src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCoul_VdwLJ_GeomP1P1_sse2_single.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 sse2_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_single.h"
  50 #include "kernelutil_x86_sse2_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomP1P1_VF_sse2_single
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_VdwLJ_GeomP1P1_VF_sse2_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  88     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     int              nvdwtype;
  93     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  97     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  98     __m128           dummy_mask,cutoff_mask;
  99     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 100     __m128           one     = _mm_set1_ps(1.0);
 101     __m128           two     = _mm_set1_ps(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm_set1_ps(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 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           = gmx_mm_invsqrt_ps(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             /* COULOMB ELECTROSTATICS */
 215             velec            = _mm_mul_ps(qq00,rinv00);
 216             felec            = _mm_mul_ps(velec,rinvsq00);
 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 40 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           = gmx_mm_invsqrt_ps(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             /* COULOMB ELECTROSTATICS */
 311             velec            = _mm_mul_ps(qq00,rinv00);
 312             felec            = _mm_mul_ps(velec,rinvsq00);
 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 40 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*40);
 373 }
 374 /*
 375  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomP1P1_F_sse2_single
 376  * Electrostatics interaction: Coulomb
 377  * VdW interaction:            LennardJones
 378  * Geometry:                   Particle-Particle
 379  * Calculate force/pot:        Force
 380  */
 381 void
 382 nb_kernel_ElecCoul_VdwLJ_GeomP1P1_F_sse2_single
 383                     (t_nblist                    * gmx_restrict       nlist,
 384                      rvec                        * gmx_restrict          xx,
 385                      rvec                        * gmx_restrict          ff,
 386                      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->epsfac);
 436     charge           = mdatoms->chargeA;
 437     nvdwtype         = fr->ntype;
 438     vdwparam         = fr->nbfp;
 439     vdwtype          = mdatoms->typeA;
 440
 441     /* Avoid stupid compiler warnings */
 442     jnrA = jnrB = jnrC = jnrD = 0;
 443     j_coord_offsetA = 0;
 444     j_coord_offsetB = 0;
 445     j_coord_offsetC = 0;
 446     j_coord_offsetD = 0;
 447
 448     outeriter        = 0;
 449     inneriter        = 0;
 450
 451     for(iidx=0;iidx<4*DIM;iidx++)
 452     {
 453         scratch[iidx] = 0.0;
 454     }
 455
 456     /* Start outer loop over neighborlists */
 457     for(iidx=0; iidx<nri; iidx++)
 458     {
 459         /* Load shift vector for this list */
 460         i_shift_offset   = DIM*shiftidx[iidx];
 461
 462         /* Load limits for loop over neighbors */
 463         j_index_start    = jindex[iidx];
 464         j_index_end      = jindex[iidx+1];
 465
 466         /* Get outer coordinate index */
 467         inr              = iinr[iidx];
 468         i_coord_offset   = DIM*inr;
 469
 470         /* Load i particle coords and add shift vector */
 471         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 472
 473         fix0             = _mm_setzero_ps();
 474         fiy0             = _mm_setzero_ps();
 475         fiz0             = _mm_setzero_ps();
 476
 477         /* Load parameters for i particles */
 478         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 479         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 480
 481         /* Start inner kernel loop */
 482         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 483         {
 484
 485             /* Get j neighbor index, and coordinate index */
 486             jnrA             = jjnr[jidx];
 487             jnrB             = jjnr[jidx+1];
 488             jnrC             = jjnr[jidx+2];
 489             jnrD             = jjnr[jidx+3];
 490             j_coord_offsetA  = DIM*jnrA;
 491             j_coord_offsetB  = DIM*jnrB;
 492             j_coord_offsetC  = DIM*jnrC;
 493             j_coord_offsetD  = DIM*jnrD;
 494
 495             /* load j atom coordinates */
 496             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 497                                               x+j_coord_offsetC,x+j_coord_offsetD,
 498                                               &jx0,&jy0,&jz0);
 499
 500             /* Calculate displacement vector */
 501             dx00             = _mm_sub_ps(ix0,jx0);
 502             dy00             = _mm_sub_ps(iy0,jy0);
 503             dz00             = _mm_sub_ps(iz0,jz0);
 504
 505             /* Calculate squared distance and things based on it */
 506             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 507
 508             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 509
 510             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 511
 512             /* Load parameters for j particles */
 513             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 514                                                               charge+jnrC+0,charge+jnrD+0);
 515             vdwjidx0A        = 2*vdwtype[jnrA+0];
 516             vdwjidx0B        = 2*vdwtype[jnrB+0];
 517             vdwjidx0C        = 2*vdwtype[jnrC+0];
 518             vdwjidx0D        = 2*vdwtype[jnrD+0];
 519
 520             /**************************
 521              * CALCULATE INTERACTIONS *
 522              **************************/
 523
 524             /* Compute parameters for interactions between i and j atoms */
 525             qq00             = _mm_mul_ps(iq0,jq0);
 526             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 527                                          vdwparam+vdwioffset0+vdwjidx0B,
 528                                          vdwparam+vdwioffset0+vdwjidx0C,
 529                                          vdwparam+vdwioffset0+vdwjidx0D,
 530                                          &c6_00,&c12_00);
 531
 532             /* COULOMB ELECTROSTATICS */
 533             velec            = _mm_mul_ps(qq00,rinv00);
 534             felec            = _mm_mul_ps(velec,rinvsq00);
 535
 536             /* LENNARD-JONES DISPERSION/REPULSION */
 537
 538             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 539             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 540
 541             fscal            = _mm_add_ps(felec,fvdw);
 542
 543             /* Calculate temporary vectorial force */
 544             tx               = _mm_mul_ps(fscal,dx00);
 545             ty               = _mm_mul_ps(fscal,dy00);
 546             tz               = _mm_mul_ps(fscal,dz00);
 547
 548             /* Update vectorial force */
 549             fix0             = _mm_add_ps(fix0,tx);
 550             fiy0             = _mm_add_ps(fiy0,ty);
 551             fiz0             = _mm_add_ps(fiz0,tz);
 552
 553             fjptrA             = f+j_coord_offsetA;
 554             fjptrB             = f+j_coord_offsetB;
 555             fjptrC             = f+j_coord_offsetC;
 556             fjptrD             = f+j_coord_offsetD;
 557             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 558
 559             /* Inner loop uses 34 flops */
 560         }
 561
 562         if(jidx<j_index_end)
 563         {
 564
 565             /* Get j neighbor index, and coordinate index */
 566             jnrlistA         = jjnr[jidx];
 567             jnrlistB         = jjnr[jidx+1];
 568             jnrlistC         = jjnr[jidx+2];
 569             jnrlistD         = jjnr[jidx+3];
 570             /* Sign of each element will be negative for non-real atoms.
 571              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 572              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 573              */
 574             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 575             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 576             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 577             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 578             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 579             j_coord_offsetA  = DIM*jnrA;
 580             j_coord_offsetB  = DIM*jnrB;
 581             j_coord_offsetC  = DIM*jnrC;
 582             j_coord_offsetD  = DIM*jnrD;
 583
 584             /* load j atom coordinates */
 585             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 586                                               x+j_coord_offsetC,x+j_coord_offsetD,
 587                                               &jx0,&jy0,&jz0);
 588
 589             /* Calculate displacement vector */
 590             dx00             = _mm_sub_ps(ix0,jx0);
 591             dy00             = _mm_sub_ps(iy0,jy0);
 592             dz00             = _mm_sub_ps(iz0,jz0);
 593
 594             /* Calculate squared distance and things based on it */
 595             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 596
 597             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 598
 599             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 600
 601             /* Load parameters for j particles */
 602             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 603                                                               charge+jnrC+0,charge+jnrD+0);
 604             vdwjidx0A        = 2*vdwtype[jnrA+0];
 605             vdwjidx0B        = 2*vdwtype[jnrB+0];
 606             vdwjidx0C        = 2*vdwtype[jnrC+0];
 607             vdwjidx0D        = 2*vdwtype[jnrD+0];
 608
 609             /**************************
 610              * CALCULATE INTERACTIONS *
 611              **************************/
 612
 613             /* Compute parameters for interactions between i and j atoms */
 614             qq00             = _mm_mul_ps(iq0,jq0);
 615             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 616                                          vdwparam+vdwioffset0+vdwjidx0B,
 617                                          vdwparam+vdwioffset0+vdwjidx0C,
 618                                          vdwparam+vdwioffset0+vdwjidx0D,
 619                                          &c6_00,&c12_00);
 620
 621             /* COULOMB ELECTROSTATICS */
 622             velec            = _mm_mul_ps(qq00,rinv00);
 623             felec            = _mm_mul_ps(velec,rinvsq00);
 624
 625             /* LENNARD-JONES DISPERSION/REPULSION */
 626
 627             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 628             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 629
 630             fscal            = _mm_add_ps(felec,fvdw);
 631
 632             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 633
 634             /* Calculate temporary vectorial force */
 635             tx               = _mm_mul_ps(fscal,dx00);
 636             ty               = _mm_mul_ps(fscal,dy00);
 637             tz               = _mm_mul_ps(fscal,dz00);
 638
 639             /* Update vectorial force */
 640             fix0             = _mm_add_ps(fix0,tx);
 641             fiy0             = _mm_add_ps(fiy0,ty);
 642             fiz0             = _mm_add_ps(fiz0,tz);
 643
 644             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 645             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 646             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 647             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 648             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 649
 650             /* Inner loop uses 34 flops */
 651         }
 652
 653         /* End of innermost loop */
 654
 655         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 656                                               f+i_coord_offset,fshift+i_shift_offset);
 657
 658         /* Increment number of inner iterations */
 659         inneriter                  += j_index_end - j_index_start;
 660
 661         /* Outer loop uses 7 flops */
 662     }
 663
 664     /* Increment number of outer iterations */
 665     outeriter        += nri;
 666
 667     /* Update outer/inner flops */
 668
 669     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*34);
 670 }