src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_avx_128_fma_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 avx_128_fma_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  48 #include "kernelutil_x86_avx_128_fma_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_VF_avx_128_fma_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_avx_128_fma_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_macc_pd(krf,rsq00,rinv00),crf));
 220             felec            = _mm_mul_pd(qq00,_mm_msub_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_msub_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_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 234
 235             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 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_msub_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             /* Update vectorial force */
 254             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 255             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 256             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 257
 258             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 259                                                    _mm_mul_pd(dx00,fscal),
 260                                                    _mm_mul_pd(dy00,fscal),
 261                                                    _mm_mul_pd(dz00,fscal));
 262
 263             }
 264
 265             /* Inner loop uses 73 flops */
 266         }
 267
 268         if(jidx<j_index_end)
 269         {
 270
 271             jnrA             = jjnr[jidx];
 272             j_coord_offsetA  = DIM*jnrA;
 273
 274             /* load j atom coordinates */
 275             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 276                                               &jx0,&jy0,&jz0);
 277
 278             /* Calculate displacement vector */
 279             dx00             = _mm_sub_pd(ix0,jx0);
 280             dy00             = _mm_sub_pd(iy0,jy0);
 281             dz00             = _mm_sub_pd(iz0,jz0);
 282
 283             /* Calculate squared distance and things based on it */
 284             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 285
 286             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 287
 288             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 289
 290             /* Load parameters for j particles */
 291             jq0              = _mm_load_sd(charge+jnrA+0);
 292             vdwjidx0A        = 2*vdwtype[jnrA+0];
 293
 294             /**************************
 295              * CALCULATE INTERACTIONS *
 296              **************************/
 297
 298             if (gmx_mm_any_lt(rsq00,rcutoff2))
 299             {
 300
 301             r00              = _mm_mul_pd(rsq00,rinv00);
 302
 303             /* Compute parameters for interactions between i and j atoms */
 304             qq00             = _mm_mul_pd(iq0,jq0);
 305             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 306
 307             /* REACTION-FIELD ELECTROSTATICS */
 308             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 309             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 310
 311             /* LENNARD-JONES DISPERSION/REPULSION */
 312
 313             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 314             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 315             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 316             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 317             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 318
 319             d                = _mm_sub_pd(r00,rswitch);
 320             d                = _mm_max_pd(d,_mm_setzero_pd());
 321             d2               = _mm_mul_pd(d,d);
 322             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 323
 324             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 325
 326             /* Evaluate switch function */
 327             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 328             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 329             vvdw             = _mm_mul_pd(vvdw,sw);
 330             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 331
 332             /* Update potential sum for this i atom from the interaction with this j atom. */
 333             velec            = _mm_and_pd(velec,cutoff_mask);
 334             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 335             velecsum         = _mm_add_pd(velecsum,velec);
 336             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 337             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 338             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 339
 340             fscal            = _mm_add_pd(felec,fvdw);
 341
 342             fscal            = _mm_and_pd(fscal,cutoff_mask);
 343
 344             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 345
 346             /* Update vectorial force */
 347             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 348             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 349             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 350
 351             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 352                                                    _mm_mul_pd(dx00,fscal),
 353                                                    _mm_mul_pd(dy00,fscal),
 354                                                    _mm_mul_pd(dz00,fscal));
 355
 356             }
 357
 358             /* Inner loop uses 73 flops */
 359         }
 360
 361         /* End of innermost loop */
 362
 363         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 364                                               f+i_coord_offset,fshift+i_shift_offset);
 365
 366         ggid                        = gid[iidx];
 367         /* Update potential energies */
 368         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 369         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 370
 371         /* Increment number of inner iterations */
 372         inneriter                  += j_index_end - j_index_start;
 373
 374         /* Outer loop uses 9 flops */
 375     }
 376
 377     /* Increment number of outer iterations */
 378     outeriter        += nri;
 379
 380     /* Update outer/inner flops */
 381
 382     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*73);
 383 }
 384 /*
 385  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_F_avx_128_fma_double
 386  * Electrostatics interaction: ReactionField
 387  * VdW interaction:            LennardJones
 388  * Geometry:                   Particle-Particle
 389  * Calculate force/pot:        Force
 390  */
 391 void
 392 nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_F_avx_128_fma_double
 393                     (t_nblist                    * gmx_restrict       nlist,
 394                      rvec                        * gmx_restrict          xx,
 395                      rvec                        * gmx_restrict          ff,
 396                      t_forcerec                  * gmx_restrict          fr,
 397                      t_mdatoms                   * gmx_restrict     mdatoms,
 398                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 399                      t_nrnb                      * gmx_restrict        nrnb)
 400 {
 401     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 402      * just 0 for non-waters.
 403      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 404      * jnr indices corresponding to data put in the four positions in the SIMD register.
 405      */
 406     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 407     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 408     int              jnrA,jnrB;
 409     int              j_coord_offsetA,j_coord_offsetB;
 410     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 411     real             rcutoff_scalar;
 412     real             *shiftvec,*fshift,*x,*f;
 413     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 414     int              vdwioffset0;
 415     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 416     int              vdwjidx0A,vdwjidx0B;
 417     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 418     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 419     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 420     real             *charge;
 421     int              nvdwtype;
 422     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 423     int              *vdwtype;
 424     real             *vdwparam;
 425     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 426     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 427     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 428     real             rswitch_scalar,d_scalar;
 429     __m128d          dummy_mask,cutoff_mask;
 430     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 431     __m128d          one     = _mm_set1_pd(1.0);
 432     __m128d          two     = _mm_set1_pd(2.0);
 433     x                = xx[0];
 434     f                = ff[0];
 435
 436     nri              = nlist->nri;
 437     iinr             = nlist->iinr;
 438     jindex           = nlist->jindex;
 439     jjnr             = nlist->jjnr;
 440     shiftidx         = nlist->shift;
 441     gid              = nlist->gid;
 442     shiftvec         = fr->shift_vec[0];
 443     fshift           = fr->fshift[0];
 444     facel            = _mm_set1_pd(fr->epsfac);
 445     charge           = mdatoms->chargeA;
 446     krf              = _mm_set1_pd(fr->ic->k_rf);
 447     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 448     crf              = _mm_set1_pd(fr->ic->c_rf);
 449     nvdwtype         = fr->ntype;
 450     vdwparam         = fr->nbfp;
 451     vdwtype          = mdatoms->typeA;
 452
 453     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 454     rcutoff_scalar   = fr->rcoulomb;
 455     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 456     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 457
 458     rswitch_scalar   = fr->rvdw_switch;
 459     rswitch          = _mm_set1_pd(rswitch_scalar);
 460     /* Setup switch parameters */
 461     d_scalar         = rcutoff_scalar-rswitch_scalar;
 462     d                = _mm_set1_pd(d_scalar);
 463     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 464     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 465     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 466     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 467     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 468     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 469
 470     /* Avoid stupid compiler warnings */
 471     jnrA = jnrB = 0;
 472     j_coord_offsetA = 0;
 473     j_coord_offsetB = 0;
 474
 475     outeriter        = 0;
 476     inneriter        = 0;
 477
 478     /* Start outer loop over neighborlists */
 479     for(iidx=0; iidx<nri; iidx++)
 480     {
 481         /* Load shift vector for this list */
 482         i_shift_offset   = DIM*shiftidx[iidx];
 483
 484         /* Load limits for loop over neighbors */
 485         j_index_start    = jindex[iidx];
 486         j_index_end      = jindex[iidx+1];
 487
 488         /* Get outer coordinate index */
 489         inr              = iinr[iidx];
 490         i_coord_offset   = DIM*inr;
 491
 492         /* Load i particle coords and add shift vector */
 493         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 494
 495         fix0             = _mm_setzero_pd();
 496         fiy0             = _mm_setzero_pd();
 497         fiz0             = _mm_setzero_pd();
 498
 499         /* Load parameters for i particles */
 500         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 501         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 502
 503         /* Start inner kernel loop */
 504         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 505         {
 506
 507             /* Get j neighbor index, and coordinate index */
 508             jnrA             = jjnr[jidx];
 509             jnrB             = jjnr[jidx+1];
 510             j_coord_offsetA  = DIM*jnrA;
 511             j_coord_offsetB  = DIM*jnrB;
 512
 513             /* load j atom coordinates */
 514             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 515                                               &jx0,&jy0,&jz0);
 516
 517             /* Calculate displacement vector */
 518             dx00             = _mm_sub_pd(ix0,jx0);
 519             dy00             = _mm_sub_pd(iy0,jy0);
 520             dz00             = _mm_sub_pd(iz0,jz0);
 521
 522             /* Calculate squared distance and things based on it */
 523             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 524
 525             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 526
 527             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 528
 529             /* Load parameters for j particles */
 530             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 531             vdwjidx0A        = 2*vdwtype[jnrA+0];
 532             vdwjidx0B        = 2*vdwtype[jnrB+0];
 533
 534             /**************************
 535              * CALCULATE INTERACTIONS *
 536              **************************/
 537
 538             if (gmx_mm_any_lt(rsq00,rcutoff2))
 539             {
 540
 541             r00              = _mm_mul_pd(rsq00,rinv00);
 542
 543             /* Compute parameters for interactions between i and j atoms */
 544             qq00             = _mm_mul_pd(iq0,jq0);
 545             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 546                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 547
 548             /* REACTION-FIELD ELECTROSTATICS */
 549             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 550
 551             /* LENNARD-JONES DISPERSION/REPULSION */
 552
 553             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 554             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 555             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 556             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 557             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 558
 559             d                = _mm_sub_pd(r00,rswitch);
 560             d                = _mm_max_pd(d,_mm_setzero_pd());
 561             d2               = _mm_mul_pd(d,d);
 562             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 563
 564             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 565
 566             /* Evaluate switch function */
 567             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 568             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 569             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 570
 571             fscal            = _mm_add_pd(felec,fvdw);
 572
 573             fscal            = _mm_and_pd(fscal,cutoff_mask);
 574
 575             /* Update vectorial force */
 576             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 577             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 578             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 579
 580             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 581                                                    _mm_mul_pd(dx00,fscal),
 582                                                    _mm_mul_pd(dy00,fscal),
 583                                                    _mm_mul_pd(dz00,fscal));
 584
 585             }
 586
 587             /* Inner loop uses 64 flops */
 588         }
 589
 590         if(jidx<j_index_end)
 591         {
 592
 593             jnrA             = jjnr[jidx];
 594             j_coord_offsetA  = DIM*jnrA;
 595
 596             /* load j atom coordinates */
 597             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 598                                               &jx0,&jy0,&jz0);
 599
 600             /* Calculate displacement vector */
 601             dx00             = _mm_sub_pd(ix0,jx0);
 602             dy00             = _mm_sub_pd(iy0,jy0);
 603             dz00             = _mm_sub_pd(iz0,jz0);
 604
 605             /* Calculate squared distance and things based on it */
 606             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 607
 608             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 609
 610             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 611
 612             /* Load parameters for j particles */
 613             jq0              = _mm_load_sd(charge+jnrA+0);
 614             vdwjidx0A        = 2*vdwtype[jnrA+0];
 615
 616             /**************************
 617              * CALCULATE INTERACTIONS *
 618              **************************/
 619
 620             if (gmx_mm_any_lt(rsq00,rcutoff2))
 621             {
 622
 623             r00              = _mm_mul_pd(rsq00,rinv00);
 624
 625             /* Compute parameters for interactions between i and j atoms */
 626             qq00             = _mm_mul_pd(iq0,jq0);
 627             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 628
 629             /* REACTION-FIELD ELECTROSTATICS */
 630             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 631
 632             /* LENNARD-JONES DISPERSION/REPULSION */
 633
 634             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 635             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 636             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 637             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 638             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 639
 640             d                = _mm_sub_pd(r00,rswitch);
 641             d                = _mm_max_pd(d,_mm_setzero_pd());
 642             d2               = _mm_mul_pd(d,d);
 643             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 644
 645             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 646
 647             /* Evaluate switch function */
 648             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 649             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 650             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 651
 652             fscal            = _mm_add_pd(felec,fvdw);
 653
 654             fscal            = _mm_and_pd(fscal,cutoff_mask);
 655
 656             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 657
 658             /* Update vectorial force */
 659             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 660             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 661             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 662
 663             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 664                                                    _mm_mul_pd(dx00,fscal),
 665                                                    _mm_mul_pd(dy00,fscal),
 666                                                    _mm_mul_pd(dz00,fscal));
 667
 668             }
 669
 670             /* Inner loop uses 64 flops */
 671         }
 672
 673         /* End of innermost loop */
 674
 675         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 676                                               f+i_coord_offset,fshift+i_shift_offset);
 677
 678         /* Increment number of inner iterations */
 679         inneriter                  += j_index_end - j_index_start;
 680
 681         /* Outer loop uses 7 flops */
 682     }
 683
 684     /* Increment number of outer iterations */
 685     outeriter        += nri;
 686
 687     /* Update outer/inner flops */
 688
 689     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*64);
 690 }