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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "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                      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->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->rcoulomb;
 126     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 127     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 128
 129     rswitch_scalar   = fr->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_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 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             /* LENNARD-JONES DISPERSION/REPULSION */
 346
 347             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 348             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 349             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 350             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 351             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 352
 353             d                = _fjsp_sub_v2r8(r00,rswitch);
 354             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 355             d2               = _fjsp_mul_v2r8(d,d);
 356             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 357
 358             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 359
 360             /* Evaluate switch function */
 361             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 362             felec            = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(velec,dsw)) );
 363             fvdw             = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
 364             velec            = _fjsp_mul_v2r8(velec,sw);
 365             vvdw             = _fjsp_mul_v2r8(vvdw,sw);
 366             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 367
 368             /* Update potential sum for this i atom from the interaction with this j atom. */
 369             velec            = _fjsp_and_v2r8(velec,cutoff_mask);
 370             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 371             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 372             vvdw             = _fjsp_and_v2r8(vvdw,cutoff_mask);
 373             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 374             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 375
 376             fscal            = _fjsp_add_v2r8(felec,fvdw);
 377
 378             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 379
 380             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 381
 382             /* Update vectorial force */
 383             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 384             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 385             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 386
 387             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 388
 389             }
 390
 391             /* Inner loop uses 86 flops */
 392         }
 393
 394         /* End of innermost loop */
 395
 396         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 397                                               f+i_coord_offset,fshift+i_shift_offset);
 398
 399         ggid                        = gid[iidx];
 400         /* Update potential energies */
 401         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 402         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 403
 404         /* Increment number of inner iterations */
 405         inneriter                  += j_index_end - j_index_start;
 406
 407         /* Outer loop uses 9 flops */
 408     }
 409
 410     /* Increment number of outer iterations */
 411     outeriter        += nri;
 412
 413     /* Update outer/inner flops */
 414
 415     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*86);
 416 }
 417 /*
 418  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_F_sparc64_hpc_ace_double
 419  * Electrostatics interaction: Ewald
 420  * VdW interaction:            LennardJones
 421  * Geometry:                   Particle-Particle
 422  * Calculate force/pot:        Force
 423  */
 424 void
 425 nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_F_sparc64_hpc_ace_double
 426                     (t_nblist                    * gmx_restrict       nlist,
 427                      rvec                        * gmx_restrict          xx,
 428                      rvec                        * gmx_restrict          ff,
 429                      t_forcerec                  * gmx_restrict          fr,
 430                      t_mdatoms                   * gmx_restrict     mdatoms,
 431                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 432                      t_nrnb                      * gmx_restrict        nrnb)
 433 {
 434     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 435      * just 0 for non-waters.
 436      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 437      * jnr indices corresponding to data put in the four positions in the SIMD register.
 438      */
 439     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 440     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 441     int              jnrA,jnrB;
 442     int              j_coord_offsetA,j_coord_offsetB;
 443     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 444     real             rcutoff_scalar;
 445     real             *shiftvec,*fshift,*x,*f;
 446     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 447     int              vdwioffset0;
 448     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 449     int              vdwjidx0A,vdwjidx0B;
 450     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 451     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 452     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 453     real             *charge;
 454     int              nvdwtype;
 455     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 456     int              *vdwtype;
 457     real             *vdwparam;
 458     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 459     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 460     _fjsp_v2r8       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 461     real             *ewtab;
 462     _fjsp_v2r8       rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 463     real             rswitch_scalar,d_scalar;
 464     _fjsp_v2r8       itab_tmp;
 465     _fjsp_v2r8       dummy_mask,cutoff_mask;
 466     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 467     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 468     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 469
 470     x                = xx[0];
 471     f                = ff[0];
 472
 473     nri              = nlist->nri;
 474     iinr             = nlist->iinr;
 475     jindex           = nlist->jindex;
 476     jjnr             = nlist->jjnr;
 477     shiftidx         = nlist->shift;
 478     gid              = nlist->gid;
 479     shiftvec         = fr->shift_vec[0];
 480     fshift           = fr->fshift[0];
 481     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 482     charge           = mdatoms->chargeA;
 483     nvdwtype         = fr->ntype;
 484     vdwparam         = fr->nbfp;
 485     vdwtype          = mdatoms->typeA;
 486
 487     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 488     ewtab            = fr->ic->tabq_coul_FDV0;
 489     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 490     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 491
 492     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 493     rcutoff_scalar   = fr->rcoulomb;
 494     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 495     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 496
 497     rswitch_scalar   = fr->rcoulomb_switch;
 498     rswitch          = gmx_fjsp_set1_v2r8(rswitch_scalar);
 499     /* Setup switch parameters */
 500     d_scalar         = rcutoff_scalar-rswitch_scalar;
 501     d                = gmx_fjsp_set1_v2r8(d_scalar);
 502     swV3             = gmx_fjsp_set1_v2r8(-10.0/(d_scalar*d_scalar*d_scalar));
 503     swV4             = gmx_fjsp_set1_v2r8( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 504     swV5             = gmx_fjsp_set1_v2r8( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 505     swF2             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar));
 506     swF3             = gmx_fjsp_set1_v2r8( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 507     swF4             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 508
 509     /* Avoid stupid compiler warnings */
 510     jnrA = jnrB = 0;
 511     j_coord_offsetA = 0;
 512     j_coord_offsetB = 0;
 513
 514     outeriter        = 0;
 515     inneriter        = 0;
 516
 517     /* Start outer loop over neighborlists */
 518     for(iidx=0; iidx<nri; iidx++)
 519     {
 520         /* Load shift vector for this list */
 521         i_shift_offset   = DIM*shiftidx[iidx];
 522
 523         /* Load limits for loop over neighbors */
 524         j_index_start    = jindex[iidx];
 525         j_index_end      = jindex[iidx+1];
 526
 527         /* Get outer coordinate index */
 528         inr              = iinr[iidx];
 529         i_coord_offset   = DIM*inr;
 530
 531         /* Load i particle coords and add shift vector */
 532         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 533
 534         fix0             = _fjsp_setzero_v2r8();
 535         fiy0             = _fjsp_setzero_v2r8();
 536         fiz0             = _fjsp_setzero_v2r8();
 537
 538         /* Load parameters for i particles */
 539         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 540         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 541
 542         /* Start inner kernel loop */
 543         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 544         {
 545
 546             /* Get j neighbor index, and coordinate index */
 547             jnrA             = jjnr[jidx];
 548             jnrB             = jjnr[jidx+1];
 549             j_coord_offsetA  = DIM*jnrA;
 550             j_coord_offsetB  = DIM*jnrB;
 551
 552             /* load j atom coordinates */
 553             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 554                                               &jx0,&jy0,&jz0);
 555
 556             /* Calculate displacement vector */
 557             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 558             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 559             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 560
 561             /* Calculate squared distance and things based on it */
 562             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 563
 564             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 565
 566             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 567
 568             /* Load parameters for j particles */
 569             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 570             vdwjidx0A        = 2*vdwtype[jnrA+0];
 571             vdwjidx0B        = 2*vdwtype[jnrB+0];
 572
 573             /**************************
 574              * CALCULATE INTERACTIONS *
 575              **************************/
 576
 577             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 578             {
 579
 580             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 581
 582             /* Compute parameters for interactions between i and j atoms */
 583             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 584             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 585                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 586
 587             /* EWALD ELECTROSTATICS */
 588
 589             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 590             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 591             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 592             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 593             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 594
 595             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 596             ewtabD           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
 597             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 598             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 599             ewtabFn          = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
 600             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 601             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 602             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 603             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 604             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 605
 606             /* LENNARD-JONES DISPERSION/REPULSION */
 607
 608             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 609             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 610             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 611             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 612             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 613
 614             d                = _fjsp_sub_v2r8(r00,rswitch);
 615             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 616             d2               = _fjsp_mul_v2r8(d,d);
 617             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 618
 619             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 620
 621             /* Evaluate switch function */
 622             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 623             felec            = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(velec,dsw)) );
 624             fvdw             = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
 625             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 626
 627             fscal            = _fjsp_add_v2r8(felec,fvdw);
 628
 629             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 630
 631             /* Update vectorial force */
 632             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 633             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 634             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 635
 636             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 637
 638             }
 639
 640             /* Inner loop uses 80 flops */
 641         }
 642
 643         if(jidx<j_index_end)
 644         {
 645
 646             jnrA             = jjnr[jidx];
 647             j_coord_offsetA  = DIM*jnrA;
 648
 649             /* load j atom coordinates */
 650             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 651                                               &jx0,&jy0,&jz0);
 652
 653             /* Calculate displacement vector */
 654             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 655             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 656             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 657
 658             /* Calculate squared distance and things based on it */
 659             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 660
 661             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 662
 663             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 664
 665             /* Load parameters for j particles */
 666             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 667             vdwjidx0A        = 2*vdwtype[jnrA+0];
 668
 669             /**************************
 670              * CALCULATE INTERACTIONS *
 671              **************************/
 672
 673             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 674             {
 675
 676             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 677
 678             /* Compute parameters for interactions between i and j atoms */
 679             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 680             gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 681
 682             /* EWALD ELECTROSTATICS */
 683
 684             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 685             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 686             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 687             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 688             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 689
 690             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 691             ewtabD           = _fjsp_setzero_v2r8();
 692             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 693             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 694             ewtabFn          = _fjsp_setzero_v2r8();
 695             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 696             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 697             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 698             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 699             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 700
 701             /* LENNARD-JONES DISPERSION/REPULSION */
 702
 703             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 704             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 705             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 706             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 707             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 708
 709             d                = _fjsp_sub_v2r8(r00,rswitch);
 710             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 711             d2               = _fjsp_mul_v2r8(d,d);
 712             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 713
 714             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 715
 716             /* Evaluate switch function */
 717             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 718             felec            = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(velec,dsw)) );
 719             fvdw             = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
 720             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 721
 722             fscal            = _fjsp_add_v2r8(felec,fvdw);
 723
 724             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 725
 726             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 727
 728             /* Update vectorial force */
 729             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 730             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 731             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 732
 733             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 734
 735             }
 736
 737             /* Inner loop uses 80 flops */
 738         }
 739
 740         /* End of innermost loop */
 741
 742         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 743                                               f+i_coord_offset,fshift+i_shift_offset);
 744
 745         /* Increment number of inner iterations */
 746         inneriter                  += j_index_end - j_index_start;
 747
 748         /* Outer loop uses 7 flops */
 749     }
 750
 751     /* Increment number of outer iterations */
 752     outeriter        += nri;
 753
 754     /* Update outer/inner flops */
 755
 756     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*80);
 757 }