src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecEw_VdwLJEw_GeomP1P1_sparc64_hpc_ace_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sparc64_hpc_ace_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_sparc64_hpc_ace_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomP1P1_VF_sparc64_hpc_ace_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LJEwald
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwLJEw_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           c6grid_00;
  93     real                 *vdwgridparam;
  94     _fjsp_v2r8           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
  95     _fjsp_v2r8           one_half = gmx_fjsp_set1_v2r8(0.5);
  96     _fjsp_v2r8           minus_one = gmx_fjsp_set1_v2r8(-1.0);
  97     _fjsp_v2r8       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  98     real             *ewtab;
  99     _fjsp_v2r8       itab_tmp;
 100     _fjsp_v2r8       dummy_mask,cutoff_mask;
 101     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 102     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 103     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 104
 105     x                = xx[0];
 106     f                = ff[0];
 107
 108     nri              = nlist->nri;
 109     iinr             = nlist->iinr;
 110     jindex           = nlist->jindex;
 111     jjnr             = nlist->jjnr;
 112     shiftidx         = nlist->shift;
 113     gid              = nlist->gid;
 114     shiftvec         = fr->shift_vec[0];
 115     fshift           = fr->fshift[0];
 116     facel            = gmx_fjsp_set1_v2r8(fr->ic->epsfac);
 117     charge           = mdatoms->chargeA;
 118     nvdwtype         = fr->ntype;
 119     vdwparam         = fr->nbfp;
 120     vdwtype          = mdatoms->typeA;
 121     vdwgridparam     = fr->ljpme_c6grid;
 122     sh_lj_ewald      = gmx_fjsp_set1_v2r8(fr->ic->sh_lj_ewald);
 123     ewclj            = gmx_fjsp_set1_v2r8(fr->ic->ewaldcoeff_lj);
 124     ewclj2           = _fjsp_mul_v2r8(minus_one,_fjsp_mul_v2r8(ewclj,ewclj));
 125
 126     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 127     ewtab            = fr->ic->tabq_coul_FDV0;
 128     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 129     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 130
 131     /* Avoid stupid compiler warnings */
 132     jnrA = jnrB = 0;
 133     j_coord_offsetA = 0;
 134     j_coord_offsetB = 0;
 135
 136     outeriter        = 0;
 137     inneriter        = 0;
 138
 139     /* Start outer loop over neighborlists */
 140     for(iidx=0; iidx<nri; iidx++)
 141     {
 142         /* Load shift vector for this list */
 143         i_shift_offset   = DIM*shiftidx[iidx];
 144
 145         /* Load limits for loop over neighbors */
 146         j_index_start    = jindex[iidx];
 147         j_index_end      = jindex[iidx+1];
 148
 149         /* Get outer coordinate index */
 150         inr              = iinr[iidx];
 151         i_coord_offset   = DIM*inr;
 152
 153         /* Load i particle coords and add shift vector */
 154         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 155
 156         fix0             = _fjsp_setzero_v2r8();
 157         fiy0             = _fjsp_setzero_v2r8();
 158         fiz0             = _fjsp_setzero_v2r8();
 159
 160         /* Load parameters for i particles */
 161         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 162         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 163
 164         /* Reset potential sums */
 165         velecsum         = _fjsp_setzero_v2r8();
 166         vvdwsum          = _fjsp_setzero_v2r8();
 167
 168         /* Start inner kernel loop */
 169         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 170         {
 171
 172             /* Get j neighbor index, and coordinate index */
 173             jnrA             = jjnr[jidx];
 174             jnrB             = jjnr[jidx+1];
 175             j_coord_offsetA  = DIM*jnrA;
 176             j_coord_offsetB  = DIM*jnrB;
 177
 178             /* load j atom coordinates */
 179             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 180                                               &jx0,&jy0,&jz0);
 181
 182             /* Calculate displacement vector */
 183             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 184             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 185             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 186
 187             /* Calculate squared distance and things based on it */
 188             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 189
 190             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 191
 192             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 193
 194             /* Load parameters for j particles */
 195             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 196             vdwjidx0A        = 2*vdwtype[jnrA+0];
 197             vdwjidx0B        = 2*vdwtype[jnrB+0];
 198
 199             /**************************
 200              * CALCULATE INTERACTIONS *
 201              **************************/
 202
 203             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 204
 205             /* Compute parameters for interactions between i and j atoms */
 206             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 207             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 208                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 209
 210             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 211                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 212
 213             /* EWALD ELECTROSTATICS */
 214
 215             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 216             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 217             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 218             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 219             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 220
 221             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 222             ewtabD           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
 223             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 224             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 225             ewtabFn          = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
 226             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 227             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 228             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 229             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 230             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 231
 232             /* Analytical LJ-PME */
 233             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 234             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 235             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 236             exponent         = gmx_simd_exp_d(ewcljrsq);
 237             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 238             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 239             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 240             vvdw6            = _fjsp_mul_v2r8(_fjsp_madd_v2r8(c6grid_00,_fjsp_sub_v2r8(poly,one),c6_00),rinvsix);
 241             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 242             vvdw             = _fjsp_msub_v2r8(vvdw12,one_twelfth,_fjsp_mul_v2r8(vvdw6,one_sixth));
 243             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 244             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);
 245
 246             /* Update potential sum for this i atom from the interaction with this j atom. */
 247             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 248             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 249
 250             fscal            = _fjsp_add_v2r8(felec,fvdw);
 251
 252             /* Update vectorial force */
 253             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 254             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 255             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 256
 257             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 258
 259             /* Inner loop uses 68 flops */
 260         }
 261
 262         if(jidx<j_index_end)
 263         {
 264
 265             jnrA             = jjnr[jidx];
 266             j_coord_offsetA  = DIM*jnrA;
 267
 268             /* load j atom coordinates */
 269             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 270                                               &jx0,&jy0,&jz0);
 271
 272             /* Calculate displacement vector */
 273             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 274             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 275             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 276
 277             /* Calculate squared distance and things based on it */
 278             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 279
 280             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 281
 282             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 283
 284             /* Load parameters for j particles */
 285             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 286             vdwjidx0A        = 2*vdwtype[jnrA+0];
 287
 288             /**************************
 289              * CALCULATE INTERACTIONS *
 290              **************************/
 291
 292             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 293
 294             /* Compute parameters for interactions between i and j atoms */
 295             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 296             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 297                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 298
 299             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 300                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 301
 302             /* EWALD ELECTROSTATICS */
 303
 304             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 305             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 306             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 307             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 308             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 309
 310             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 311             ewtabD           = _fjsp_setzero_v2r8();
 312             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 313             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 314             ewtabFn          = _fjsp_setzero_v2r8();
 315             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 316             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 317             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 318             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 319             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 320
 321             /* Analytical LJ-PME */
 322             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 323             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 324             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 325             exponent         = gmx_simd_exp_d(ewcljrsq);
 326             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 327             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 328             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 329             vvdw6            = _fjsp_mul_v2r8(_fjsp_madd_v2r8(c6grid_00,_fjsp_sub_v2r8(poly,one),c6_00),rinvsix);
 330             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 331             vvdw             = _fjsp_msub_v2r8(vvdw12,one_twelfth,_fjsp_mul_v2r8(vvdw6,one_sixth));
 332             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 333             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);
 334
 335             /* Update potential sum for this i atom from the interaction with this j atom. */
 336             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 337             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 338             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 339             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 340
 341             fscal            = _fjsp_add_v2r8(felec,fvdw);
 342
 343             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 344
 345             /* Update vectorial force */
 346             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 347             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 348             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 349
 350             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 351
 352             /* Inner loop uses 68 flops */
 353         }
 354
 355         /* End of innermost loop */
 356
 357         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 358                                               f+i_coord_offset,fshift+i_shift_offset);
 359
 360         ggid                        = gid[iidx];
 361         /* Update potential energies */
 362         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 363         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 364
 365         /* Increment number of inner iterations */
 366         inneriter                  += j_index_end - j_index_start;
 367
 368         /* Outer loop uses 9 flops */
 369     }
 370
 371     /* Increment number of outer iterations */
 372     outeriter        += nri;
 373
 374     /* Update outer/inner flops */
 375
 376     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*68);
 377 }
 378 /*
 379  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomP1P1_F_sparc64_hpc_ace_double
 380  * Electrostatics interaction: Ewald
 381  * VdW interaction:            LJEwald
 382  * Geometry:                   Particle-Particle
 383  * Calculate force/pot:        Force
 384  */
 385 void
 386 nb_kernel_ElecEw_VdwLJEw_GeomP1P1_F_sparc64_hpc_ace_double
 387                     (t_nblist                    * gmx_restrict       nlist,
 388                      rvec                        * gmx_restrict          xx,
 389                      rvec                        * gmx_restrict          ff,
 390                      struct t_forcerec           * gmx_restrict          fr,
 391                      t_mdatoms                   * gmx_restrict     mdatoms,
 392                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 393                      t_nrnb                      * gmx_restrict        nrnb)
 394 {
 395     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 396      * just 0 for non-waters.
 397      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 398      * jnr indices corresponding to data put in the four positions in the SIMD register.
 399      */
 400     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 401     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 402     int              jnrA,jnrB;
 403     int              j_coord_offsetA,j_coord_offsetB;
 404     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 405     real             rcutoff_scalar;
 406     real             *shiftvec,*fshift,*x,*f;
 407     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 408     int              vdwioffset0;
 409     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 410     int              vdwjidx0A,vdwjidx0B;
 411     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 412     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 413     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 414     real             *charge;
 415     int              nvdwtype;
 416     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 417     int              *vdwtype;
 418     real             *vdwparam;
 419     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 420     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 421     _fjsp_v2r8           c6grid_00;
 422     real                 *vdwgridparam;
 423     _fjsp_v2r8           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 424     _fjsp_v2r8           one_half = gmx_fjsp_set1_v2r8(0.5);
 425     _fjsp_v2r8           minus_one = gmx_fjsp_set1_v2r8(-1.0);
 426     _fjsp_v2r8       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 427     real             *ewtab;
 428     _fjsp_v2r8       itab_tmp;
 429     _fjsp_v2r8       dummy_mask,cutoff_mask;
 430     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 431     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 432     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 433
 434     x                = xx[0];
 435     f                = ff[0];
 436
 437     nri              = nlist->nri;
 438     iinr             = nlist->iinr;
 439     jindex           = nlist->jindex;
 440     jjnr             = nlist->jjnr;
 441     shiftidx         = nlist->shift;
 442     gid              = nlist->gid;
 443     shiftvec         = fr->shift_vec[0];
 444     fshift           = fr->fshift[0];
 445     facel            = gmx_fjsp_set1_v2r8(fr->ic->epsfac);
 446     charge           = mdatoms->chargeA;
 447     nvdwtype         = fr->ntype;
 448     vdwparam         = fr->nbfp;
 449     vdwtype          = mdatoms->typeA;
 450     vdwgridparam     = fr->ljpme_c6grid;
 451     sh_lj_ewald      = gmx_fjsp_set1_v2r8(fr->ic->sh_lj_ewald);
 452     ewclj            = gmx_fjsp_set1_v2r8(fr->ic->ewaldcoeff_lj);
 453     ewclj2           = _fjsp_mul_v2r8(minus_one,_fjsp_mul_v2r8(ewclj,ewclj));
 454
 455     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 456     ewtab            = fr->ic->tabq_coul_F;
 457     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 458     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 459
 460     /* Avoid stupid compiler warnings */
 461     jnrA = jnrB = 0;
 462     j_coord_offsetA = 0;
 463     j_coord_offsetB = 0;
 464
 465     outeriter        = 0;
 466     inneriter        = 0;
 467
 468     /* Start outer loop over neighborlists */
 469     for(iidx=0; iidx<nri; iidx++)
 470     {
 471         /* Load shift vector for this list */
 472         i_shift_offset   = DIM*shiftidx[iidx];
 473
 474         /* Load limits for loop over neighbors */
 475         j_index_start    = jindex[iidx];
 476         j_index_end      = jindex[iidx+1];
 477
 478         /* Get outer coordinate index */
 479         inr              = iinr[iidx];
 480         i_coord_offset   = DIM*inr;
 481
 482         /* Load i particle coords and add shift vector */
 483         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 484
 485         fix0             = _fjsp_setzero_v2r8();
 486         fiy0             = _fjsp_setzero_v2r8();
 487         fiz0             = _fjsp_setzero_v2r8();
 488
 489         /* Load parameters for i particles */
 490         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 491         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 492
 493         /* Start inner kernel loop */
 494         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 495         {
 496
 497             /* Get j neighbor index, and coordinate index */
 498             jnrA             = jjnr[jidx];
 499             jnrB             = jjnr[jidx+1];
 500             j_coord_offsetA  = DIM*jnrA;
 501             j_coord_offsetB  = DIM*jnrB;
 502
 503             /* load j atom coordinates */
 504             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 505                                               &jx0,&jy0,&jz0);
 506
 507             /* Calculate displacement vector */
 508             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 509             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 510             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 511
 512             /* Calculate squared distance and things based on it */
 513             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 514
 515             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 516
 517             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 518
 519             /* Load parameters for j particles */
 520             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 521             vdwjidx0A        = 2*vdwtype[jnrA+0];
 522             vdwjidx0B        = 2*vdwtype[jnrB+0];
 523
 524             /**************************
 525              * CALCULATE INTERACTIONS *
 526              **************************/
 527
 528             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 529
 530             /* Compute parameters for interactions between i and j atoms */
 531             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 532             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 533                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 534
 535             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 536                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 537
 538             /* EWALD ELECTROSTATICS */
 539
 540             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 541             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 542             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 543             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 544             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 545
 546             gmx_fjsp_load_2pair_swizzle_v2r8(ewtab+ewconv.i[0],ewtab+ewconv.i[1],
 547                                          &ewtabF,&ewtabFn);
 548             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 549             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 550
 551             /* Analytical LJ-PME */
 552             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 553             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 554             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 555             exponent         = gmx_simd_exp_d(ewcljrsq);
 556             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 557             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 558             /* f6A = 6 * C6grid * (1 - poly) */
 559             f6A              = _fjsp_mul_v2r8(c6grid_00,_fjsp_sub_v2r8(one,poly));
 560             /* f6B = C6grid * exponent * beta^6 */
 561             f6B              = _fjsp_mul_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6));
 562             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 563             fvdw              = _fjsp_mul_v2r8(_fjsp_madd_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,_fjsp_sub_v2r8(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 564
 565             fscal            = _fjsp_add_v2r8(felec,fvdw);
 566
 567             /* Update vectorial force */
 568             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 569             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 570             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 571
 572             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 573
 574             /* Inner loop uses 61 flops */
 575         }
 576
 577         if(jidx<j_index_end)
 578         {
 579
 580             jnrA             = jjnr[jidx];
 581             j_coord_offsetA  = DIM*jnrA;
 582
 583             /* load j atom coordinates */
 584             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 585                                               &jx0,&jy0,&jz0);
 586
 587             /* Calculate displacement vector */
 588             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 589             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 590             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 591
 592             /* Calculate squared distance and things based on it */
 593             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 594
 595             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 596
 597             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 598
 599             /* Load parameters for j particles */
 600             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 601             vdwjidx0A        = 2*vdwtype[jnrA+0];
 602
 603             /**************************
 604              * CALCULATE INTERACTIONS *
 605              **************************/
 606
 607             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 608
 609             /* Compute parameters for interactions between i and j atoms */
 610             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 611             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 612                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 613
 614             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 615                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 616
 617             /* EWALD ELECTROSTATICS */
 618
 619             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 620             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 621             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 622             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 623             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 624
 625             gmx_fjsp_load_1pair_swizzle_v2r8(ewtab+ewconv.i[0],&ewtabF,&ewtabFn);
 626             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 627             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 628
 629             /* Analytical LJ-PME */
 630             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 631             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 632             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 633             exponent         = gmx_simd_exp_d(ewcljrsq);
 634             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 635             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 636             /* f6A = 6 * C6grid * (1 - poly) */
 637             f6A              = _fjsp_mul_v2r8(c6grid_00,_fjsp_sub_v2r8(one,poly));
 638             /* f6B = C6grid * exponent * beta^6 */
 639             f6B              = _fjsp_mul_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6));
 640             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 641             fvdw              = _fjsp_mul_v2r8(_fjsp_madd_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,_fjsp_sub_v2r8(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 642
 643             fscal            = _fjsp_add_v2r8(felec,fvdw);
 644
 645             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 646
 647             /* Update vectorial force */
 648             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 649             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 650             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 651
 652             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 653
 654             /* Inner loop uses 61 flops */
 655         }
 656
 657         /* End of innermost loop */
 658
 659         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 660                                               f+i_coord_offset,fshift+i_shift_offset);
 661
 662         /* Increment number of inner iterations */
 663         inneriter                  += j_index_end - j_index_start;
 664
 665         /* Outer loop uses 7 flops */
 666     }
 667
 668     /* Increment number of outer iterations */
 669     outeriter        += nri;
 670
 671     /* Update outer/inner flops */
 672
 673     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*61);
 674 }