src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecEw_VdwCSTab_GeomP1P1_sparc64_hpc_ace_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_sparc64_hpc_ace_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomP1P1_VF_sparc64_hpc_ace_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwCSTab_GeomP1P1_VF_sparc64_hpc_ace_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct 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       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  95     real             *ewtab;
  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->ic->epsfac);
 114     charge           = mdatoms->chargeA;
 115     nvdwtype         = fr->ntype;
 116     vdwparam         = fr->nbfp;
 117     vdwtype          = mdatoms->typeA;
 118
 119     vftab            = kernel_data->table_vdw->data;
 120     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_vdw->scale);
 121
 122     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 123     ewtab            = fr->ic->tabq_coul_FDV0;
 124     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 125     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 126
 127     /* Avoid stupid compiler warnings */
 128     jnrA = jnrB = 0;
 129     j_coord_offsetA = 0;
 130     j_coord_offsetB = 0;
 131
 132     outeriter        = 0;
 133     inneriter        = 0;
 134
 135     /* Start outer loop over neighborlists */
 136     for(iidx=0; iidx<nri; iidx++)
 137     {
 138         /* Load shift vector for this list */
 139         i_shift_offset   = DIM*shiftidx[iidx];
 140
 141         /* Load limits for loop over neighbors */
 142         j_index_start    = jindex[iidx];
 143         j_index_end      = jindex[iidx+1];
 144
 145         /* Get outer coordinate index */
 146         inr              = iinr[iidx];
 147         i_coord_offset   = DIM*inr;
 148
 149         /* Load i particle coords and add shift vector */
 150         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 151
 152         fix0             = _fjsp_setzero_v2r8();
 153         fiy0             = _fjsp_setzero_v2r8();
 154         fiz0             = _fjsp_setzero_v2r8();
 155
 156         /* Load parameters for i particles */
 157         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 158         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 159
 160         /* Reset potential sums */
 161         velecsum         = _fjsp_setzero_v2r8();
 162         vvdwsum          = _fjsp_setzero_v2r8();
 163
 164         /* Start inner kernel loop */
 165         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 166         {
 167
 168             /* Get j neighbor index, and coordinate index */
 169             jnrA             = jjnr[jidx];
 170             jnrB             = jjnr[jidx+1];
 171             j_coord_offsetA  = DIM*jnrA;
 172             j_coord_offsetB  = DIM*jnrB;
 173
 174             /* load j atom coordinates */
 175             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 176                                               &jx0,&jy0,&jz0);
 177
 178             /* Calculate displacement vector */
 179             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 180             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 181             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 182
 183             /* Calculate squared distance and things based on it */
 184             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 185
 186             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 187
 188             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 189
 190             /* Load parameters for j particles */
 191             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 192             vdwjidx0A        = 2*vdwtype[jnrA+0];
 193             vdwjidx0B        = 2*vdwtype[jnrB+0];
 194
 195             /**************************
 196              * CALCULATE INTERACTIONS *
 197              **************************/
 198
 199             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 200
 201             /* Compute parameters for interactions between i and j atoms */
 202             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 203             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 204                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 205
 206             /* Calculate table index by multiplying r with table scale and truncate to integer */
 207             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 208             itab_tmp         = _fjsp_dtox_v2r8(rt);
 209             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 210             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 211             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 212
 213             vfconv.i[0]     *= 8;
 214             vfconv.i[1]     *= 8;
 215
 216             /* EWALD ELECTROSTATICS */
 217
 218             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 219             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 220             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 221             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 222             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 223
 224             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 225             ewtabD           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
 226             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 227             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 228             ewtabFn          = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
 229             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 230             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 231             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 232             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 233             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 234
 235             /* CUBIC SPLINE TABLE DISPERSION */
 236             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 237             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 238             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 239             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 240             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 2 );
 241             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 242             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 243             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 244             vvdw6            = _fjsp_mul_v2r8(c6_00,VV);
 245             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 246             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 247
 248             /* CUBIC SPLINE TABLE REPULSION */
 249             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 250             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 4 );
 251             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 252             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 253             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 6 );
 254             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 255             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 256             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 257             vvdw12           = _fjsp_mul_v2r8(c12_00,VV);
 258             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 259             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 260             vvdw             = _fjsp_add_v2r8(vvdw12,vvdw6);
 261             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 262
 263             /* Update potential sum for this i atom from the interaction with this j atom. */
 264             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 265             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 266
 267             fscal            = _fjsp_add_v2r8(felec,fvdw);
 268
 269             /* Update vectorial force */
 270             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 271             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 272             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 273
 274             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 275
 276             /* Inner loop uses 78 flops */
 277         }
 278
 279         if(jidx<j_index_end)
 280         {
 281
 282             jnrA             = jjnr[jidx];
 283             j_coord_offsetA  = DIM*jnrA;
 284
 285             /* load j atom coordinates */
 286             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 287                                               &jx0,&jy0,&jz0);
 288
 289             /* Calculate displacement vector */
 290             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 291             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 292             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 293
 294             /* Calculate squared distance and things based on it */
 295             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 296
 297             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 298
 299             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 300
 301             /* Load parameters for j particles */
 302             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 303             vdwjidx0A        = 2*vdwtype[jnrA+0];
 304
 305             /**************************
 306              * CALCULATE INTERACTIONS *
 307              **************************/
 308
 309             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 310
 311             /* Compute parameters for interactions between i and j atoms */
 312             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 313             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 314                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 315
 316             /* Calculate table index by multiplying r with table scale and truncate to integer */
 317             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 318             itab_tmp         = _fjsp_dtox_v2r8(rt);
 319             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 320             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 321             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 322
 323             vfconv.i[0]     *= 8;
 324             vfconv.i[1]     *= 8;
 325
 326             /* EWALD ELECTROSTATICS */
 327
 328             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 329             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 330             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 331             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 332             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 333
 334             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 335             ewtabD           = _fjsp_setzero_v2r8();
 336             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 337             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 338             ewtabFn          = _fjsp_setzero_v2r8();
 339             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 340             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 341             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 342             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 343             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 344
 345             /* CUBIC SPLINE TABLE DISPERSION */
 346             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 347             F                = _fjsp_setzero_v2r8();
 348             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 349             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 350             H                = _fjsp_setzero_v2r8();
 351             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 352             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 353             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 354             vvdw6            = _fjsp_mul_v2r8(c6_00,VV);
 355             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 356             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 357
 358             /* CUBIC SPLINE TABLE REPULSION */
 359             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 360             F                = _fjsp_setzero_v2r8();
 361             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 362             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 363             H                = _fjsp_setzero_v2r8();
 364             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 365             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 366             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 367             vvdw12           = _fjsp_mul_v2r8(c12_00,VV);
 368             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 369             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 370             vvdw             = _fjsp_add_v2r8(vvdw12,vvdw6);
 371             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 372
 373             /* Update potential sum for this i atom from the interaction with this j atom. */
 374             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 375             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 376             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 377             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 378
 379             fscal            = _fjsp_add_v2r8(felec,fvdw);
 380
 381             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 382
 383             /* Update vectorial force */
 384             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 385             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 386             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 387
 388             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 389
 390             /* Inner loop uses 78 flops */
 391         }
 392
 393         /* End of innermost loop */
 394
 395         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 396                                               f+i_coord_offset,fshift+i_shift_offset);
 397
 398         ggid                        = gid[iidx];
 399         /* Update potential energies */
 400         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 401         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 402
 403         /* Increment number of inner iterations */
 404         inneriter                  += j_index_end - j_index_start;
 405
 406         /* Outer loop uses 9 flops */
 407     }
 408
 409     /* Increment number of outer iterations */
 410     outeriter        += nri;
 411
 412     /* Update outer/inner flops */
 413
 414     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*78);
 415 }
 416 /*
 417  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomP1P1_F_sparc64_hpc_ace_double
 418  * Electrostatics interaction: Ewald
 419  * VdW interaction:            CubicSplineTable
 420  * Geometry:                   Particle-Particle
 421  * Calculate force/pot:        Force
 422  */
 423 void
 424 nb_kernel_ElecEw_VdwCSTab_GeomP1P1_F_sparc64_hpc_ace_double
 425                     (t_nblist                    * gmx_restrict       nlist,
 426                      rvec                        * gmx_restrict          xx,
 427                      rvec                        * gmx_restrict          ff,
 428                      struct t_forcerec           * gmx_restrict          fr,
 429                      t_mdatoms                   * gmx_restrict     mdatoms,
 430                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 431                      t_nrnb                      * gmx_restrict        nrnb)
 432 {
 433     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 434      * just 0 for non-waters.
 435      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 436      * jnr indices corresponding to data put in the four positions in the SIMD register.
 437      */
 438     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 439     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 440     int              jnrA,jnrB;
 441     int              j_coord_offsetA,j_coord_offsetB;
 442     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 443     real             rcutoff_scalar;
 444     real             *shiftvec,*fshift,*x,*f;
 445     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 446     int              vdwioffset0;
 447     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 448     int              vdwjidx0A,vdwjidx0B;
 449     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 450     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 451     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 452     real             *charge;
 453     int              nvdwtype;
 454     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 455     int              *vdwtype;
 456     real             *vdwparam;
 457     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 458     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 459     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 460     real             *vftab;
 461     _fjsp_v2r8       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 462     real             *ewtab;
 463     _fjsp_v2r8       itab_tmp;
 464     _fjsp_v2r8       dummy_mask,cutoff_mask;
 465     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 466     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 467     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 468
 469     x                = xx[0];
 470     f                = ff[0];
 471
 472     nri              = nlist->nri;
 473     iinr             = nlist->iinr;
 474     jindex           = nlist->jindex;
 475     jjnr             = nlist->jjnr;
 476     shiftidx         = nlist->shift;
 477     gid              = nlist->gid;
 478     shiftvec         = fr->shift_vec[0];
 479     fshift           = fr->fshift[0];
 480     facel            = gmx_fjsp_set1_v2r8(fr->ic->epsfac);
 481     charge           = mdatoms->chargeA;
 482     nvdwtype         = fr->ntype;
 483     vdwparam         = fr->nbfp;
 484     vdwtype          = mdatoms->typeA;
 485
 486     vftab            = kernel_data->table_vdw->data;
 487     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_vdw->scale);
 488
 489     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 490     ewtab            = fr->ic->tabq_coul_F;
 491     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 492     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 493
 494     /* Avoid stupid compiler warnings */
 495     jnrA = jnrB = 0;
 496     j_coord_offsetA = 0;
 497     j_coord_offsetB = 0;
 498
 499     outeriter        = 0;
 500     inneriter        = 0;
 501
 502     /* Start outer loop over neighborlists */
 503     for(iidx=0; iidx<nri; iidx++)
 504     {
 505         /* Load shift vector for this list */
 506         i_shift_offset   = DIM*shiftidx[iidx];
 507
 508         /* Load limits for loop over neighbors */
 509         j_index_start    = jindex[iidx];
 510         j_index_end      = jindex[iidx+1];
 511
 512         /* Get outer coordinate index */
 513         inr              = iinr[iidx];
 514         i_coord_offset   = DIM*inr;
 515
 516         /* Load i particle coords and add shift vector */
 517         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 518
 519         fix0             = _fjsp_setzero_v2r8();
 520         fiy0             = _fjsp_setzero_v2r8();
 521         fiz0             = _fjsp_setzero_v2r8();
 522
 523         /* Load parameters for i particles */
 524         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 525         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 526
 527         /* Start inner kernel loop */
 528         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 529         {
 530
 531             /* Get j neighbor index, and coordinate index */
 532             jnrA             = jjnr[jidx];
 533             jnrB             = jjnr[jidx+1];
 534             j_coord_offsetA  = DIM*jnrA;
 535             j_coord_offsetB  = DIM*jnrB;
 536
 537             /* load j atom coordinates */
 538             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 539                                               &jx0,&jy0,&jz0);
 540
 541             /* Calculate displacement vector */
 542             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 543             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 544             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 545
 546             /* Calculate squared distance and things based on it */
 547             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 548
 549             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 550
 551             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 552
 553             /* Load parameters for j particles */
 554             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 555             vdwjidx0A        = 2*vdwtype[jnrA+0];
 556             vdwjidx0B        = 2*vdwtype[jnrB+0];
 557
 558             /**************************
 559              * CALCULATE INTERACTIONS *
 560              **************************/
 561
 562             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 563
 564             /* Compute parameters for interactions between i and j atoms */
 565             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 566             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 567                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 568
 569             /* Calculate table index by multiplying r with table scale and truncate to integer */
 570             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 571             itab_tmp         = _fjsp_dtox_v2r8(rt);
 572             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 573             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 574             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 575
 576             vfconv.i[0]     *= 8;
 577             vfconv.i[1]     *= 8;
 578
 579             /* EWALD ELECTROSTATICS */
 580
 581             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 582             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 583             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 584             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 585             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 586
 587             gmx_fjsp_load_2pair_swizzle_v2r8(ewtab+ewconv.i[0],ewtab+ewconv.i[1],
 588                                          &ewtabF,&ewtabFn);
 589             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 590             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 591
 592             /* CUBIC SPLINE TABLE DISPERSION */
 593             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 594             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 595             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 596             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 597             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 2 );
 598             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 599             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 600             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 601             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 602
 603             /* CUBIC SPLINE TABLE REPULSION */
 604             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 605             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 4 );
 606             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 607             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 608             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 6 );
 609             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 610             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 611             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 612             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 613             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 614
 615             fscal            = _fjsp_add_v2r8(felec,fvdw);
 616
 617             /* Update vectorial force */
 618             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 619             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 620             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 621
 622             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 623
 624             /* Inner loop uses 65 flops */
 625         }
 626
 627         if(jidx<j_index_end)
 628         {
 629
 630             jnrA             = jjnr[jidx];
 631             j_coord_offsetA  = DIM*jnrA;
 632
 633             /* load j atom coordinates */
 634             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 635                                               &jx0,&jy0,&jz0);
 636
 637             /* Calculate displacement vector */
 638             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 639             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 640             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 641
 642             /* Calculate squared distance and things based on it */
 643             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 644
 645             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 646
 647             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 648
 649             /* Load parameters for j particles */
 650             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 651             vdwjidx0A        = 2*vdwtype[jnrA+0];
 652
 653             /**************************
 654              * CALCULATE INTERACTIONS *
 655              **************************/
 656
 657             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 658
 659             /* Compute parameters for interactions between i and j atoms */
 660             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 661             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 662                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 663
 664             /* Calculate table index by multiplying r with table scale and truncate to integer */
 665             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 666             itab_tmp         = _fjsp_dtox_v2r8(rt);
 667             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 668             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 669             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 670
 671             vfconv.i[0]     *= 8;
 672             vfconv.i[1]     *= 8;
 673
 674             /* EWALD ELECTROSTATICS */
 675
 676             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 677             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 678             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 679             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 680             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 681
 682             gmx_fjsp_load_1pair_swizzle_v2r8(ewtab+ewconv.i[0],&ewtabF,&ewtabFn);
 683             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 684             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 685
 686             /* CUBIC SPLINE TABLE DISPERSION */
 687             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 688             F                = _fjsp_setzero_v2r8();
 689             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 690             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 691             H                = _fjsp_setzero_v2r8();
 692             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 693             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 694             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 695             fvdw6            = _fjsp_mul_v2r8(c6_00,FF);
 696
 697             /* CUBIC SPLINE TABLE REPULSION */
 698             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 699             F                = _fjsp_setzero_v2r8();
 700             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 701             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 702             H                = _fjsp_setzero_v2r8();
 703             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 704             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 705             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 706             fvdw12           = _fjsp_mul_v2r8(c12_00,FF);
 707             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv00)));
 708
 709             fscal            = _fjsp_add_v2r8(felec,fvdw);
 710
 711             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 712
 713             /* Update vectorial force */
 714             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 715             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 716             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 717
 718             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 719
 720             /* Inner loop uses 65 flops */
 721         }
 722
 723         /* End of innermost loop */
 724
 725         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 726                                               f+i_coord_offset,fshift+i_shift_offset);
 727
 728         /* Increment number of inner iterations */
 729         inneriter                  += j_index_end - j_index_start;
 730
 731         /* Outer loop uses 7 flops */
 732     }
 733
 734     /* Increment number of outer iterations */
 735     outeriter        += nri;
 736
 737     /* Update outer/inner flops */
 738
 739     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*65);
 740 }