src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRF_VdwLJ_GeomP1P1_sse2_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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 sse2_single 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_x86_sse2_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomP1P1_VF_sse2_single
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRF_VdwLJ_GeomP1P1_VF_sse2_single
  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,C,D refer to j loop unrolling done with SSE, e.g. for the four 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,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  85     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  88     real             *charge;
  89     int              nvdwtype;
  90     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  91     int              *vdwtype;
  92     real             *vdwparam;
  93     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  94     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  95     __m128           dummy_mask,cutoff_mask;
  96     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  97     __m128           one     = _mm_set1_ps(1.0);
  98     __m128           two     = _mm_set1_ps(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_ps(fr->ic->epsfac);
 111     charge           = mdatoms->chargeA;
 112     krf              = _mm_set1_ps(fr->ic->k_rf);
 113     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 114     crf              = _mm_set1_ps(fr->ic->c_rf);
 115     nvdwtype         = fr->ntype;
 116     vdwparam         = fr->nbfp;
 117     vdwtype          = mdatoms->typeA;
 118
 119     /* Avoid stupid compiler warnings */
 120     jnrA = jnrB = jnrC = jnrD = 0;
 121     j_coord_offsetA = 0;
 122     j_coord_offsetB = 0;
 123     j_coord_offsetC = 0;
 124     j_coord_offsetD = 0;
 125
 126     outeriter        = 0;
 127     inneriter        = 0;
 128
 129     for(iidx=0;iidx<4*DIM;iidx++)
 130     {
 131         scratch[iidx] = 0.0;
 132     }
 133
 134     /* Start outer loop over neighborlists */
 135     for(iidx=0; iidx<nri; iidx++)
 136     {
 137         /* Load shift vector for this list */
 138         i_shift_offset   = DIM*shiftidx[iidx];
 139
 140         /* Load limits for loop over neighbors */
 141         j_index_start    = jindex[iidx];
 142         j_index_end      = jindex[iidx+1];
 143
 144         /* Get outer coordinate index */
 145         inr              = iinr[iidx];
 146         i_coord_offset   = DIM*inr;
 147
 148         /* Load i particle coords and add shift vector */
 149         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 150
 151         fix0             = _mm_setzero_ps();
 152         fiy0             = _mm_setzero_ps();
 153         fiz0             = _mm_setzero_ps();
 154
 155         /* Load parameters for i particles */
 156         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 157         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 158
 159         /* Reset potential sums */
 160         velecsum         = _mm_setzero_ps();
 161         vvdwsum          = _mm_setzero_ps();
 162
 163         /* Start inner kernel loop */
 164         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 165         {
 166
 167             /* Get j neighbor index, and coordinate index */
 168             jnrA             = jjnr[jidx];
 169             jnrB             = jjnr[jidx+1];
 170             jnrC             = jjnr[jidx+2];
 171             jnrD             = jjnr[jidx+3];
 172             j_coord_offsetA  = DIM*jnrA;
 173             j_coord_offsetB  = DIM*jnrB;
 174             j_coord_offsetC  = DIM*jnrC;
 175             j_coord_offsetD  = DIM*jnrD;
 176
 177             /* load j atom coordinates */
 178             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 179                                               x+j_coord_offsetC,x+j_coord_offsetD,
 180                                               &jx0,&jy0,&jz0);
 181
 182             /* Calculate displacement vector */
 183             dx00             = _mm_sub_ps(ix0,jx0);
 184             dy00             = _mm_sub_ps(iy0,jy0);
 185             dz00             = _mm_sub_ps(iz0,jz0);
 186
 187             /* Calculate squared distance and things based on it */
 188             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 189
 190             rinv00           = sse2_invsqrt_f(rsq00);
 191
 192             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 193
 194             /* Load parameters for j particles */
 195             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 196                                                               charge+jnrC+0,charge+jnrD+0);
 197             vdwjidx0A        = 2*vdwtype[jnrA+0];
 198             vdwjidx0B        = 2*vdwtype[jnrB+0];
 199             vdwjidx0C        = 2*vdwtype[jnrC+0];
 200             vdwjidx0D        = 2*vdwtype[jnrD+0];
 201
 202             /**************************
 203              * CALCULATE INTERACTIONS *
 204              **************************/
 205
 206             /* Compute parameters for interactions between i and j atoms */
 207             qq00             = _mm_mul_ps(iq0,jq0);
 208             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 209                                          vdwparam+vdwioffset0+vdwjidx0B,
 210                                          vdwparam+vdwioffset0+vdwjidx0C,
 211                                          vdwparam+vdwioffset0+vdwjidx0D,
 212                                          &c6_00,&c12_00);
 213
 214             /* REACTION-FIELD ELECTROSTATICS */
 215             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 216             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 217
 218             /* LENNARD-JONES DISPERSION/REPULSION */
 219
 220             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 221             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 222             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 223             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 224             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 225
 226             /* Update potential sum for this i atom from the interaction with this j atom. */
 227             velecsum         = _mm_add_ps(velecsum,velec);
 228             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 229
 230             fscal            = _mm_add_ps(felec,fvdw);
 231
 232             /* Calculate temporary vectorial force */
 233             tx               = _mm_mul_ps(fscal,dx00);
 234             ty               = _mm_mul_ps(fscal,dy00);
 235             tz               = _mm_mul_ps(fscal,dz00);
 236
 237             /* Update vectorial force */
 238             fix0             = _mm_add_ps(fix0,tx);
 239             fiy0             = _mm_add_ps(fiy0,ty);
 240             fiz0             = _mm_add_ps(fiz0,tz);
 241
 242             fjptrA             = f+j_coord_offsetA;
 243             fjptrB             = f+j_coord_offsetB;
 244             fjptrC             = f+j_coord_offsetC;
 245             fjptrD             = f+j_coord_offsetD;
 246             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 247
 248             /* Inner loop uses 44 flops */
 249         }
 250
 251         if(jidx<j_index_end)
 252         {
 253
 254             /* Get j neighbor index, and coordinate index */
 255             jnrlistA         = jjnr[jidx];
 256             jnrlistB         = jjnr[jidx+1];
 257             jnrlistC         = jjnr[jidx+2];
 258             jnrlistD         = jjnr[jidx+3];
 259             /* Sign of each element will be negative for non-real atoms.
 260              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 261              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 262              */
 263             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 264             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 265             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 266             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 267             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 268             j_coord_offsetA  = DIM*jnrA;
 269             j_coord_offsetB  = DIM*jnrB;
 270             j_coord_offsetC  = DIM*jnrC;
 271             j_coord_offsetD  = DIM*jnrD;
 272
 273             /* load j atom coordinates */
 274             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 275                                               x+j_coord_offsetC,x+j_coord_offsetD,
 276                                               &jx0,&jy0,&jz0);
 277
 278             /* Calculate displacement vector */
 279             dx00             = _mm_sub_ps(ix0,jx0);
 280             dy00             = _mm_sub_ps(iy0,jy0);
 281             dz00             = _mm_sub_ps(iz0,jz0);
 282
 283             /* Calculate squared distance and things based on it */
 284             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 285
 286             rinv00           = sse2_invsqrt_f(rsq00);
 287
 288             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 289
 290             /* Load parameters for j particles */
 291             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 292                                                               charge+jnrC+0,charge+jnrD+0);
 293             vdwjidx0A        = 2*vdwtype[jnrA+0];
 294             vdwjidx0B        = 2*vdwtype[jnrB+0];
 295             vdwjidx0C        = 2*vdwtype[jnrC+0];
 296             vdwjidx0D        = 2*vdwtype[jnrD+0];
 297
 298             /**************************
 299              * CALCULATE INTERACTIONS *
 300              **************************/
 301
 302             /* Compute parameters for interactions between i and j atoms */
 303             qq00             = _mm_mul_ps(iq0,jq0);
 304             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 305                                          vdwparam+vdwioffset0+vdwjidx0B,
 306                                          vdwparam+vdwioffset0+vdwjidx0C,
 307                                          vdwparam+vdwioffset0+vdwjidx0D,
 308                                          &c6_00,&c12_00);
 309
 310             /* REACTION-FIELD ELECTROSTATICS */
 311             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 312             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 313
 314             /* LENNARD-JONES DISPERSION/REPULSION */
 315
 316             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 317             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 318             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 319             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 320             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 321
 322             /* Update potential sum for this i atom from the interaction with this j atom. */
 323             velec            = _mm_andnot_ps(dummy_mask,velec);
 324             velecsum         = _mm_add_ps(velecsum,velec);
 325             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 326             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 327
 328             fscal            = _mm_add_ps(felec,fvdw);
 329
 330             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 331
 332             /* Calculate temporary vectorial force */
 333             tx               = _mm_mul_ps(fscal,dx00);
 334             ty               = _mm_mul_ps(fscal,dy00);
 335             tz               = _mm_mul_ps(fscal,dz00);
 336
 337             /* Update vectorial force */
 338             fix0             = _mm_add_ps(fix0,tx);
 339             fiy0             = _mm_add_ps(fiy0,ty);
 340             fiz0             = _mm_add_ps(fiz0,tz);
 341
 342             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 343             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 344             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 345             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 346             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 347
 348             /* Inner loop uses 44 flops */
 349         }
 350
 351         /* End of innermost loop */
 352
 353         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 354                                               f+i_coord_offset,fshift+i_shift_offset);
 355
 356         ggid                        = gid[iidx];
 357         /* Update potential energies */
 358         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 359         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 360
 361         /* Increment number of inner iterations */
 362         inneriter                  += j_index_end - j_index_start;
 363
 364         /* Outer loop uses 9 flops */
 365     }
 366
 367     /* Increment number of outer iterations */
 368     outeriter        += nri;
 369
 370     /* Update outer/inner flops */
 371
 372     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*44);
 373 }
 374 /*
 375  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_sse2_single
 376  * Electrostatics interaction: ReactionField
 377  * VdW interaction:            LennardJones
 378  * Geometry:                   Particle-Particle
 379  * Calculate force/pot:        Force
 380  */
 381 void
 382 nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_sse2_single
 383                     (t_nblist                    * gmx_restrict       nlist,
 384                      rvec                        * gmx_restrict          xx,
 385                      rvec                        * gmx_restrict          ff,
 386                      struct t_forcerec           * gmx_restrict          fr,
 387                      t_mdatoms                   * gmx_restrict     mdatoms,
 388                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 389                      t_nrnb                      * gmx_restrict        nrnb)
 390 {
 391     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 392      * just 0 for non-waters.
 393      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 394      * jnr indices corresponding to data put in the four positions in the SIMD register.
 395      */
 396     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 397     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 398     int              jnrA,jnrB,jnrC,jnrD;
 399     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 400     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 401     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 402     real             rcutoff_scalar;
 403     real             *shiftvec,*fshift,*x,*f;
 404     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 405     real             scratch[4*DIM];
 406     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 407     int              vdwioffset0;
 408     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 409     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 410     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 411     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 412     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 413     real             *charge;
 414     int              nvdwtype;
 415     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 416     int              *vdwtype;
 417     real             *vdwparam;
 418     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 419     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 420     __m128           dummy_mask,cutoff_mask;
 421     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 422     __m128           one     = _mm_set1_ps(1.0);
 423     __m128           two     = _mm_set1_ps(2.0);
 424     x                = xx[0];
 425     f                = ff[0];
 426
 427     nri              = nlist->nri;
 428     iinr             = nlist->iinr;
 429     jindex           = nlist->jindex;
 430     jjnr             = nlist->jjnr;
 431     shiftidx         = nlist->shift;
 432     gid              = nlist->gid;
 433     shiftvec         = fr->shift_vec[0];
 434     fshift           = fr->fshift[0];
 435     facel            = _mm_set1_ps(fr->ic->epsfac);
 436     charge           = mdatoms->chargeA;
 437     krf              = _mm_set1_ps(fr->ic->k_rf);
 438     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 439     crf              = _mm_set1_ps(fr->ic->c_rf);
 440     nvdwtype         = fr->ntype;
 441     vdwparam         = fr->nbfp;
 442     vdwtype          = mdatoms->typeA;
 443
 444     /* Avoid stupid compiler warnings */
 445     jnrA = jnrB = jnrC = jnrD = 0;
 446     j_coord_offsetA = 0;
 447     j_coord_offsetB = 0;
 448     j_coord_offsetC = 0;
 449     j_coord_offsetD = 0;
 450
 451     outeriter        = 0;
 452     inneriter        = 0;
 453
 454     for(iidx=0;iidx<4*DIM;iidx++)
 455     {
 456         scratch[iidx] = 0.0;
 457     }
 458
 459     /* Start outer loop over neighborlists */
 460     for(iidx=0; iidx<nri; iidx++)
 461     {
 462         /* Load shift vector for this list */
 463         i_shift_offset   = DIM*shiftidx[iidx];
 464
 465         /* Load limits for loop over neighbors */
 466         j_index_start    = jindex[iidx];
 467         j_index_end      = jindex[iidx+1];
 468
 469         /* Get outer coordinate index */
 470         inr              = iinr[iidx];
 471         i_coord_offset   = DIM*inr;
 472
 473         /* Load i particle coords and add shift vector */
 474         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 475
 476         fix0             = _mm_setzero_ps();
 477         fiy0             = _mm_setzero_ps();
 478         fiz0             = _mm_setzero_ps();
 479
 480         /* Load parameters for i particles */
 481         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 482         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 483
 484         /* Start inner kernel loop */
 485         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 486         {
 487
 488             /* Get j neighbor index, and coordinate index */
 489             jnrA             = jjnr[jidx];
 490             jnrB             = jjnr[jidx+1];
 491             jnrC             = jjnr[jidx+2];
 492             jnrD             = jjnr[jidx+3];
 493             j_coord_offsetA  = DIM*jnrA;
 494             j_coord_offsetB  = DIM*jnrB;
 495             j_coord_offsetC  = DIM*jnrC;
 496             j_coord_offsetD  = DIM*jnrD;
 497
 498             /* load j atom coordinates */
 499             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 500                                               x+j_coord_offsetC,x+j_coord_offsetD,
 501                                               &jx0,&jy0,&jz0);
 502
 503             /* Calculate displacement vector */
 504             dx00             = _mm_sub_ps(ix0,jx0);
 505             dy00             = _mm_sub_ps(iy0,jy0);
 506             dz00             = _mm_sub_ps(iz0,jz0);
 507
 508             /* Calculate squared distance and things based on it */
 509             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 510
 511             rinv00           = sse2_invsqrt_f(rsq00);
 512
 513             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 514
 515             /* Load parameters for j particles */
 516             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 517                                                               charge+jnrC+0,charge+jnrD+0);
 518             vdwjidx0A        = 2*vdwtype[jnrA+0];
 519             vdwjidx0B        = 2*vdwtype[jnrB+0];
 520             vdwjidx0C        = 2*vdwtype[jnrC+0];
 521             vdwjidx0D        = 2*vdwtype[jnrD+0];
 522
 523             /**************************
 524              * CALCULATE INTERACTIONS *
 525              **************************/
 526
 527             /* Compute parameters for interactions between i and j atoms */
 528             qq00             = _mm_mul_ps(iq0,jq0);
 529             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 530                                          vdwparam+vdwioffset0+vdwjidx0B,
 531                                          vdwparam+vdwioffset0+vdwjidx0C,
 532                                          vdwparam+vdwioffset0+vdwjidx0D,
 533                                          &c6_00,&c12_00);
 534
 535             /* REACTION-FIELD ELECTROSTATICS */
 536             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 537
 538             /* LENNARD-JONES DISPERSION/REPULSION */
 539
 540             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 541             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 542
 543             fscal            = _mm_add_ps(felec,fvdw);
 544
 545             /* Calculate temporary vectorial force */
 546             tx               = _mm_mul_ps(fscal,dx00);
 547             ty               = _mm_mul_ps(fscal,dy00);
 548             tz               = _mm_mul_ps(fscal,dz00);
 549
 550             /* Update vectorial force */
 551             fix0             = _mm_add_ps(fix0,tx);
 552             fiy0             = _mm_add_ps(fiy0,ty);
 553             fiz0             = _mm_add_ps(fiz0,tz);
 554
 555             fjptrA             = f+j_coord_offsetA;
 556             fjptrB             = f+j_coord_offsetB;
 557             fjptrC             = f+j_coord_offsetC;
 558             fjptrD             = f+j_coord_offsetD;
 559             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 560
 561             /* Inner loop uses 34 flops */
 562         }
 563
 564         if(jidx<j_index_end)
 565         {
 566
 567             /* Get j neighbor index, and coordinate index */
 568             jnrlistA         = jjnr[jidx];
 569             jnrlistB         = jjnr[jidx+1];
 570             jnrlistC         = jjnr[jidx+2];
 571             jnrlistD         = jjnr[jidx+3];
 572             /* Sign of each element will be negative for non-real atoms.
 573              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 574              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 575              */
 576             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 577             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 578             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 579             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 580             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 581             j_coord_offsetA  = DIM*jnrA;
 582             j_coord_offsetB  = DIM*jnrB;
 583             j_coord_offsetC  = DIM*jnrC;
 584             j_coord_offsetD  = DIM*jnrD;
 585
 586             /* load j atom coordinates */
 587             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 588                                               x+j_coord_offsetC,x+j_coord_offsetD,
 589                                               &jx0,&jy0,&jz0);
 590
 591             /* Calculate displacement vector */
 592             dx00             = _mm_sub_ps(ix0,jx0);
 593             dy00             = _mm_sub_ps(iy0,jy0);
 594             dz00             = _mm_sub_ps(iz0,jz0);
 595
 596             /* Calculate squared distance and things based on it */
 597             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 598
 599             rinv00           = sse2_invsqrt_f(rsq00);
 600
 601             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 602
 603             /* Load parameters for j particles */
 604             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 605                                                               charge+jnrC+0,charge+jnrD+0);
 606             vdwjidx0A        = 2*vdwtype[jnrA+0];
 607             vdwjidx0B        = 2*vdwtype[jnrB+0];
 608             vdwjidx0C        = 2*vdwtype[jnrC+0];
 609             vdwjidx0D        = 2*vdwtype[jnrD+0];
 610
 611             /**************************
 612              * CALCULATE INTERACTIONS *
 613              **************************/
 614
 615             /* Compute parameters for interactions between i and j atoms */
 616             qq00             = _mm_mul_ps(iq0,jq0);
 617             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 618                                          vdwparam+vdwioffset0+vdwjidx0B,
 619                                          vdwparam+vdwioffset0+vdwjidx0C,
 620                                          vdwparam+vdwioffset0+vdwjidx0D,
 621                                          &c6_00,&c12_00);
 622
 623             /* REACTION-FIELD ELECTROSTATICS */
 624             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 625
 626             /* LENNARD-JONES DISPERSION/REPULSION */
 627
 628             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 629             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 630
 631             fscal            = _mm_add_ps(felec,fvdw);
 632
 633             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 634
 635             /* Calculate temporary vectorial force */
 636             tx               = _mm_mul_ps(fscal,dx00);
 637             ty               = _mm_mul_ps(fscal,dy00);
 638             tz               = _mm_mul_ps(fscal,dz00);
 639
 640             /* Update vectorial force */
 641             fix0             = _mm_add_ps(fix0,tx);
 642             fiy0             = _mm_add_ps(fiy0,ty);
 643             fiz0             = _mm_add_ps(fiz0,tz);
 644
 645             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 646             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 647             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 648             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 649             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 650
 651             /* Inner loop uses 34 flops */
 652         }
 653
 654         /* End of innermost loop */
 655
 656         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 657                                               f+i_coord_offset,fshift+i_shift_offset);
 658
 659         /* Increment number of inner iterations */
 660         inneriter                  += j_index_end - j_index_start;
 661
 662         /* Outer loop uses 7 flops */
 663     }
 664
 665     /* Increment number of outer iterations */
 666     outeriter        += nri;
 667
 668     /* Update outer/inner flops */
 669
 670     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*34);
 671 }