src/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_sparc64_hpc_ace_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012, by the GROMACS development team, led by
   5  * David van der Spoel, Berk Hess, Erik Lindahl, and including many
   6  * others, as listed in the AUTHORS file in the top-level source
   7  * 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "kernelutil_sparc64_hpc_ace_double.h"
  50
  51 /*
  52  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_VF_sparc64_hpc_ace_double
  53  * Electrostatics interaction: ReactionField
  54  * VdW interaction:            CubicSplineTable
  55  * Geometry:                   Particle-Particle
  56  * Calculate force/pot:        PotentialAndForce
  57  */
  58 void
  59 nb_kernel_ElecRFCut_VdwCSTab_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_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     int              nvdwtype;
  89     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  90     int              *vdwtype;
  91     real             *vdwparam;
  92     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
  93     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
  94     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  95     real             *vftab;
  96     _fjsp_v2r8       itab_tmp;
  97     _fjsp_v2r8       dummy_mask,cutoff_mask;
  98     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
  99     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 100     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 101
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115     krf              = gmx_fjsp_set1_v2r8(fr->ic->k_rf);
 116     krf2             = gmx_fjsp_set1_v2r8(fr->ic->k_rf*2.0);
 117     crf              = gmx_fjsp_set1_v2r8(fr->ic->c_rf);
 118     nvdwtype         = fr->ntype;
 119     vdwparam         = fr->nbfp;
 120     vdwtype          = mdatoms->typeA;
 121
 122     vftab            = kernel_data->table_vdw->data;
 123     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_vdw->scale);
 124
 125     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 126     rcutoff_scalar   = fr->rcoulomb;
 127     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 128     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 129
 130     /* Avoid stupid compiler warnings */
 131     jnrA = jnrB = 0;
 132     j_coord_offsetA = 0;
 133     j_coord_offsetB = 0;
 134
 135     outeriter        = 0;
 136     inneriter        = 0;
 137
 138     /* Start outer loop over neighborlists */
 139     for(iidx=0; iidx<nri; iidx++)
 140     {
 141         /* Load shift vector for this list */
 142         i_shift_offset   = DIM*shiftidx[iidx];
 143
 144         /* Load limits for loop over neighbors */
 145         j_index_start    = jindex[iidx];
 146         j_index_end      = jindex[iidx+1];
 147
 148         /* Get outer coordinate index */
 149         inr              = iinr[iidx];
 150         i_coord_offset   = DIM*inr;
 151
 152         /* Load i particle coords and add shift vector */
 153         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 154
 155         fix0             = _fjsp_setzero_v2r8();
 156         fiy0             = _fjsp_setzero_v2r8();
 157         fiz0             = _fjsp_setzero_v2r8();
 158
 159         /* Load parameters for i particles */
 160         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 161         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 162
 163         /* Reset potential sums */
 164         velecsum         = _fjsp_setzero_v2r8();
 165         vvdwsum          = _fjsp_setzero_v2r8();
 166
 167         /* Start inner kernel loop */
 168         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 169         {
 170
 171             /* Get j neighbor index, and coordinate index */
 172             jnrA             = jjnr[jidx];
 173             jnrB             = jjnr[jidx+1];
 174             j_coord_offsetA  = DIM*jnrA;
 175             j_coord_offsetB  = DIM*jnrB;
 176
 177             /* load j atom coordinates */
 178             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 179                                               &jx0,&jy0,&jz0);
 180
 181             /* Calculate displacement vector */
 182             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 183             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 184             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 185
 186             /* Calculate squared distance and things based on it */
 187             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 188
 189             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 190
 191             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 192
 193             /* Load parameters for j particles */
 194             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 195             vdwjidx0A        = 2*vdwtype[jnrA+0];
 196             vdwjidx0B        = 2*vdwtype[jnrB+0];
 197
 198             /**************************
 199              * CALCULATE INTERACTIONS *
 200              **************************/
 201
 202             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 203             {
 204
 205             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 206
 207             /* Compute parameters for interactions between i and j atoms */
 208             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 209             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 210                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 211
 212             /* Calculate table index by multiplying r with table scale and truncate to integer */
 213             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 214             itab_tmp         = _fjsp_dtox_v2r8(rt);
 215             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 216             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 217             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 218
 219             vfconv.i[0]     *= 8;
 220             vfconv.i[1]     *= 8;
 221
 222             /* REACTION-FIELD ELECTROSTATICS */
 223             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(_fjsp_madd_v2r8(krf,rsq00,rinv00),crf));
 224             felec            = _fjsp_mul_v2r8(qq00,_fjsp_msub_v2r8(rinv00,rinvsq00,krf2));
 225
 226             /* CUBIC SPLINE TABLE DISPERSION */
 227             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 228             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 229             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 230             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 231             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 2 );
 232             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 233             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 234             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 235             vvdw6            = _fjsp_mul_v2r8(c6_00,VV);
 236             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 237             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 238
 239             /* CUBIC SPLINE TABLE REPULSION */
 240             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 241             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 4 );
 242             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 243             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 244             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 6 );
 245             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 246             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 247             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 248             vvdw12           = _fjsp_mul_v2r8(c12_00,VV);
 249             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 250             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 251             vvdw             = _fjsp_add_v2r8(vvdw12,vvdw6);
 252             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 253
 254             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 255
 256             /* Update potential sum for this i atom from the interaction with this j atom. */
 257             velec            = _fjsp_and_v2r8(velec,cutoff_mask);
 258             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 259             vvdw             = _fjsp_and_v2r8(vvdw,cutoff_mask);
 260             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 261
 262             fscal            = _fjsp_add_v2r8(felec,fvdw);
 263
 264             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 265
 266             /* Update vectorial force */
 267             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 268             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 269             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 270
 271             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 272
 273             }
 274
 275             /* Inner loop uses 75 flops */
 276         }
 277
 278         if(jidx<j_index_end)
 279         {
 280
 281             jnrA             = jjnr[jidx];
 282             j_coord_offsetA  = DIM*jnrA;
 283
 284             /* load j atom coordinates */
 285             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 286                                               &jx0,&jy0,&jz0);
 287
 288             /* Calculate displacement vector */
 289             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 290             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 291             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 292
 293             /* Calculate squared distance and things based on it */
 294             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 295
 296             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 297
 298             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 299
 300             /* Load parameters for j particles */
 301             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 302             vdwjidx0A        = 2*vdwtype[jnrA+0];
 303
 304             /**************************
 305              * CALCULATE INTERACTIONS *
 306              **************************/
 307
 308             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 309             {
 310
 311             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 312
 313             /* Compute parameters for interactions between i and j atoms */
 314             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 315             gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 316
 317             /* Calculate table index by multiplying r with table scale and truncate to integer */
 318             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 319             itab_tmp         = _fjsp_dtox_v2r8(rt);
 320             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 321             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 322             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 323
 324             vfconv.i[0]     *= 8;
 325             vfconv.i[1]     *= 8;
 326
 327             /* REACTION-FIELD ELECTROSTATICS */
 328             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(_fjsp_madd_v2r8(krf,rsq00,rinv00),crf));
 329             felec            = _fjsp_mul_v2r8(qq00,_fjsp_msub_v2r8(rinv00,rinvsq00,krf2));
 330
 331             /* CUBIC SPLINE TABLE DISPERSION */
 332             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 333             F                = _fjsp_setzero_v2r8();
 334             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 335             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 336             H                = _fjsp_setzero_v2r8();
 337             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 338             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 339             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 340             vvdw6            = _fjsp_mul_v2r8(c6_00,VV);
 341             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 342             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 343
 344             /* CUBIC SPLINE TABLE REPULSION */
 345             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 346             F                = _fjsp_setzero_v2r8();
 347             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 348             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 349             H                = _fjsp_setzero_v2r8();
 350             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 351             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 352             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 353             vvdw12           = _fjsp_mul_v2r8(c12_00,VV);
 354             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 355             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 356             vvdw             = _fjsp_add_v2r8(vvdw12,vvdw6);
 357             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 358
 359             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 360
 361             /* Update potential sum for this i atom from the interaction with this j atom. */
 362             velec            = _fjsp_and_v2r8(velec,cutoff_mask);
 363             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 364             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 365             vvdw             = _fjsp_and_v2r8(vvdw,cutoff_mask);
 366             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 367             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 368
 369             fscal            = _fjsp_add_v2r8(felec,fvdw);
 370
 371             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 372
 373             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 374
 375             /* Update vectorial force */
 376             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 377             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 378             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 379
 380             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 381
 382             }
 383
 384             /* Inner loop uses 75 flops */
 385         }
 386
 387         /* End of innermost loop */
 388
 389         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 390                                               f+i_coord_offset,fshift+i_shift_offset);
 391
 392         ggid                        = gid[iidx];
 393         /* Update potential energies */
 394         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 395         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 396
 397         /* Increment number of inner iterations */
 398         inneriter                  += j_index_end - j_index_start;
 399
 400         /* Outer loop uses 9 flops */
 401     }
 402
 403     /* Increment number of outer iterations */
 404     outeriter        += nri;
 405
 406     /* Update outer/inner flops */
 407
 408     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*75);
 409 }
 410 /*
 411  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_F_sparc64_hpc_ace_double
 412  * Electrostatics interaction: ReactionField
 413  * VdW interaction:            CubicSplineTable
 414  * Geometry:                   Particle-Particle
 415  * Calculate force/pot:        Force
 416  */
 417 void
 418 nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_F_sparc64_hpc_ace_double
 419                     (t_nblist * gmx_restrict                nlist,
 420                      rvec * gmx_restrict                    xx,
 421                      rvec * gmx_restrict                    ff,
 422                      t_forcerec * gmx_restrict              fr,
 423                      t_mdatoms * gmx_restrict               mdatoms,
 424                      nb_kernel_data_t * gmx_restrict        kernel_data,
 425                      t_nrnb * gmx_restrict                  nrnb)
 426 {
 427     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 428      * just 0 for non-waters.
 429      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 430      * jnr indices corresponding to data put in the four positions in the SIMD register.
 431      */
 432     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 433     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 434     int              jnrA,jnrB;
 435     int              j_coord_offsetA,j_coord_offsetB;
 436     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 437     real             rcutoff_scalar;
 438     real             *shiftvec,*fshift,*x,*f;
 439     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 440     int              vdwioffset0;
 441     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 442     int              vdwjidx0A,vdwjidx0B;
 443     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 444     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 445     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 446     real             *charge;
 447     int              nvdwtype;
 448     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 449     int              *vdwtype;
 450     real             *vdwparam;
 451     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 452     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 453     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 454     real             *vftab;
 455     _fjsp_v2r8       itab_tmp;
 456     _fjsp_v2r8       dummy_mask,cutoff_mask;
 457     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 458     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 459     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 460
 461     x                = xx[0];
 462     f                = ff[0];
 463
 464     nri              = nlist->nri;
 465     iinr             = nlist->iinr;
 466     jindex           = nlist->jindex;
 467     jjnr             = nlist->jjnr;
 468     shiftidx         = nlist->shift;
 469     gid              = nlist->gid;
 470     shiftvec         = fr->shift_vec[0];
 471     fshift           = fr->fshift[0];
 472     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 473     charge           = mdatoms->chargeA;
 474     krf              = gmx_fjsp_set1_v2r8(fr->ic->k_rf);
 475     krf2             = gmx_fjsp_set1_v2r8(fr->ic->k_rf*2.0);
 476     crf              = gmx_fjsp_set1_v2r8(fr->ic->c_rf);
 477     nvdwtype         = fr->ntype;
 478     vdwparam         = fr->nbfp;
 479     vdwtype          = mdatoms->typeA;
 480
 481     vftab            = kernel_data->table_vdw->data;
 482     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_vdw->scale);
 483
 484     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 485     rcutoff_scalar   = fr->rcoulomb;
 486     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 487     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 488
 489     /* Avoid stupid compiler warnings */
 490     jnrA = jnrB = 0;
 491     j_coord_offsetA = 0;
 492     j_coord_offsetB = 0;
 493
 494     outeriter        = 0;
 495     inneriter        = 0;
 496
 497     /* Start outer loop over neighborlists */
 498     for(iidx=0; iidx<nri; iidx++)
 499     {
 500         /* Load shift vector for this list */
 501         i_shift_offset   = DIM*shiftidx[iidx];
 502
 503         /* Load limits for loop over neighbors */
 504         j_index_start    = jindex[iidx];
 505         j_index_end      = jindex[iidx+1];
 506
 507         /* Get outer coordinate index */
 508         inr              = iinr[iidx];
 509         i_coord_offset   = DIM*inr;
 510
 511         /* Load i particle coords and add shift vector */
 512         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 513
 514         fix0             = _fjsp_setzero_v2r8();
 515         fiy0             = _fjsp_setzero_v2r8();
 516         fiz0             = _fjsp_setzero_v2r8();
 517
 518         /* Load parameters for i particles */
 519         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 520         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 521
 522         /* Start inner kernel loop */
 523         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 524         {
 525
 526             /* Get j neighbor index, and coordinate index */
 527             jnrA             = jjnr[jidx];
 528             jnrB             = jjnr[jidx+1];
 529             j_coord_offsetA  = DIM*jnrA;
 530             j_coord_offsetB  = DIM*jnrB;
 531
 532             /* load j atom coordinates */
 533             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 534                                               &jx0,&jy0,&jz0);
 535
 536             /* Calculate displacement vector */
 537             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 538             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 539             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 540
 541             /* Calculate squared distance and things based on it */
 542             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 543
 544             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 545
 546             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 547
 548             /* Load parameters for j particles */
 549             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 550             vdwjidx0A        = 2*vdwtype[jnrA+0];
 551             vdwjidx0B        = 2*vdwtype[jnrB+0];
 552
 553             /**************************
 554              * CALCULATE INTERACTIONS *
 555              **************************/
 556
 557             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 558             {
 559
 560             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 561
 562             /* Compute parameters for interactions between i and j atoms */
 563             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 564             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 565                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 566
 567             /* Calculate table index by multiplying r with table scale and truncate to integer */
 568             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 569             itab_tmp         = _fjsp_dtox_v2r8(rt);
 570             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 571             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 572             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 573
 574             vfconv.i[0]     *= 8;
 575             vfconv.i[1]     *= 8;
 576
 577             /* REACTION-FIELD ELECTROSTATICS */
 578             felec            = _fjsp_mul_v2r8(qq00,_fjsp_msub_v2r8(rinv00,rinvsq00,krf2));
 579
 580             /* CUBIC SPLINE TABLE DISPERSION */
 581             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 582             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 583             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 584             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 585             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 2 );
 586             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 587             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 588             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 589             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 590
 591             /* CUBIC SPLINE TABLE REPULSION */
 592             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 593             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 4 );
 594             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 595             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 596             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 6 );
 597             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 598             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 599             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 600             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 601             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 602
 603             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 604
 605             fscal            = _fjsp_add_v2r8(felec,fvdw);
 606
 607             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 608
 609             /* Update vectorial force */
 610             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 611             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 612             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 613
 614             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 615
 616             }
 617
 618             /* Inner loop uses 60 flops */
 619         }
 620
 621         if(jidx<j_index_end)
 622         {
 623
 624             jnrA             = jjnr[jidx];
 625             j_coord_offsetA  = DIM*jnrA;
 626
 627             /* load j atom coordinates */
 628             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 629                                               &jx0,&jy0,&jz0);
 630
 631             /* Calculate displacement vector */
 632             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 633             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 634             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 635
 636             /* Calculate squared distance and things based on it */
 637             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 638
 639             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 640
 641             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 642
 643             /* Load parameters for j particles */
 644             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 645             vdwjidx0A        = 2*vdwtype[jnrA+0];
 646
 647             /**************************
 648              * CALCULATE INTERACTIONS *
 649              **************************/
 650
 651             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 652             {
 653
 654             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 655
 656             /* Compute parameters for interactions between i and j atoms */
 657             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 658             gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 659
 660             /* Calculate table index by multiplying r with table scale and truncate to integer */
 661             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 662             itab_tmp         = _fjsp_dtox_v2r8(rt);
 663             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 664             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 665             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 666
 667             vfconv.i[0]     *= 8;
 668             vfconv.i[1]     *= 8;
 669
 670             /* REACTION-FIELD ELECTROSTATICS */
 671             felec            = _fjsp_mul_v2r8(qq00,_fjsp_msub_v2r8(rinv00,rinvsq00,krf2));
 672
 673             /* CUBIC SPLINE TABLE DISPERSION */
 674             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 675             F                = _fjsp_setzero_v2r8();
 676             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 677             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 678             H                = _fjsp_setzero_v2r8();
 679             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 680             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 681             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 682             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 683
 684             /* CUBIC SPLINE TABLE REPULSION */
 685             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 686             F                = _fjsp_setzero_v2r8();
 687             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 688             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 689             H                = _fjsp_setzero_v2r8();
 690             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 691             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 692             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 693             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 694             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 695
 696             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 697
 698             fscal            = _fjsp_add_v2r8(felec,fvdw);
 699
 700             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 701
 702             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 703
 704             /* Update vectorial force */
 705             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 706             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 707             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 708
 709             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 710
 711             }
 712
 713             /* Inner loop uses 60 flops */
 714         }
 715
 716         /* End of innermost loop */
 717
 718         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 719                                               f+i_coord_offset,fshift+i_shift_offset);
 720
 721         /* Increment number of inner iterations */
 722         inneriter                  += j_index_end - j_index_start;
 723
 724         /* Outer loop uses 7 flops */
 725     }
 726
 727     /* Increment number of outer iterations */
 728     outeriter        += nri;
 729
 730     /* Update outer/inner flops */
 731
 732     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*60);
 733 }