src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecEw_VdwLJEw_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_ElecEw_VdwLJEw_GeomP1P1_VF_sparc64_hpc_ace_double
  53  * Electrostatics interaction: Ewald
  54  * VdW interaction:            LJEwald
  55  * Geometry:                   Particle-Particle
  56  * Calculate force/pot:        PotentialAndForce
  57  */
  58 void
  59 nb_kernel_ElecEw_VdwLJEw_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           c6grid_00;
  95     real                 *vdwgridparam;
  96     _fjsp_v2r8           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
  97     _fjsp_v2r8           one_half = gmx_fjsp_set1_v2r8(0.5);
  98     _fjsp_v2r8           minus_one = gmx_fjsp_set1_v2r8(-1.0);
  99     _fjsp_v2r8       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 100     real             *ewtab;
 101     _fjsp_v2r8       itab_tmp;
 102     _fjsp_v2r8       dummy_mask,cutoff_mask;
 103     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 104     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 105     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 106
 107     x                = xx[0];
 108     f                = ff[0];
 109
 110     nri              = nlist->nri;
 111     iinr             = nlist->iinr;
 112     jindex           = nlist->jindex;
 113     jjnr             = nlist->jjnr;
 114     shiftidx         = nlist->shift;
 115     gid              = nlist->gid;
 116     shiftvec         = fr->shift_vec[0];
 117     fshift           = fr->fshift[0];
 118     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 119     charge           = mdatoms->chargeA;
 120     nvdwtype         = fr->ntype;
 121     vdwparam         = fr->nbfp;
 122     vdwtype          = mdatoms->typeA;
 123     vdwgridparam     = fr->ljpme_c6grid;
 124     sh_lj_ewald      = gmx_fjsp_set1_v2r8(fr->ic->sh_lj_ewald);
 125     ewclj            = gmx_fjsp_set1_v2r8(fr->ewaldcoeff_lj);
 126     ewclj2           = _fjsp_mul_v2r8(minus_one,_fjsp_mul_v2r8(ewclj,ewclj));
 127
 128     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 129     ewtab            = fr->ic->tabq_coul_FDV0;
 130     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 131     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 132
 133     /* Avoid stupid compiler warnings */
 134     jnrA = jnrB = 0;
 135     j_coord_offsetA = 0;
 136     j_coord_offsetB = 0;
 137
 138     outeriter        = 0;
 139     inneriter        = 0;
 140
 141     /* Start outer loop over neighborlists */
 142     for(iidx=0; iidx<nri; iidx++)
 143     {
 144         /* Load shift vector for this list */
 145         i_shift_offset   = DIM*shiftidx[iidx];
 146
 147         /* Load limits for loop over neighbors */
 148         j_index_start    = jindex[iidx];
 149         j_index_end      = jindex[iidx+1];
 150
 151         /* Get outer coordinate index */
 152         inr              = iinr[iidx];
 153         i_coord_offset   = DIM*inr;
 154
 155         /* Load i particle coords and add shift vector */
 156         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 157
 158         fix0             = _fjsp_setzero_v2r8();
 159         fiy0             = _fjsp_setzero_v2r8();
 160         fiz0             = _fjsp_setzero_v2r8();
 161
 162         /* Load parameters for i particles */
 163         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 164         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 165
 166         /* Reset potential sums */
 167         velecsum         = _fjsp_setzero_v2r8();
 168         vvdwsum          = _fjsp_setzero_v2r8();
 169
 170         /* Start inner kernel loop */
 171         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 172         {
 173
 174             /* Get j neighbor index, and coordinate index */
 175             jnrA             = jjnr[jidx];
 176             jnrB             = jjnr[jidx+1];
 177             j_coord_offsetA  = DIM*jnrA;
 178             j_coord_offsetB  = DIM*jnrB;
 179
 180             /* load j atom coordinates */
 181             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 182                                               &jx0,&jy0,&jz0);
 183
 184             /* Calculate displacement vector */
 185             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 186             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 187             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 188
 189             /* Calculate squared distance and things based on it */
 190             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 191
 192             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 193
 194             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 195
 196             /* Load parameters for j particles */
 197             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 198             vdwjidx0A        = 2*vdwtype[jnrA+0];
 199             vdwjidx0B        = 2*vdwtype[jnrB+0];
 200
 201             /**************************
 202              * CALCULATE INTERACTIONS *
 203              **************************/
 204
 205             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 206
 207             /* Compute parameters for interactions between i and j atoms */
 208             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 209             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 210                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 211
 212             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 213                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 214
 215             /* EWALD ELECTROSTATICS */
 216
 217             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 218             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 219             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 220             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 221             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 222
 223             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 224             ewtabD           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
 225             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 226             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 227             ewtabFn          = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
 228             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 229             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 230             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 231             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 232             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 233
 234             /* Analytical LJ-PME */
 235             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 236             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 237             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 238             exponent         = gmx_simd_exp_d(ewcljrsq);
 239             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 240             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 241             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 242             vvdw6            = _fjsp_mul_v2r8(_fjsp_madd_v2r8(c6grid_00,_fjsp_sub_v2r8(poly,one),c6_00),rinvsix);
 243             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 244             vvdw             = _fjsp_msub_v2r8(vvdw12,one_twelfth,_fjsp_mul_v2r8(vvdw6,one_sixth));
 245             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 246             fvdw             = _fjsp_mul_v2r8(_fjsp_add_v2r8(vvdw12,_fjsp_msub_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6),vvdw6)),rinvsq00);
 247
 248             /* Update potential sum for this i atom from the interaction with this j atom. */
 249             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 250             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 251
 252             fscal            = _fjsp_add_v2r8(felec,fvdw);
 253
 254             /* Update vectorial force */
 255             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 256             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 257             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 258
 259             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 260
 261             /* Inner loop uses 68 flops */
 262         }
 263
 264         if(jidx<j_index_end)
 265         {
 266
 267             jnrA             = jjnr[jidx];
 268             j_coord_offsetA  = DIM*jnrA;
 269
 270             /* load j atom coordinates */
 271             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 272                                               &jx0,&jy0,&jz0);
 273
 274             /* Calculate displacement vector */
 275             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 276             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 277             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 278
 279             /* Calculate squared distance and things based on it */
 280             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 281
 282             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 283
 284             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 285
 286             /* Load parameters for j particles */
 287             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 288             vdwjidx0A        = 2*vdwtype[jnrA+0];
 289
 290             /**************************
 291              * CALCULATE INTERACTIONS *
 292              **************************/
 293
 294             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 295
 296             /* Compute parameters for interactions between i and j atoms */
 297             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 298             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 299                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 300
 301             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 302                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 303
 304             /* EWALD ELECTROSTATICS */
 305
 306             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 307             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 308             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 309             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 310             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 311
 312             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 313             ewtabD           = _fjsp_setzero_v2r8();
 314             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 315             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 316             ewtabFn          = _fjsp_setzero_v2r8();
 317             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 318             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 319             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 320             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 321             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 322
 323             /* Analytical LJ-PME */
 324             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 325             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 326             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 327             exponent         = gmx_simd_exp_d(ewcljrsq);
 328             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 329             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 330             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 331             vvdw6            = _fjsp_mul_v2r8(_fjsp_madd_v2r8(c6grid_00,_fjsp_sub_v2r8(poly,one),c6_00),rinvsix);
 332             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 333             vvdw             = _fjsp_msub_v2r8(vvdw12,one_twelfth,_fjsp_mul_v2r8(vvdw6,one_sixth));
 334             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 335             fvdw             = _fjsp_mul_v2r8(_fjsp_add_v2r8(vvdw12,_fjsp_msub_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6),vvdw6)),rinvsq00);
 336
 337             /* Update potential sum for this i atom from the interaction with this j atom. */
 338             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 339             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 340             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 341             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 342
 343             fscal            = _fjsp_add_v2r8(felec,fvdw);
 344
 345             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 346
 347             /* Update vectorial force */
 348             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 349             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 350             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 351
 352             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 353
 354             /* Inner loop uses 68 flops */
 355         }
 356
 357         /* End of innermost loop */
 358
 359         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 360                                               f+i_coord_offset,fshift+i_shift_offset);
 361
 362         ggid                        = gid[iidx];
 363         /* Update potential energies */
 364         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 365         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 366
 367         /* Increment number of inner iterations */
 368         inneriter                  += j_index_end - j_index_start;
 369
 370         /* Outer loop uses 9 flops */
 371     }
 372
 373     /* Increment number of outer iterations */
 374     outeriter        += nri;
 375
 376     /* Update outer/inner flops */
 377
 378     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*68);
 379 }
 380 /*
 381  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomP1P1_F_sparc64_hpc_ace_double
 382  * Electrostatics interaction: Ewald
 383  * VdW interaction:            LJEwald
 384  * Geometry:                   Particle-Particle
 385  * Calculate force/pot:        Force
 386  */
 387 void
 388 nb_kernel_ElecEw_VdwLJEw_GeomP1P1_F_sparc64_hpc_ace_double
 389                     (t_nblist                    * gmx_restrict       nlist,
 390                      rvec                        * gmx_restrict          xx,
 391                      rvec                        * gmx_restrict          ff,
 392                      t_forcerec                  * gmx_restrict          fr,
 393                      t_mdatoms                   * gmx_restrict     mdatoms,
 394                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 395                      t_nrnb                      * gmx_restrict        nrnb)
 396 {
 397     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 398      * just 0 for non-waters.
 399      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 400      * jnr indices corresponding to data put in the four positions in the SIMD register.
 401      */
 402     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 403     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 404     int              jnrA,jnrB;
 405     int              j_coord_offsetA,j_coord_offsetB;
 406     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 407     real             rcutoff_scalar;
 408     real             *shiftvec,*fshift,*x,*f;
 409     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 410     int              vdwioffset0;
 411     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 412     int              vdwjidx0A,vdwjidx0B;
 413     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 414     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 415     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 416     real             *charge;
 417     int              nvdwtype;
 418     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 419     int              *vdwtype;
 420     real             *vdwparam;
 421     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 422     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 423     _fjsp_v2r8           c6grid_00;
 424     real                 *vdwgridparam;
 425     _fjsp_v2r8           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 426     _fjsp_v2r8           one_half = gmx_fjsp_set1_v2r8(0.5);
 427     _fjsp_v2r8           minus_one = gmx_fjsp_set1_v2r8(-1.0);
 428     _fjsp_v2r8       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 429     real             *ewtab;
 430     _fjsp_v2r8       itab_tmp;
 431     _fjsp_v2r8       dummy_mask,cutoff_mask;
 432     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 433     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 434     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 435
 436     x                = xx[0];
 437     f                = ff[0];
 438
 439     nri              = nlist->nri;
 440     iinr             = nlist->iinr;
 441     jindex           = nlist->jindex;
 442     jjnr             = nlist->jjnr;
 443     shiftidx         = nlist->shift;
 444     gid              = nlist->gid;
 445     shiftvec         = fr->shift_vec[0];
 446     fshift           = fr->fshift[0];
 447     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 448     charge           = mdatoms->chargeA;
 449     nvdwtype         = fr->ntype;
 450     vdwparam         = fr->nbfp;
 451     vdwtype          = mdatoms->typeA;
 452     vdwgridparam     = fr->ljpme_c6grid;
 453     sh_lj_ewald      = gmx_fjsp_set1_v2r8(fr->ic->sh_lj_ewald);
 454     ewclj            = gmx_fjsp_set1_v2r8(fr->ewaldcoeff_lj);
 455     ewclj2           = _fjsp_mul_v2r8(minus_one,_fjsp_mul_v2r8(ewclj,ewclj));
 456
 457     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 458     ewtab            = fr->ic->tabq_coul_F;
 459     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 460     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 461
 462     /* Avoid stupid compiler warnings */
 463     jnrA = jnrB = 0;
 464     j_coord_offsetA = 0;
 465     j_coord_offsetB = 0;
 466
 467     outeriter        = 0;
 468     inneriter        = 0;
 469
 470     /* Start outer loop over neighborlists */
 471     for(iidx=0; iidx<nri; iidx++)
 472     {
 473         /* Load shift vector for this list */
 474         i_shift_offset   = DIM*shiftidx[iidx];
 475
 476         /* Load limits for loop over neighbors */
 477         j_index_start    = jindex[iidx];
 478         j_index_end      = jindex[iidx+1];
 479
 480         /* Get outer coordinate index */
 481         inr              = iinr[iidx];
 482         i_coord_offset   = DIM*inr;
 483
 484         /* Load i particle coords and add shift vector */
 485         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 486
 487         fix0             = _fjsp_setzero_v2r8();
 488         fiy0             = _fjsp_setzero_v2r8();
 489         fiz0             = _fjsp_setzero_v2r8();
 490
 491         /* Load parameters for i particles */
 492         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 493         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 494
 495         /* Start inner kernel loop */
 496         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 497         {
 498
 499             /* Get j neighbor index, and coordinate index */
 500             jnrA             = jjnr[jidx];
 501             jnrB             = jjnr[jidx+1];
 502             j_coord_offsetA  = DIM*jnrA;
 503             j_coord_offsetB  = DIM*jnrB;
 504
 505             /* load j atom coordinates */
 506             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 507                                               &jx0,&jy0,&jz0);
 508
 509             /* Calculate displacement vector */
 510             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 511             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 512             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 513
 514             /* Calculate squared distance and things based on it */
 515             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 516
 517             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 518
 519             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 520
 521             /* Load parameters for j particles */
 522             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 523             vdwjidx0A        = 2*vdwtype[jnrA+0];
 524             vdwjidx0B        = 2*vdwtype[jnrB+0];
 525
 526             /**************************
 527              * CALCULATE INTERACTIONS *
 528              **************************/
 529
 530             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 531
 532             /* Compute parameters for interactions between i and j atoms */
 533             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 534             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 535                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 536
 537             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 538                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 539
 540             /* EWALD ELECTROSTATICS */
 541
 542             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 543             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 544             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 545             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 546             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 547
 548             gmx_fjsp_load_2pair_swizzle_v2r8(ewtab+ewconv.i[0],ewtab+ewconv.i[1],
 549                                          &ewtabF,&ewtabFn);
 550             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 551             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 552
 553             /* Analytical LJ-PME */
 554             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 555             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 556             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 557             exponent         = gmx_simd_exp_d(ewcljrsq);
 558             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 559             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 560             /* f6A = 6 * C6grid * (1 - poly) */
 561             f6A              = _fjsp_mul_v2r8(c6grid_00,_fjsp_sub_v2r8(one,poly));
 562             /* f6B = C6grid * exponent * beta^6 */
 563             f6B              = _fjsp_mul_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6));
 564             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 565             fvdw              = _fjsp_mul_v2r8(_fjsp_madd_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,_fjsp_sub_v2r8(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 566
 567             fscal            = _fjsp_add_v2r8(felec,fvdw);
 568
 569             /* Update vectorial force */
 570             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 571             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 572             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 573
 574             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 575
 576             /* Inner loop uses 61 flops */
 577         }
 578
 579         if(jidx<j_index_end)
 580         {
 581
 582             jnrA             = jjnr[jidx];
 583             j_coord_offsetA  = DIM*jnrA;
 584
 585             /* load j atom coordinates */
 586             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 587                                               &jx0,&jy0,&jz0);
 588
 589             /* Calculate displacement vector */
 590             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 591             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 592             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 593
 594             /* Calculate squared distance and things based on it */
 595             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 596
 597             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 598
 599             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 600
 601             /* Load parameters for j particles */
 602             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 603             vdwjidx0A        = 2*vdwtype[jnrA+0];
 604
 605             /**************************
 606              * CALCULATE INTERACTIONS *
 607              **************************/
 608
 609             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 610
 611             /* Compute parameters for interactions between i and j atoms */
 612             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 613             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 614                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 615
 616             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 617                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 618
 619             /* EWALD ELECTROSTATICS */
 620
 621             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 622             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 623             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 624             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 625             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 626
 627             gmx_fjsp_load_1pair_swizzle_v2r8(ewtab+ewconv.i[0],&ewtabF,&ewtabFn);
 628             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 629             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 630
 631             /* Analytical LJ-PME */
 632             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 633             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 634             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 635             exponent         = gmx_simd_exp_d(ewcljrsq);
 636             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 637             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 638             /* f6A = 6 * C6grid * (1 - poly) */
 639             f6A              = _fjsp_mul_v2r8(c6grid_00,_fjsp_sub_v2r8(one,poly));
 640             /* f6B = C6grid * exponent * beta^6 */
 641             f6B              = _fjsp_mul_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6));
 642             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 643             fvdw              = _fjsp_mul_v2r8(_fjsp_madd_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,_fjsp_sub_v2r8(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 644
 645             fscal            = _fjsp_add_v2r8(felec,fvdw);
 646
 647             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 648
 649             /* Update vectorial force */
 650             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 651             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 652             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 653
 654             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 655
 656             /* Inner loop uses 61 flops */
 657         }
 658
 659         /* End of innermost loop */
 660
 661         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 662                                               f+i_coord_offset,fshift+i_shift_offset);
 663
 664         /* Increment number of inner iterations */
 665         inneriter                  += j_index_end - j_index_start;
 666
 667         /* Outer loop uses 7 flops */
 668     }
 669
 670     /* Increment number of outer iterations */
 671     outeriter        += nri;
 672
 673     /* Update outer/inner flops */
 674
 675     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*61);
 676 }