src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecEwSw_VdwLJSw_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_ElecEwSw_VdwLJSw_GeomP1P1_VF_sparc64_hpc_ace_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSw_VdwLJSw_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       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  93     real             *ewtab;
  94     _fjsp_v2r8       rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  95     real             rswitch_scalar,d_scalar;
  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     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 120     ewtab            = fr->ic->tabq_coul_FDV0;
 121     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 122     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 123
 124     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 125     rcutoff_scalar   = fr->ic->rcoulomb;
 126     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 127     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 128
 129     rswitch_scalar   = fr->ic->rcoulomb_switch;
 130     rswitch          = gmx_fjsp_set1_v2r8(rswitch_scalar);
 131     /* Setup switch parameters */
 132     d_scalar         = rcutoff_scalar-rswitch_scalar;
 133     d                = gmx_fjsp_set1_v2r8(d_scalar);
 134     swV3             = gmx_fjsp_set1_v2r8(-10.0/(d_scalar*d_scalar*d_scalar));
 135     swV4             = gmx_fjsp_set1_v2r8( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 136     swV5             = gmx_fjsp_set1_v2r8( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 137     swF2             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar));
 138     swF3             = gmx_fjsp_set1_v2r8( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 139     swF4             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 140
 141     /* Avoid stupid compiler warnings */
 142     jnrA = jnrB = 0;
 143     j_coord_offsetA = 0;
 144     j_coord_offsetB = 0;
 145
 146     outeriter        = 0;
 147     inneriter        = 0;
 148
 149     /* Start outer loop over neighborlists */
 150     for(iidx=0; iidx<nri; iidx++)
 151     {
 152         /* Load shift vector for this list */
 153         i_shift_offset   = DIM*shiftidx[iidx];
 154
 155         /* Load limits for loop over neighbors */
 156         j_index_start    = jindex[iidx];
 157         j_index_end      = jindex[iidx+1];
 158
 159         /* Get outer coordinate index */
 160         inr              = iinr[iidx];
 161         i_coord_offset   = DIM*inr;
 162
 163         /* Load i particle coords and add shift vector */
 164         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 165
 166         fix0             = _fjsp_setzero_v2r8();
 167         fiy0             = _fjsp_setzero_v2r8();
 168         fiz0             = _fjsp_setzero_v2r8();
 169
 170         /* Load parameters for i particles */
 171         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 172         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 173
 174         /* Reset potential sums */
 175         velecsum         = _fjsp_setzero_v2r8();
 176         vvdwsum          = _fjsp_setzero_v2r8();
 177
 178         /* Start inner kernel loop */
 179         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 180         {
 181
 182             /* Get j neighbor index, and coordinate index */
 183             jnrA             = jjnr[jidx];
 184             jnrB             = jjnr[jidx+1];
 185             j_coord_offsetA  = DIM*jnrA;
 186             j_coord_offsetB  = DIM*jnrB;
 187
 188             /* load j atom coordinates */
 189             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 190                                               &jx0,&jy0,&jz0);
 191
 192             /* Calculate displacement vector */
 193             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 194             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 195             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 196
 197             /* Calculate squared distance and things based on it */
 198             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 199
 200             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 201
 202             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 203
 204             /* Load parameters for j particles */
 205             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 206             vdwjidx0A        = 2*vdwtype[jnrA+0];
 207             vdwjidx0B        = 2*vdwtype[jnrB+0];
 208
 209             /**************************
 210              * CALCULATE INTERACTIONS *
 211              **************************/
 212
 213             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 214             {
 215
 216             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 217
 218             /* Compute parameters for interactions between i and j atoms */
 219             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 220             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 221                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 222
 223             /* EWALD ELECTROSTATICS */
 224
 225             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 226             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 227             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 228             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 229             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 230
 231             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 232             ewtabD           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
 233             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 234             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 235             ewtabFn          = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
 236             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 237             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 238             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 239             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 240             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 241
 242             /* LENNARD-JONES DISPERSION/REPULSION */
 243
 244             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 245             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 246             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 247             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 248             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 249
 250             d                = _fjsp_sub_v2r8(r00,rswitch);
 251             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 252             d2               = _fjsp_mul_v2r8(d,d);
 253             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 254
 255             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 256
 257             /* Evaluate switch function */
 258             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 259             felec            = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(velec,dsw)) );
 260             fvdw             = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
 261             velec            = _fjsp_mul_v2r8(velec,sw);
 262             vvdw             = _fjsp_mul_v2r8(vvdw,sw);
 263             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 264
 265             /* Update potential sum for this i atom from the interaction with this j atom. */
 266             velec            = _fjsp_and_v2r8(velec,cutoff_mask);
 267             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 268             vvdw             = _fjsp_and_v2r8(vvdw,cutoff_mask);
 269             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 270
 271             fscal            = _fjsp_add_v2r8(felec,fvdw);
 272
 273             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 274
 275             /* Update vectorial force */
 276             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 277             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 278             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 279
 280             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 281
 282             }
 283
 284             /* Inner loop uses 86 flops */
 285         }
 286
 287         if(jidx<j_index_end)
 288         {
 289
 290             jnrA             = jjnr[jidx];
 291             j_coord_offsetA  = DIM*jnrA;
 292
 293             /* load j atom coordinates */
 294             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 295                                               &jx0,&jy0,&jz0);
 296
 297             /* Calculate displacement vector */
 298             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 299             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 300             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 301
 302             /* Calculate squared distance and things based on it */
 303             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 304
 305             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 306
 307             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 308
 309             /* Load parameters for j particles */
 310             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 311             vdwjidx0A        = 2*vdwtype[jnrA+0];
 312
 313             /**************************
 314              * CALCULATE INTERACTIONS *
 315              **************************/
 316
 317             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 318             {
 319
 320             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 321
 322             /* Compute parameters for interactions between i and j atoms */
 323             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 324             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 325                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 326
 327             /* EWALD ELECTROSTATICS */
 328
 329             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 330             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 331             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 332             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 333             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 334
 335             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 336             ewtabD           = _fjsp_setzero_v2r8();
 337             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 338             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 339             ewtabFn          = _fjsp_setzero_v2r8();
 340             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 341             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 342             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 343             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 344             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 345
 346             /* LENNARD-JONES DISPERSION/REPULSION */
 347
 348             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 349             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 350             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 351             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 352             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 353
 354             d                = _fjsp_sub_v2r8(r00,rswitch);
 355             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 356             d2               = _fjsp_mul_v2r8(d,d);
 357             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 358
 359             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 360
 361             /* Evaluate switch function */
 362             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 363             felec            = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(velec,dsw)) );
 364             fvdw             = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
 365             velec            = _fjsp_mul_v2r8(velec,sw);
 366             vvdw             = _fjsp_mul_v2r8(vvdw,sw);
 367             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 368
 369             /* Update potential sum for this i atom from the interaction with this j atom. */
 370             velec            = _fjsp_and_v2r8(velec,cutoff_mask);
 371             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 372             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 373             vvdw             = _fjsp_and_v2r8(vvdw,cutoff_mask);
 374             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 375             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 376
 377             fscal            = _fjsp_add_v2r8(felec,fvdw);
 378
 379             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 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             }
 391
 392             /* Inner loop uses 86 flops */
 393         }
 394
 395         /* End of innermost loop */
 396
 397         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 398                                               f+i_coord_offset,fshift+i_shift_offset);
 399
 400         ggid                        = gid[iidx];
 401         /* Update potential energies */
 402         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 403         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 404
 405         /* Increment number of inner iterations */
 406         inneriter                  += j_index_end - j_index_start;
 407
 408         /* Outer loop uses 9 flops */
 409     }
 410
 411     /* Increment number of outer iterations */
 412     outeriter        += nri;
 413
 414     /* Update outer/inner flops */
 415
 416     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*86);
 417 }
 418 /*
 419  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_F_sparc64_hpc_ace_double
 420  * Electrostatics interaction: Ewald
 421  * VdW interaction:            LennardJones
 422  * Geometry:                   Particle-Particle
 423  * Calculate force/pot:        Force
 424  */
 425 void
 426 nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_F_sparc64_hpc_ace_double
 427                     (t_nblist                    * gmx_restrict       nlist,
 428                      rvec                        * gmx_restrict          xx,
 429                      rvec                        * gmx_restrict          ff,
 430                      struct t_forcerec           * gmx_restrict          fr,
 431                      t_mdatoms                   * gmx_restrict     mdatoms,
 432                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 433                      t_nrnb                      * gmx_restrict        nrnb)
 434 {
 435     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 436      * just 0 for non-waters.
 437      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 438      * jnr indices corresponding to data put in the four positions in the SIMD register.
 439      */
 440     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 441     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 442     int              jnrA,jnrB;
 443     int              j_coord_offsetA,j_coord_offsetB;
 444     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 445     real             rcutoff_scalar;
 446     real             *shiftvec,*fshift,*x,*f;
 447     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 448     int              vdwioffset0;
 449     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 450     int              vdwjidx0A,vdwjidx0B;
 451     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 452     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 453     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 454     real             *charge;
 455     int              nvdwtype;
 456     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 457     int              *vdwtype;
 458     real             *vdwparam;
 459     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 460     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 461     _fjsp_v2r8       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 462     real             *ewtab;
 463     _fjsp_v2r8       rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 464     real             rswitch_scalar,d_scalar;
 465     _fjsp_v2r8       itab_tmp;
 466     _fjsp_v2r8       dummy_mask,cutoff_mask;
 467     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 468     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 469     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 470
 471     x                = xx[0];
 472     f                = ff[0];
 473
 474     nri              = nlist->nri;
 475     iinr             = nlist->iinr;
 476     jindex           = nlist->jindex;
 477     jjnr             = nlist->jjnr;
 478     shiftidx         = nlist->shift;
 479     gid              = nlist->gid;
 480     shiftvec         = fr->shift_vec[0];
 481     fshift           = fr->fshift[0];
 482     facel            = gmx_fjsp_set1_v2r8(fr->ic->epsfac);
 483     charge           = mdatoms->chargeA;
 484     nvdwtype         = fr->ntype;
 485     vdwparam         = fr->nbfp;
 486     vdwtype          = mdatoms->typeA;
 487
 488     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 489     ewtab            = fr->ic->tabq_coul_FDV0;
 490     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 491     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 492
 493     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 494     rcutoff_scalar   = fr->ic->rcoulomb;
 495     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 496     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 497
 498     rswitch_scalar   = fr->ic->rcoulomb_switch;
 499     rswitch          = gmx_fjsp_set1_v2r8(rswitch_scalar);
 500     /* Setup switch parameters */
 501     d_scalar         = rcutoff_scalar-rswitch_scalar;
 502     d                = gmx_fjsp_set1_v2r8(d_scalar);
 503     swV3             = gmx_fjsp_set1_v2r8(-10.0/(d_scalar*d_scalar*d_scalar));
 504     swV4             = gmx_fjsp_set1_v2r8( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 505     swV5             = gmx_fjsp_set1_v2r8( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 506     swF2             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar));
 507     swF3             = gmx_fjsp_set1_v2r8( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 508     swF4             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 509
 510     /* Avoid stupid compiler warnings */
 511     jnrA = jnrB = 0;
 512     j_coord_offsetA = 0;
 513     j_coord_offsetB = 0;
 514
 515     outeriter        = 0;
 516     inneriter        = 0;
 517
 518     /* Start outer loop over neighborlists */
 519     for(iidx=0; iidx<nri; iidx++)
 520     {
 521         /* Load shift vector for this list */
 522         i_shift_offset   = DIM*shiftidx[iidx];
 523
 524         /* Load limits for loop over neighbors */
 525         j_index_start    = jindex[iidx];
 526         j_index_end      = jindex[iidx+1];
 527
 528         /* Get outer coordinate index */
 529         inr              = iinr[iidx];
 530         i_coord_offset   = DIM*inr;
 531
 532         /* Load i particle coords and add shift vector */
 533         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 534
 535         fix0             = _fjsp_setzero_v2r8();
 536         fiy0             = _fjsp_setzero_v2r8();
 537         fiz0             = _fjsp_setzero_v2r8();
 538
 539         /* Load parameters for i particles */
 540         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 541         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 542
 543         /* Start inner kernel loop */
 544         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 545         {
 546
 547             /* Get j neighbor index, and coordinate index */
 548             jnrA             = jjnr[jidx];
 549             jnrB             = jjnr[jidx+1];
 550             j_coord_offsetA  = DIM*jnrA;
 551             j_coord_offsetB  = DIM*jnrB;
 552
 553             /* load j atom coordinates */
 554             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 555                                               &jx0,&jy0,&jz0);
 556
 557             /* Calculate displacement vector */
 558             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 559             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 560             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 561
 562             /* Calculate squared distance and things based on it */
 563             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 564
 565             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 566
 567             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 568
 569             /* Load parameters for j particles */
 570             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 571             vdwjidx0A        = 2*vdwtype[jnrA+0];
 572             vdwjidx0B        = 2*vdwtype[jnrB+0];
 573
 574             /**************************
 575              * CALCULATE INTERACTIONS *
 576              **************************/
 577
 578             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 579             {
 580
 581             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 582
 583             /* Compute parameters for interactions between i and j atoms */
 584             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 585             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 586                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 587
 588             /* EWALD ELECTROSTATICS */
 589
 590             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 591             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 592             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 593             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 594             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 595
 596             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 597             ewtabD           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
 598             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 599             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 600             ewtabFn          = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
 601             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 602             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 603             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 604             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 605             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 606
 607             /* LENNARD-JONES DISPERSION/REPULSION */
 608
 609             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 610             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 611             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 612             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 613             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 614
 615             d                = _fjsp_sub_v2r8(r00,rswitch);
 616             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 617             d2               = _fjsp_mul_v2r8(d,d);
 618             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 619
 620             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 621
 622             /* Evaluate switch function */
 623             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 624             felec            = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(velec,dsw)) );
 625             fvdw             = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
 626             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 627
 628             fscal            = _fjsp_add_v2r8(felec,fvdw);
 629
 630             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 631
 632             /* Update vectorial force */
 633             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 634             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 635             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 636
 637             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 638
 639             }
 640
 641             /* Inner loop uses 80 flops */
 642         }
 643
 644         if(jidx<j_index_end)
 645         {
 646
 647             jnrA             = jjnr[jidx];
 648             j_coord_offsetA  = DIM*jnrA;
 649
 650             /* load j atom coordinates */
 651             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 652                                               &jx0,&jy0,&jz0);
 653
 654             /* Calculate displacement vector */
 655             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 656             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 657             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 658
 659             /* Calculate squared distance and things based on it */
 660             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 661
 662             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 663
 664             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 665
 666             /* Load parameters for j particles */
 667             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 668             vdwjidx0A        = 2*vdwtype[jnrA+0];
 669
 670             /**************************
 671              * CALCULATE INTERACTIONS *
 672              **************************/
 673
 674             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 675             {
 676
 677             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 678
 679             /* Compute parameters for interactions between i and j atoms */
 680             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 681             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 682                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 683
 684             /* EWALD ELECTROSTATICS */
 685
 686             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 687             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 688             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 689             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 690             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 691
 692             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 693             ewtabD           = _fjsp_setzero_v2r8();
 694             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 695             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 696             ewtabFn          = _fjsp_setzero_v2r8();
 697             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 698             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 699             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 700             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 701             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 702
 703             /* LENNARD-JONES DISPERSION/REPULSION */
 704
 705             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 706             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 707             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 708             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 709             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 710
 711             d                = _fjsp_sub_v2r8(r00,rswitch);
 712             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 713             d2               = _fjsp_mul_v2r8(d,d);
 714             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 715
 716             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 717
 718             /* Evaluate switch function */
 719             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 720             felec            = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(velec,dsw)) );
 721             fvdw             = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
 722             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 723
 724             fscal            = _fjsp_add_v2r8(felec,fvdw);
 725
 726             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 727
 728             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 729
 730             /* Update vectorial force */
 731             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 732             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 733             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 734
 735             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 736
 737             }
 738
 739             /* Inner loop uses 80 flops */
 740         }
 741
 742         /* End of innermost loop */
 743
 744         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 745                                               f+i_coord_offset,fshift+i_shift_offset);
 746
 747         /* Increment number of inner iterations */
 748         inneriter                  += j_index_end - j_index_start;
 749
 750         /* Outer loop uses 7 flops */
 751     }
 752
 753     /* Increment number of outer iterations */
 754     outeriter        += nri;
 755
 756     /* Update outer/inner flops */
 757
 758     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*80);
 759 }