src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_sparc64_hpc_ace_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sparc64_hpc_ace_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "kernelutil_sparc64_hpc_ace_double.h"
  50
  51 /*
  52  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_VF_sparc64_hpc_ace_double
  53  * Electrostatics interaction: Ewald
  54  * VdW interaction:            LennardJones
  55  * Geometry:                   Particle-Particle
  56  * Calculate force/pot:        PotentialAndForce
  57  */
  58 void
  59 nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_VF_sparc64_hpc_ace_double
  60                     (t_nblist                    * gmx_restrict       nlist,
  61                      rvec                        * gmx_restrict          xx,
  62                      rvec                        * gmx_restrict          ff,
  63                      t_forcerec                  * gmx_restrict          fr,
  64                      t_mdatoms                   * gmx_restrict     mdatoms,
  65                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  66                      t_nrnb                      * gmx_restrict        nrnb)
  67 {
  68     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  69      * just 0 for non-waters.
  70      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
  71      * jnr indices corresponding to data put in the four positions in the SIMD register.
  72      */
  73     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  74     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  75     int              jnrA,jnrB;
  76     int              j_coord_offsetA,j_coord_offsetB;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  81     int              vdwioffset0;
  82     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  83     int              vdwjidx0A,vdwjidx0B;
  84     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  85     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  86     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
  87     real             *charge;
  88     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       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  95     real             *ewtab;
  96     _fjsp_v2r8       rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  97     real             rswitch_scalar,d_scalar;
  98     _fjsp_v2r8       itab_tmp;
  99     _fjsp_v2r8       dummy_mask,cutoff_mask;
 100     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 101     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 102     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 103
 104     x                = xx[0];
 105     f                = ff[0];
 106
 107     nri              = nlist->nri;
 108     iinr             = nlist->iinr;
 109     jindex           = nlist->jindex;
 110     jjnr             = nlist->jjnr;
 111     shiftidx         = nlist->shift;
 112     gid              = nlist->gid;
 113     shiftvec         = fr->shift_vec[0];
 114     fshift           = fr->fshift[0];
 115     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 116     charge           = mdatoms->chargeA;
 117     nvdwtype         = fr->ntype;
 118     vdwparam         = fr->nbfp;
 119     vdwtype          = mdatoms->typeA;
 120
 121     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 122     ewtab            = fr->ic->tabq_coul_FDV0;
 123     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 124     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 125
 126     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 127     rcutoff_scalar   = fr->rcoulomb;
 128     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 129     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 130
 131     rswitch_scalar   = fr->rcoulomb_switch;
 132     rswitch          = gmx_fjsp_set1_v2r8(rswitch_scalar);
 133     /* Setup switch parameters */
 134     d_scalar         = rcutoff_scalar-rswitch_scalar;
 135     d                = gmx_fjsp_set1_v2r8(d_scalar);
 136     swV3             = gmx_fjsp_set1_v2r8(-10.0/(d_scalar*d_scalar*d_scalar));
 137     swV4             = gmx_fjsp_set1_v2r8( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 138     swV5             = gmx_fjsp_set1_v2r8( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 139     swF2             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar));
 140     swF3             = gmx_fjsp_set1_v2r8( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 141     swF4             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 142
 143     /* Avoid stupid compiler warnings */
 144     jnrA = jnrB = 0;
 145     j_coord_offsetA = 0;
 146     j_coord_offsetB = 0;
 147
 148     outeriter        = 0;
 149     inneriter        = 0;
 150
 151     /* Start outer loop over neighborlists */
 152     for(iidx=0; iidx<nri; iidx++)
 153     {
 154         /* Load shift vector for this list */
 155         i_shift_offset   = DIM*shiftidx[iidx];
 156
 157         /* Load limits for loop over neighbors */
 158         j_index_start    = jindex[iidx];
 159         j_index_end      = jindex[iidx+1];
 160
 161         /* Get outer coordinate index */
 162         inr              = iinr[iidx];
 163         i_coord_offset   = DIM*inr;
 164
 165         /* Load i particle coords and add shift vector */
 166         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 167
 168         fix0             = _fjsp_setzero_v2r8();
 169         fiy0             = _fjsp_setzero_v2r8();
 170         fiz0             = _fjsp_setzero_v2r8();
 171
 172         /* Load parameters for i particles */
 173         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 174         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 175
 176         /* Reset potential sums */
 177         velecsum         = _fjsp_setzero_v2r8();
 178         vvdwsum          = _fjsp_setzero_v2r8();
 179
 180         /* Start inner kernel loop */
 181         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 182         {
 183
 184             /* Get j neighbor index, and coordinate index */
 185             jnrA             = jjnr[jidx];
 186             jnrB             = jjnr[jidx+1];
 187             j_coord_offsetA  = DIM*jnrA;
 188             j_coord_offsetB  = DIM*jnrB;
 189
 190             /* load j atom coordinates */
 191             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 192                                               &jx0,&jy0,&jz0);
 193
 194             /* Calculate displacement vector */
 195             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 196             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 197             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 198
 199             /* Calculate squared distance and things based on it */
 200             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 201
 202             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 203
 204             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 205
 206             /* Load parameters for j particles */
 207             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 208             vdwjidx0A        = 2*vdwtype[jnrA+0];
 209             vdwjidx0B        = 2*vdwtype[jnrB+0];
 210
 211             /**************************
 212              * CALCULATE INTERACTIONS *
 213              **************************/
 214
 215             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 216             {
 217
 218             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 219
 220             /* Compute parameters for interactions between i and j atoms */
 221             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 222             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 223                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 224
 225             /* EWALD ELECTROSTATICS */
 226
 227             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 228             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 229             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 230             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 231             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 232
 233             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 234             ewtabD           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
 235             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 236             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 237             ewtabFn          = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
 238             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 239             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 240             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 241             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 242             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 243
 244             /* LENNARD-JONES DISPERSION/REPULSION */
 245
 246             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 247             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 248             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 249             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 250             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 251
 252             d                = _fjsp_sub_v2r8(r00,rswitch);
 253             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 254             d2               = _fjsp_mul_v2r8(d,d);
 255             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 256
 257             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 258
 259             /* Evaluate switch function */
 260             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 261             felec            = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(velec,dsw)) );
 262             fvdw             = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
 263             velec            = _fjsp_mul_v2r8(velec,sw);
 264             vvdw             = _fjsp_mul_v2r8(vvdw,sw);
 265             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 266
 267             /* Update potential sum for this i atom from the interaction with this j atom. */
 268             velec            = _fjsp_and_v2r8(velec,cutoff_mask);
 269             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 270             vvdw             = _fjsp_and_v2r8(vvdw,cutoff_mask);
 271             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 272
 273             fscal            = _fjsp_add_v2r8(felec,fvdw);
 274
 275             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 276
 277             /* Update vectorial force */
 278             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 279             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 280             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 281
 282             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 283
 284             }
 285
 286             /* Inner loop uses 86 flops */
 287         }
 288
 289         if(jidx<j_index_end)
 290         {
 291
 292             jnrA             = jjnr[jidx];
 293             j_coord_offsetA  = DIM*jnrA;
 294
 295             /* load j atom coordinates */
 296             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 297                                               &jx0,&jy0,&jz0);
 298
 299             /* Calculate displacement vector */
 300             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 301             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 302             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 303
 304             /* Calculate squared distance and things based on it */
 305             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 306
 307             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 308
 309             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 310
 311             /* Load parameters for j particles */
 312             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 313             vdwjidx0A        = 2*vdwtype[jnrA+0];
 314
 315             /**************************
 316              * CALCULATE INTERACTIONS *
 317              **************************/
 318
 319             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 320             {
 321
 322             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 323
 324             /* Compute parameters for interactions between i and j atoms */
 325             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 326             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 327                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 328
 329             /* EWALD ELECTROSTATICS */
 330
 331             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 332             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 333             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 334             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 335             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 336
 337             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 338             ewtabD           = _fjsp_setzero_v2r8();
 339             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 340             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 341             ewtabFn          = _fjsp_setzero_v2r8();
 342             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 343             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 344             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 345             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 346             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 347
 348             /* LENNARD-JONES DISPERSION/REPULSION */
 349
 350             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 351             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 352             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 353             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 354             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 355
 356             d                = _fjsp_sub_v2r8(r00,rswitch);
 357             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 358             d2               = _fjsp_mul_v2r8(d,d);
 359             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 360
 361             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 362
 363             /* Evaluate switch function */
 364             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 365             felec            = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(velec,dsw)) );
 366             fvdw             = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
 367             velec            = _fjsp_mul_v2r8(velec,sw);
 368             vvdw             = _fjsp_mul_v2r8(vvdw,sw);
 369             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 370
 371             /* Update potential sum for this i atom from the interaction with this j atom. */
 372             velec            = _fjsp_and_v2r8(velec,cutoff_mask);
 373             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 374             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 375             vvdw             = _fjsp_and_v2r8(vvdw,cutoff_mask);
 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_and_v2r8(fscal,cutoff_mask);
 382
 383             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 384
 385             /* Update vectorial force */
 386             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 387             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 388             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 389
 390             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 391
 392             }
 393
 394             /* Inner loop uses 86 flops */
 395         }
 396
 397         /* End of innermost loop */
 398
 399         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 400                                               f+i_coord_offset,fshift+i_shift_offset);
 401
 402         ggid                        = gid[iidx];
 403         /* Update potential energies */
 404         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 405         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 406
 407         /* Increment number of inner iterations */
 408         inneriter                  += j_index_end - j_index_start;
 409
 410         /* Outer loop uses 9 flops */
 411     }
 412
 413     /* Increment number of outer iterations */
 414     outeriter        += nri;
 415
 416     /* Update outer/inner flops */
 417
 418     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*86);
 419 }
 420 /*
 421  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_F_sparc64_hpc_ace_double
 422  * Electrostatics interaction: Ewald
 423  * VdW interaction:            LennardJones
 424  * Geometry:                   Particle-Particle
 425  * Calculate force/pot:        Force
 426  */
 427 void
 428 nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_F_sparc64_hpc_ace_double
 429                     (t_nblist                    * gmx_restrict       nlist,
 430                      rvec                        * gmx_restrict          xx,
 431                      rvec                        * gmx_restrict          ff,
 432                      t_forcerec                  * gmx_restrict          fr,
 433                      t_mdatoms                   * gmx_restrict     mdatoms,
 434                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 435                      t_nrnb                      * gmx_restrict        nrnb)
 436 {
 437     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 438      * just 0 for non-waters.
 439      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 440      * jnr indices corresponding to data put in the four positions in the SIMD register.
 441      */
 442     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 443     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 444     int              jnrA,jnrB;
 445     int              j_coord_offsetA,j_coord_offsetB;
 446     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 447     real             rcutoff_scalar;
 448     real             *shiftvec,*fshift,*x,*f;
 449     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 450     int              vdwioffset0;
 451     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 452     int              vdwjidx0A,vdwjidx0B;
 453     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 454     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 455     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 456     real             *charge;
 457     int              nvdwtype;
 458     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 459     int              *vdwtype;
 460     real             *vdwparam;
 461     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 462     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 463     _fjsp_v2r8       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 464     real             *ewtab;
 465     _fjsp_v2r8       rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 466     real             rswitch_scalar,d_scalar;
 467     _fjsp_v2r8       itab_tmp;
 468     _fjsp_v2r8       dummy_mask,cutoff_mask;
 469     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 470     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 471     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 472
 473     x                = xx[0];
 474     f                = ff[0];
 475
 476     nri              = nlist->nri;
 477     iinr             = nlist->iinr;
 478     jindex           = nlist->jindex;
 479     jjnr             = nlist->jjnr;
 480     shiftidx         = nlist->shift;
 481     gid              = nlist->gid;
 482     shiftvec         = fr->shift_vec[0];
 483     fshift           = fr->fshift[0];
 484     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 485     charge           = mdatoms->chargeA;
 486     nvdwtype         = fr->ntype;
 487     vdwparam         = fr->nbfp;
 488     vdwtype          = mdatoms->typeA;
 489
 490     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 491     ewtab            = fr->ic->tabq_coul_FDV0;
 492     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 493     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 494
 495     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 496     rcutoff_scalar   = fr->rcoulomb;
 497     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 498     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 499
 500     rswitch_scalar   = fr->rcoulomb_switch;
 501     rswitch          = gmx_fjsp_set1_v2r8(rswitch_scalar);
 502     /* Setup switch parameters */
 503     d_scalar         = rcutoff_scalar-rswitch_scalar;
 504     d                = gmx_fjsp_set1_v2r8(d_scalar);
 505     swV3             = gmx_fjsp_set1_v2r8(-10.0/(d_scalar*d_scalar*d_scalar));
 506     swV4             = gmx_fjsp_set1_v2r8( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 507     swV5             = gmx_fjsp_set1_v2r8( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 508     swF2             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar));
 509     swF3             = gmx_fjsp_set1_v2r8( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 510     swF4             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 511
 512     /* Avoid stupid compiler warnings */
 513     jnrA = jnrB = 0;
 514     j_coord_offsetA = 0;
 515     j_coord_offsetB = 0;
 516
 517     outeriter        = 0;
 518     inneriter        = 0;
 519
 520     /* Start outer loop over neighborlists */
 521     for(iidx=0; iidx<nri; iidx++)
 522     {
 523         /* Load shift vector for this list */
 524         i_shift_offset   = DIM*shiftidx[iidx];
 525
 526         /* Load limits for loop over neighbors */
 527         j_index_start    = jindex[iidx];
 528         j_index_end      = jindex[iidx+1];
 529
 530         /* Get outer coordinate index */
 531         inr              = iinr[iidx];
 532         i_coord_offset   = DIM*inr;
 533
 534         /* Load i particle coords and add shift vector */
 535         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 536
 537         fix0             = _fjsp_setzero_v2r8();
 538         fiy0             = _fjsp_setzero_v2r8();
 539         fiz0             = _fjsp_setzero_v2r8();
 540
 541         /* Load parameters for i particles */
 542         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 543         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 544
 545         /* Start inner kernel loop */
 546         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 547         {
 548
 549             /* Get j neighbor index, and coordinate index */
 550             jnrA             = jjnr[jidx];
 551             jnrB             = jjnr[jidx+1];
 552             j_coord_offsetA  = DIM*jnrA;
 553             j_coord_offsetB  = DIM*jnrB;
 554
 555             /* load j atom coordinates */
 556             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 557                                               &jx0,&jy0,&jz0);
 558
 559             /* Calculate displacement vector */
 560             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 561             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 562             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 563
 564             /* Calculate squared distance and things based on it */
 565             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 566
 567             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 568
 569             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 570
 571             /* Load parameters for j particles */
 572             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 573             vdwjidx0A        = 2*vdwtype[jnrA+0];
 574             vdwjidx0B        = 2*vdwtype[jnrB+0];
 575
 576             /**************************
 577              * CALCULATE INTERACTIONS *
 578              **************************/
 579
 580             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 581             {
 582
 583             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 584
 585             /* Compute parameters for interactions between i and j atoms */
 586             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 587             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 588                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 589
 590             /* EWALD ELECTROSTATICS */
 591
 592             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 593             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 594             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 595             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 596             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 597
 598             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 599             ewtabD           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
 600             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 601             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 602             ewtabFn          = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
 603             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 604             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 605             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 606             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 607             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 608
 609             /* LENNARD-JONES DISPERSION/REPULSION */
 610
 611             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 612             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 613             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 614             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 615             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 616
 617             d                = _fjsp_sub_v2r8(r00,rswitch);
 618             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 619             d2               = _fjsp_mul_v2r8(d,d);
 620             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 621
 622             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 623
 624             /* Evaluate switch function */
 625             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 626             felec            = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(velec,dsw)) );
 627             fvdw             = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
 628             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 629
 630             fscal            = _fjsp_add_v2r8(felec,fvdw);
 631
 632             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 633
 634             /* Update vectorial force */
 635             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 636             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 637             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 638
 639             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 640
 641             }
 642
 643             /* Inner loop uses 80 flops */
 644         }
 645
 646         if(jidx<j_index_end)
 647         {
 648
 649             jnrA             = jjnr[jidx];
 650             j_coord_offsetA  = DIM*jnrA;
 651
 652             /* load j atom coordinates */
 653             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 654                                               &jx0,&jy0,&jz0);
 655
 656             /* Calculate displacement vector */
 657             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 658             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 659             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 660
 661             /* Calculate squared distance and things based on it */
 662             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 663
 664             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 665
 666             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 667
 668             /* Load parameters for j particles */
 669             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 670             vdwjidx0A        = 2*vdwtype[jnrA+0];
 671
 672             /**************************
 673              * CALCULATE INTERACTIONS *
 674              **************************/
 675
 676             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 677             {
 678
 679             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 680
 681             /* Compute parameters for interactions between i and j atoms */
 682             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 683             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 684                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 685
 686             /* EWALD ELECTROSTATICS */
 687
 688             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 689             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 690             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 691             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 692             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 693
 694             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 695             ewtabD           = _fjsp_setzero_v2r8();
 696             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 697             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 698             ewtabFn          = _fjsp_setzero_v2r8();
 699             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 700             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 701             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 702             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 703             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 704
 705             /* LENNARD-JONES DISPERSION/REPULSION */
 706
 707             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 708             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 709             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 710             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 711             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 712
 713             d                = _fjsp_sub_v2r8(r00,rswitch);
 714             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 715             d2               = _fjsp_mul_v2r8(d,d);
 716             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 717
 718             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 719
 720             /* Evaluate switch function */
 721             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 722             felec            = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(velec,dsw)) );
 723             fvdw             = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
 724             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 725
 726             fscal            = _fjsp_add_v2r8(felec,fvdw);
 727
 728             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 729
 730             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 731
 732             /* Update vectorial force */
 733             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 734             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 735             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 736
 737             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 738
 739             }
 740
 741             /* Inner loop uses 80 flops */
 742         }
 743
 744         /* End of innermost loop */
 745
 746         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 747                                               f+i_coord_offset,fshift+i_shift_offset);
 748
 749         /* Increment number of inner iterations */
 750         inneriter                  += j_index_end - j_index_start;
 751
 752         /* Outer loop uses 7 flops */
 753     }
 754
 755     /* Increment number of outer iterations */
 756     outeriter        += nri;
 757
 758     /* Update outer/inner flops */
 759
 760     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*80);
 761 }