src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_sse4_1_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 sse4_1_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 "gromacs/simd/math_x86_sse4_1_double.h"
  48 #include "kernelutil_x86_sse4_1_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_VF_sse4_1_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_VF_sse4_1_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwjidx0A,vdwjidx0B;
  83     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  84     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  85     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  86     real             *charge;
  87     int              nvdwtype;
  88     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  89     int              *vdwtype;
  90     real             *vdwparam;
  91     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  92     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  93     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  94     real             rswitch_scalar,d_scalar;
  95     __m128d          dummy_mask,cutoff_mask;
  96     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  97     __m128d          one     = _mm_set1_pd(1.0);
  98     __m128d          two     = _mm_set1_pd(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     facel            = _mm_set1_pd(fr->epsfac);
 111     charge           = mdatoms->chargeA;
 112     krf              = _mm_set1_pd(fr->ic->k_rf);
 113     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 114     crf              = _mm_set1_pd(fr->ic->c_rf);
 115     nvdwtype         = fr->ntype;
 116     vdwparam         = fr->nbfp;
 117     vdwtype          = mdatoms->typeA;
 118
 119     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 120     rcutoff_scalar   = fr->rcoulomb;
 121     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 122     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 123
 124     rswitch_scalar   = fr->rvdw_switch;
 125     rswitch          = _mm_set1_pd(rswitch_scalar);
 126     /* Setup switch parameters */
 127     d_scalar         = rcutoff_scalar-rswitch_scalar;
 128     d                = _mm_set1_pd(d_scalar);
 129     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 130     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 131     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 132     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 133     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 134     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 135
 136     /* Avoid stupid compiler warnings */
 137     jnrA = jnrB = 0;
 138     j_coord_offsetA = 0;
 139     j_coord_offsetB = 0;
 140
 141     outeriter        = 0;
 142     inneriter        = 0;
 143
 144     /* Start outer loop over neighborlists */
 145     for(iidx=0; iidx<nri; iidx++)
 146     {
 147         /* Load shift vector for this list */
 148         i_shift_offset   = DIM*shiftidx[iidx];
 149
 150         /* Load limits for loop over neighbors */
 151         j_index_start    = jindex[iidx];
 152         j_index_end      = jindex[iidx+1];
 153
 154         /* Get outer coordinate index */
 155         inr              = iinr[iidx];
 156         i_coord_offset   = DIM*inr;
 157
 158         /* Load i particle coords and add shift vector */
 159         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 160
 161         fix0             = _mm_setzero_pd();
 162         fiy0             = _mm_setzero_pd();
 163         fiz0             = _mm_setzero_pd();
 164
 165         /* Load parameters for i particles */
 166         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 167         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 168
 169         /* Reset potential sums */
 170         velecsum         = _mm_setzero_pd();
 171         vvdwsum          = _mm_setzero_pd();
 172
 173         /* Start inner kernel loop */
 174         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 175         {
 176
 177             /* Get j neighbor index, and coordinate index */
 178             jnrA             = jjnr[jidx];
 179             jnrB             = jjnr[jidx+1];
 180             j_coord_offsetA  = DIM*jnrA;
 181             j_coord_offsetB  = DIM*jnrB;
 182
 183             /* load j atom coordinates */
 184             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 185                                               &jx0,&jy0,&jz0);
 186
 187             /* Calculate displacement vector */
 188             dx00             = _mm_sub_pd(ix0,jx0);
 189             dy00             = _mm_sub_pd(iy0,jy0);
 190             dz00             = _mm_sub_pd(iz0,jz0);
 191
 192             /* Calculate squared distance and things based on it */
 193             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 194
 195             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 196
 197             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 198
 199             /* Load parameters for j particles */
 200             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 201             vdwjidx0A        = 2*vdwtype[jnrA+0];
 202             vdwjidx0B        = 2*vdwtype[jnrB+0];
 203
 204             /**************************
 205              * CALCULATE INTERACTIONS *
 206              **************************/
 207
 208             if (gmx_mm_any_lt(rsq00,rcutoff2))
 209             {
 210
 211             r00              = _mm_mul_pd(rsq00,rinv00);
 212
 213             /* Compute parameters for interactions between i and j atoms */
 214             qq00             = _mm_mul_pd(iq0,jq0);
 215             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 216                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 217
 218             /* REACTION-FIELD ELECTROSTATICS */
 219             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 220             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 221
 222             /* LENNARD-JONES DISPERSION/REPULSION */
 223
 224             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 225             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 226             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 227             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 228             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 229
 230             d                = _mm_sub_pd(r00,rswitch);
 231             d                = _mm_max_pd(d,_mm_setzero_pd());
 232             d2               = _mm_mul_pd(d,d);
 233             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 234
 235             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 236
 237             /* Evaluate switch function */
 238             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 239             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 240             vvdw             = _mm_mul_pd(vvdw,sw);
 241             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 242
 243             /* Update potential sum for this i atom from the interaction with this j atom. */
 244             velec            = _mm_and_pd(velec,cutoff_mask);
 245             velecsum         = _mm_add_pd(velecsum,velec);
 246             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 247             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 248
 249             fscal            = _mm_add_pd(felec,fvdw);
 250
 251             fscal            = _mm_and_pd(fscal,cutoff_mask);
 252
 253             /* Calculate temporary vectorial force */
 254             tx               = _mm_mul_pd(fscal,dx00);
 255             ty               = _mm_mul_pd(fscal,dy00);
 256             tz               = _mm_mul_pd(fscal,dz00);
 257
 258             /* Update vectorial force */
 259             fix0             = _mm_add_pd(fix0,tx);
 260             fiy0             = _mm_add_pd(fiy0,ty);
 261             fiz0             = _mm_add_pd(fiz0,tz);
 262
 263             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 264
 265             }
 266
 267             /* Inner loop uses 70 flops */
 268         }
 269
 270         if(jidx<j_index_end)
 271         {
 272
 273             jnrA             = jjnr[jidx];
 274             j_coord_offsetA  = DIM*jnrA;
 275
 276             /* load j atom coordinates */
 277             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 278                                               &jx0,&jy0,&jz0);
 279
 280             /* Calculate displacement vector */
 281             dx00             = _mm_sub_pd(ix0,jx0);
 282             dy00             = _mm_sub_pd(iy0,jy0);
 283             dz00             = _mm_sub_pd(iz0,jz0);
 284
 285             /* Calculate squared distance and things based on it */
 286             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 287
 288             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 289
 290             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 291
 292             /* Load parameters for j particles */
 293             jq0              = _mm_load_sd(charge+jnrA+0);
 294             vdwjidx0A        = 2*vdwtype[jnrA+0];
 295
 296             /**************************
 297              * CALCULATE INTERACTIONS *
 298              **************************/
 299
 300             if (gmx_mm_any_lt(rsq00,rcutoff2))
 301             {
 302
 303             r00              = _mm_mul_pd(rsq00,rinv00);
 304
 305             /* Compute parameters for interactions between i and j atoms */
 306             qq00             = _mm_mul_pd(iq0,jq0);
 307             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 308
 309             /* REACTION-FIELD ELECTROSTATICS */
 310             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 311             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 312
 313             /* LENNARD-JONES DISPERSION/REPULSION */
 314
 315             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 316             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 317             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 318             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 319             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 320
 321             d                = _mm_sub_pd(r00,rswitch);
 322             d                = _mm_max_pd(d,_mm_setzero_pd());
 323             d2               = _mm_mul_pd(d,d);
 324             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 325
 326             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 327
 328             /* Evaluate switch function */
 329             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 330             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 331             vvdw             = _mm_mul_pd(vvdw,sw);
 332             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 333
 334             /* Update potential sum for this i atom from the interaction with this j atom. */
 335             velec            = _mm_and_pd(velec,cutoff_mask);
 336             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 337             velecsum         = _mm_add_pd(velecsum,velec);
 338             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 339             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 340             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 341
 342             fscal            = _mm_add_pd(felec,fvdw);
 343
 344             fscal            = _mm_and_pd(fscal,cutoff_mask);
 345
 346             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 347
 348             /* Calculate temporary vectorial force */
 349             tx               = _mm_mul_pd(fscal,dx00);
 350             ty               = _mm_mul_pd(fscal,dy00);
 351             tz               = _mm_mul_pd(fscal,dz00);
 352
 353             /* Update vectorial force */
 354             fix0             = _mm_add_pd(fix0,tx);
 355             fiy0             = _mm_add_pd(fiy0,ty);
 356             fiz0             = _mm_add_pd(fiz0,tz);
 357
 358             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 359
 360             }
 361
 362             /* Inner loop uses 70 flops */
 363         }
 364
 365         /* End of innermost loop */
 366
 367         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 368                                               f+i_coord_offset,fshift+i_shift_offset);
 369
 370         ggid                        = gid[iidx];
 371         /* Update potential energies */
 372         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 373         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 374
 375         /* Increment number of inner iterations */
 376         inneriter                  += j_index_end - j_index_start;
 377
 378         /* Outer loop uses 9 flops */
 379     }
 380
 381     /* Increment number of outer iterations */
 382     outeriter        += nri;
 383
 384     /* Update outer/inner flops */
 385
 386     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*70);
 387 }
 388 /*
 389  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_F_sse4_1_double
 390  * Electrostatics interaction: ReactionField
 391  * VdW interaction:            LennardJones
 392  * Geometry:                   Particle-Particle
 393  * Calculate force/pot:        Force
 394  */
 395 void
 396 nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_F_sse4_1_double
 397                     (t_nblist                    * gmx_restrict       nlist,
 398                      rvec                        * gmx_restrict          xx,
 399                      rvec                        * gmx_restrict          ff,
 400                      t_forcerec                  * gmx_restrict          fr,
 401                      t_mdatoms                   * gmx_restrict     mdatoms,
 402                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 403                      t_nrnb                      * gmx_restrict        nrnb)
 404 {
 405     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 406      * just 0 for non-waters.
 407      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 408      * jnr indices corresponding to data put in the four positions in the SIMD register.
 409      */
 410     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 411     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 412     int              jnrA,jnrB;
 413     int              j_coord_offsetA,j_coord_offsetB;
 414     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 415     real             rcutoff_scalar;
 416     real             *shiftvec,*fshift,*x,*f;
 417     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 418     int              vdwioffset0;
 419     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 420     int              vdwjidx0A,vdwjidx0B;
 421     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 422     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 423     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 424     real             *charge;
 425     int              nvdwtype;
 426     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 427     int              *vdwtype;
 428     real             *vdwparam;
 429     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 430     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 431     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 432     real             rswitch_scalar,d_scalar;
 433     __m128d          dummy_mask,cutoff_mask;
 434     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 435     __m128d          one     = _mm_set1_pd(1.0);
 436     __m128d          two     = _mm_set1_pd(2.0);
 437     x                = xx[0];
 438     f                = ff[0];
 439
 440     nri              = nlist->nri;
 441     iinr             = nlist->iinr;
 442     jindex           = nlist->jindex;
 443     jjnr             = nlist->jjnr;
 444     shiftidx         = nlist->shift;
 445     gid              = nlist->gid;
 446     shiftvec         = fr->shift_vec[0];
 447     fshift           = fr->fshift[0];
 448     facel            = _mm_set1_pd(fr->epsfac);
 449     charge           = mdatoms->chargeA;
 450     krf              = _mm_set1_pd(fr->ic->k_rf);
 451     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 452     crf              = _mm_set1_pd(fr->ic->c_rf);
 453     nvdwtype         = fr->ntype;
 454     vdwparam         = fr->nbfp;
 455     vdwtype          = mdatoms->typeA;
 456
 457     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 458     rcutoff_scalar   = fr->rcoulomb;
 459     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 460     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 461
 462     rswitch_scalar   = fr->rvdw_switch;
 463     rswitch          = _mm_set1_pd(rswitch_scalar);
 464     /* Setup switch parameters */
 465     d_scalar         = rcutoff_scalar-rswitch_scalar;
 466     d                = _mm_set1_pd(d_scalar);
 467     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 468     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 469     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 470     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 471     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 472     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 473
 474     /* Avoid stupid compiler warnings */
 475     jnrA = jnrB = 0;
 476     j_coord_offsetA = 0;
 477     j_coord_offsetB = 0;
 478
 479     outeriter        = 0;
 480     inneriter        = 0;
 481
 482     /* Start outer loop over neighborlists */
 483     for(iidx=0; iidx<nri; iidx++)
 484     {
 485         /* Load shift vector for this list */
 486         i_shift_offset   = DIM*shiftidx[iidx];
 487
 488         /* Load limits for loop over neighbors */
 489         j_index_start    = jindex[iidx];
 490         j_index_end      = jindex[iidx+1];
 491
 492         /* Get outer coordinate index */
 493         inr              = iinr[iidx];
 494         i_coord_offset   = DIM*inr;
 495
 496         /* Load i particle coords and add shift vector */
 497         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 498
 499         fix0             = _mm_setzero_pd();
 500         fiy0             = _mm_setzero_pd();
 501         fiz0             = _mm_setzero_pd();
 502
 503         /* Load parameters for i particles */
 504         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 505         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 506
 507         /* Start inner kernel loop */
 508         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 509         {
 510
 511             /* Get j neighbor index, and coordinate index */
 512             jnrA             = jjnr[jidx];
 513             jnrB             = jjnr[jidx+1];
 514             j_coord_offsetA  = DIM*jnrA;
 515             j_coord_offsetB  = DIM*jnrB;
 516
 517             /* load j atom coordinates */
 518             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 519                                               &jx0,&jy0,&jz0);
 520
 521             /* Calculate displacement vector */
 522             dx00             = _mm_sub_pd(ix0,jx0);
 523             dy00             = _mm_sub_pd(iy0,jy0);
 524             dz00             = _mm_sub_pd(iz0,jz0);
 525
 526             /* Calculate squared distance and things based on it */
 527             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 528
 529             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 530
 531             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 532
 533             /* Load parameters for j particles */
 534             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 535             vdwjidx0A        = 2*vdwtype[jnrA+0];
 536             vdwjidx0B        = 2*vdwtype[jnrB+0];
 537
 538             /**************************
 539              * CALCULATE INTERACTIONS *
 540              **************************/
 541
 542             if (gmx_mm_any_lt(rsq00,rcutoff2))
 543             {
 544
 545             r00              = _mm_mul_pd(rsq00,rinv00);
 546
 547             /* Compute parameters for interactions between i and j atoms */
 548             qq00             = _mm_mul_pd(iq0,jq0);
 549             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 550                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 551
 552             /* REACTION-FIELD ELECTROSTATICS */
 553             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 554
 555             /* LENNARD-JONES DISPERSION/REPULSION */
 556
 557             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 558             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 559             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 560             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 561             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 562
 563             d                = _mm_sub_pd(r00,rswitch);
 564             d                = _mm_max_pd(d,_mm_setzero_pd());
 565             d2               = _mm_mul_pd(d,d);
 566             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 567
 568             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 569
 570             /* Evaluate switch function */
 571             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 572             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 573             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 574
 575             fscal            = _mm_add_pd(felec,fvdw);
 576
 577             fscal            = _mm_and_pd(fscal,cutoff_mask);
 578
 579             /* Calculate temporary vectorial force */
 580             tx               = _mm_mul_pd(fscal,dx00);
 581             ty               = _mm_mul_pd(fscal,dy00);
 582             tz               = _mm_mul_pd(fscal,dz00);
 583
 584             /* Update vectorial force */
 585             fix0             = _mm_add_pd(fix0,tx);
 586             fiy0             = _mm_add_pd(fiy0,ty);
 587             fiz0             = _mm_add_pd(fiz0,tz);
 588
 589             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 590
 591             }
 592
 593             /* Inner loop uses 61 flops */
 594         }
 595
 596         if(jidx<j_index_end)
 597         {
 598
 599             jnrA             = jjnr[jidx];
 600             j_coord_offsetA  = DIM*jnrA;
 601
 602             /* load j atom coordinates */
 603             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 604                                               &jx0,&jy0,&jz0);
 605
 606             /* Calculate displacement vector */
 607             dx00             = _mm_sub_pd(ix0,jx0);
 608             dy00             = _mm_sub_pd(iy0,jy0);
 609             dz00             = _mm_sub_pd(iz0,jz0);
 610
 611             /* Calculate squared distance and things based on it */
 612             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 613
 614             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 615
 616             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 617
 618             /* Load parameters for j particles */
 619             jq0              = _mm_load_sd(charge+jnrA+0);
 620             vdwjidx0A        = 2*vdwtype[jnrA+0];
 621
 622             /**************************
 623              * CALCULATE INTERACTIONS *
 624              **************************/
 625
 626             if (gmx_mm_any_lt(rsq00,rcutoff2))
 627             {
 628
 629             r00              = _mm_mul_pd(rsq00,rinv00);
 630
 631             /* Compute parameters for interactions between i and j atoms */
 632             qq00             = _mm_mul_pd(iq0,jq0);
 633             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 634
 635             /* REACTION-FIELD ELECTROSTATICS */
 636             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 637
 638             /* LENNARD-JONES DISPERSION/REPULSION */
 639
 640             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 641             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 642             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 643             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 644             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 645
 646             d                = _mm_sub_pd(r00,rswitch);
 647             d                = _mm_max_pd(d,_mm_setzero_pd());
 648             d2               = _mm_mul_pd(d,d);
 649             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 650
 651             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 652
 653             /* Evaluate switch function */
 654             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 655             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 656             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 657
 658             fscal            = _mm_add_pd(felec,fvdw);
 659
 660             fscal            = _mm_and_pd(fscal,cutoff_mask);
 661
 662             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 663
 664             /* Calculate temporary vectorial force */
 665             tx               = _mm_mul_pd(fscal,dx00);
 666             ty               = _mm_mul_pd(fscal,dy00);
 667             tz               = _mm_mul_pd(fscal,dz00);
 668
 669             /* Update vectorial force */
 670             fix0             = _mm_add_pd(fix0,tx);
 671             fiy0             = _mm_add_pd(fiy0,ty);
 672             fiz0             = _mm_add_pd(fiz0,tz);
 673
 674             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 675
 676             }
 677
 678             /* Inner loop uses 61 flops */
 679         }
 680
 681         /* End of innermost loop */
 682
 683         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 684                                               f+i_coord_offset,fshift+i_shift_offset);
 685
 686         /* Increment number of inner iterations */
 687         inneriter                  += j_index_end - j_index_start;
 688
 689         /* Outer loop uses 7 flops */
 690     }
 691
 692     /* Increment number of outer iterations */
 693     outeriter        += nri;
 694
 695     /* Update outer/inner flops */
 696
 697     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*61);
 698 }