0A        = 2*vdwtype[jnrA+0];
 271
 272             /**************************
 273              * CALCULATE INTERACTIONS *
 274              **************************/
 275
 276             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 277
 278             /* Compute parameters for interactions between i and j atoms */
 279             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 280             gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 281
 282             /* Calculate table index by multiplying r with table scale and truncate to integer */
 283             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 284             itab_tmp         = _fjsp_dtox_v2r8(rt);
 285             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 286             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 287             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 288
 289             vfconv.i[0]     *= 4;
 290             vfconv.i[1]     *= 4;
 291
 292             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 293             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 294             F                = _fjsp_setzero_v2r8();
 295             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 296             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] +2);
 297             H                = _fjsp_setzero_v2r8();
 298             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 299             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(vfeps,H,G),F);
 300             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 301             velec            = _fjsp_mul_v2r8(qq00,VV);
 302             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twovfeps,H,G),vfeps,Fp);
 303             felec            = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,FF),_fjsp_mul_v2r8(vftabscale,rinv00)));
 304
 305             /* LENNARD-JONES DISPERSION/REPULSION */
 306
 307             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 308             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 309             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 310             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 311             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 312
 313             /* Update potential sum for this i atom from the interaction with this j atom. */
 314             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 315             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 316             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 317             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 318
 319             fscal            = _fjsp_add_v2r8(felec,fvdw);
 320
 321             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 322
 323             /* Update vectorial force */
 324             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 325             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 326             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 327
 328             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 329
 330             /* Inner loop uses 59 flops */
 331         }
 332
 333         /* End of innermost loop */
 334
 335         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 336                                               f+i_coord_offset,fshift+i_shift_offset);
 337
 338         ggid                        = gid[iidx];
 339         /* Update potential energies */
 340         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 341         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 342
 343         /* Increment number of inner iterations */
 344         inneriter                  += j_index_end - j_index_start;
 345
 346         /* Outer loop uses 9 flops */
 347     }
 348
 349     /* Increment number of outer iterations */
 350     outeriter        += nri;
 351
 352     /* Update outer/inner flops */
 353
 354     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*59);
 355 }
 356 /*
 357  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_sparc64_hpc_ace_double
 358  * Electrostatics interaction: CubicSplineTable
 359  * VdW interaction:            LennardJones
 360  * Geometry:                   Particle-Particle
 361  * Calculate force/pot:        Force
 362  */
 363 void
 364 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_sparc64_hpc_ace_double
 365                     (t_nblist                    * gmx_restrict       nlist,
 366                      rvec                        * gmx_restrict          xx,
 367                      rvec                        * gmx_restrict          ff,
 368                      t_forcerec                  * gmx_restrict          fr,
 369                      t_mdatoms                   * gmx_restrict     mdatoms,
 370                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 371                      t_nrnb                      * gmx_restrict        nrnb)
 372 {
 373     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 374      * just 0 for non-waters.
 375      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 376      * jnr indices corresponding to data put in the four positions in the SIMD register.
 377      */
 378     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 379     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 380     int              jnrA,jnrB;
 381     int              j_coord_offsetA,j_coord_offsetB;
 382     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 383     real             rcutoff_scalar;
 384     real             *shiftvec,*fshift,*x,*f;
 385     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 386     int              vdwioffset0;
 387     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 388     int              vdwjidx0A,vdwjidx0B;
 389     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 390     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 391     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 392     real             *charge;
 393     int              nvdwtype;
 394     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 395     int              *vdwtype;
 396     real             *vdwparam;
 397     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 398     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 399     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 400     real             *vftab;
 401     _fjsp_v2r8       itab_tmp;
 402     _fjsp_v2r8       dummy_mask,cutoff_mask;
 403     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 404     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 405     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 406
 407     x                = xx[0];
 408     f                = ff[0];
 409
 410     nri              = nlist->nri;
 411     iinr             = nlist->iinr;
 412     jindex           = nlist->jindex;
 413     jjnr             = nlist->jjnr;
 414     shiftidx         = nlist->shift;
 415     gid              = nlist->gid;
 416     shiftvec         = fr->shift_vec[0];
 417     fshift           = fr->fshift[0];
 418     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 419     charge           = mdatoms->chargeA;
 420     nvdwtype         = fr->ntype;
 421     vdwparam         = fr->nbfp;
 422     vdwtype          = mdatoms->typeA;
 423
 424     vftab            = kernel_data->table_elec->data;
 425     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_elec->scale);
 426
 427     /* Avoid stupid compiler warnings */
 428     jnrA = jnrB = 0;
 429     j_coord_offsetA = 0;
 430     j_coord_offsetB = 0;
 431
 432     outeriter        = 0;
 433     inneriter        = 0;
 434
 435     /* Start outer loop over neighborlists */
 436     for(iidx=0; iidx<nri; iidx++)
 437     {
 438         /* Load shift vector for this list */
 439         i_shift_offset   = DIM*shiftidx[iidx];
 440
 441         /* Load limits for loop over neighbors */
 442         j_index_start    = jindex[iidx];
 443         j_index_end      = jindex[iidx+1];
 444
 445         /* Get outer coordinate index */
 446         inr              = iinr[iidx];
 447         i_coord_offset   = DIM*inr;
 448
 449         /* Load i particle coords and add shift vector */
 450         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 451
 452         fix0             = _fjsp_setzero_v2r8();
 453         fiy0             = _fjsp_setzero_v2r8();
 454         fiz0             = _fjsp_setzero_v2r8();
 455
 456         /* Load parameters for i particles */
 457         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 458         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 459
 460         /* Start inner kernel loop */
 461         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 462         {
 463
 464             /* Get j neighbor index, and coordinate index */
 465             jnrA             = jjnr[jidx];
 466             jnrB             = jjnr[jidx+1];
 467             j_coord_offsetA  = DIM*jnrA;
 468             j_coord_offsetB  = DIM*jnrB;
 469
 470             /* load j atom coordinates */
 471             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 472                                               &jx0,&jy0,&jz0);
 473
 474             /* Calculate displacement vector */
 475             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 476             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 477             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 478
 479             /* Calculate squared distance and things based on it */
 480             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 481
 482             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 483
 484             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 485
 486             /* Load parameters for j particles */
 487             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 488             vdwjidx0A        = 2*vdwtype[jnrA+0];
 489             vdwjidx0B        = 2*vdwtype[jnrB+0];
 490
 491             /**************************
 492              * CALCULATE INTERACTIONS *
 493              **************************/
 494
 495             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 496
 497             /* Compute parameters for interactions between i and j atoms */
 498             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 499             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 500                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 501
 502             /* Calculate table index by multiplying r with table scale and truncate to integer */
 503             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 504             itab_tmp         = _fjsp_dtox_v2r8(rt);
 505             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 506             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 507             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 508
 509             vfconv.i[0]     *= 4;
 510             vfconv.i[1]     *= 4;
 511
 512             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 513             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 514             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 515             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 516             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] +2);
 517             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] +2);
 518             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 519             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(vfeps,H,G),F);
 520             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twovfeps,H,G),vfeps,Fp);
 521             felec            = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,FF),_fjsp_mul_v2r8(vftabscale,rinv00)));
 522
 523             /* LENNARD-JONES DISPERSION/REPULSION */
 524
 525             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 526             fvdw             = _fjsp_mul_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,c6_00),_fjsp_mul_v2r8(rinvsix,rinvsq00));
 527
 528             fscal            = _fjsp_add_v2r8(felec,fvdw);
 529
 530             /* Update vectorial force */
 531             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 532             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 533             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 534
 535             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 536
 537             /* Inner loop uses 50 flops */
 538         }
 539
 540         if(jidx<j_index_end)
 541         {
 542
 543             jnrA             = jjnr[jidx];
 544             j_coord_offsetA  = DIM*jnrA;
 545
 546             /* load j atom coordinates */
 547             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 548                                               &jx0,&jy0,&jz0);
 549
 550             /* Calculate displacement vector */
 551             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 552             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 553             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 554
 555             /* Calculate squared distance and things based on it */
 556             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 557
 558             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 559
 560             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 561
 562             /* Load parameters for j particles */
 563             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 564             vdwjidx0A        = 2*vdwtype[jnrA+0];
 565
 566             /**************************
 567              * CALCULATE INTERACTIONS *
 568              **************************/
 569
 570             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 571
 572             /* Compute parameters for interactions between i and j atoms */
 573             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 574             gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 575
 576             /* Calculate table index by multiplying r with table scale and truncate to integer */
 577             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 578             itab_tmp         = _fjsp_dtox_v2r8(rt);
 579             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 580             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 581             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 582
 583             vfconv.i[0]     *= 4;
 584             vfconv.i[1]     *= 4;
 585
 586             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 587             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 588             F                = _fjsp_setzero_v2r8();
 589             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 590             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] +2);
 591             H                = _fjsp_setzero_v2r8();
 592             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 593             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(vfeps,H,G),F);
 594             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twovfeps,H,G),vfeps,Fp);
 595             felec            = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,FF),_fjsp_mul_v2r8(vftabscale,rinv00)));
 596
 597             /* LENNARD-JONES DISPERSION/REPULSION */
 598
 599             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 600             fvdw             = _fjsp_mul_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,c6_00),_fjsp_mul_v2r8(rinvsix,rinvsq00));
 601
 602             fscal            = _fjsp_add_v2r8(felec,fvdw);
 603
 604             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 605
 606             /* Update vectorial force */
 607             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 608             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 609             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 610
 611             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 612
 613             /* Inner loop uses 50 flops */
 614         }
 615
 616         /* End of innermost loop */
 617
 618         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 619                                               f+i_coord_offset,fshift+i_shift_offset);
 620
 621         /* Increment number of inner iterations */
 622         inneriter                  += j_index_end - j_index_start;
 623
 624         /* Outer loop uses 7 flops */
 625     }
 626
 627     /* Increment number of outer iterations */
 628     outeriter        += nri;
 629
 630     /* Update outer/inner flops */
 631
 632     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*50);
 633 }