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, 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 "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_ElecRF_VdwLJ_GeomP1P1_VF_sse2_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_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     krf              = _mm_set1_ps(fr->ic->k_rf);
 116     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 117     crf              = _mm_set1_ps(fr->ic->c_rf);
 118     nvdwtype         = fr->ntype;
 119     vdwparam         = fr->nbfp;
 120     vdwtype          = mdatoms->typeA;
 121
 122     /* Avoid stupid compiler warnings */
 123     jnrA = jnrB = jnrC = jnrD = 0;
 124     j_coord_offsetA = 0;
 125     j_coord_offsetB = 0;
 126     j_coord_offsetC = 0;
 127     j_coord_offsetD = 0;
 128
 129     outeriter        = 0;
 130     inneriter        = 0;
 131
 132     for(iidx=0;iidx<4*DIM;iidx++)
 133     {
 134         scratch[iidx] = 0.0;
 135     }
 136
 137     /* Start outer loop over neighborlists */
 138     for(iidx=0; iidx<nri; iidx++)
 139     {
 140         /* Load shift vector for this list */
 141         i_shift_offset   = DIM*shiftidx[iidx];
 142
 143         /* Load limits for loop over neighbors */
 144         j_index_start    = jindex[iidx];
 145         j_index_end      = jindex[iidx+1];
 146
 147         /* Get outer coordinate index */
 148         inr              = iinr[iidx];
 149         i_coord_offset   = DIM*inr;
 150
 151         /* Load i particle coords and add shift vector */
 152         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 153
 154         fix0             = _mm_setzero_ps();
 155         fiy0             = _mm_setzero_ps();
 156         fiz0             = _mm_setzero_ps();
 157
 158         /* Load parameters for i particles */
 159         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 160         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 161
 162         /* Reset potential sums */
 163         velecsum         = _mm_setzero_ps();
 164         vvdwsum          = _mm_setzero_ps();
 165
 166         /* Start inner kernel loop */
 167         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 168         {
 169
 170             /* Get j neighbor index, and coordinate index */
 171             jnrA             = jjnr[jidx];
 172             jnrB             = jjnr[jidx+1];
 173             jnrC             = jjnr[jidx+2];
 174             jnrD             = jjnr[jidx+3];
 175             j_coord_offsetA  = DIM*jnrA;
 176             j_coord_offsetB  = DIM*jnrB;
 177             j_coord_offsetC  = DIM*jnrC;
 178             j_coord_offsetD  = DIM*jnrD;
 179
 180             /* load j atom coordinates */
 181             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 182                                               x+j_coord_offsetC,x+j_coord_offsetD,
 183                                               &jx0,&jy0,&jz0);
 184
 185             /* Calculate displacement vector */
 186             dx00             = _mm_sub_ps(ix0,jx0);
 187             dy00             = _mm_sub_ps(iy0,jy0);
 188             dz00             = _mm_sub_ps(iz0,jz0);
 189
 190             /* Calculate squared distance and things based on it */
 191             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 192
 193             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 194
 195             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 196
 197             /* Load parameters for j particles */
 198             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 199                                                               charge+jnrC+0,charge+jnrD+0);
 200             vdwjidx0A        = 2*vdwtype[jnrA+0];
 201             vdwjidx0B        = 2*vdwtype[jnrB+0];
 202             vdwjidx0C        = 2*vdwtype[jnrC+0];
 203             vdwjidx0D        = 2*vdwtype[jnrD+0];
 204
 205             /**************************
 206              * CALCULATE INTERACTIONS *
 207              **************************/
 208
 209             /* Compute parameters for interactions between i and j atoms */
 210             qq00             = _mm_mul_ps(iq0,jq0);
 211             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 212                                          vdwparam+vdwioffset0+vdwjidx0B,
 213                                          vdwparam+vdwioffset0+vdwjidx0C,
 214                                          vdwparam+vdwioffset0+vdwjidx0D,
 215                                          &c6_00,&c12_00);
 216
 217             /* REACTION-FIELD ELECTROSTATICS */
 218             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 219             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 220
 221             /* LENNARD-JONES DISPERSION/REPULSION */
 222
 223             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 224             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 225             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 226             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 227             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 228
 229             /* Update potential sum for this i atom from the interaction with this j atom. */
 230             velecsum         = _mm_add_ps(velecsum,velec);
 231             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 232
 233             fscal            = _mm_add_ps(felec,fvdw);
 234
 235             /* Calculate temporary vectorial force */
 236             tx               = _mm_mul_ps(fscal,dx00);
 237             ty               = _mm_mul_ps(fscal,dy00);
 238             tz               = _mm_mul_ps(fscal,dz00);
 239
 240             /* Update vectorial force */
 241             fix0             = _mm_add_ps(fix0,tx);
 242             fiy0             = _mm_add_ps(fiy0,ty);
 243             fiz0             = _mm_add_ps(fiz0,tz);
 244
 245             fjptrA             = f+j_coord_offsetA;
 246             fjptrB             = f+j_coord_offsetB;
 247             fjptrC             = f+j_coord_offsetC;
 248             fjptrD             = f+j_coord_offsetD;
 249             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 250
 251             /* Inner loop uses 44 flops */
 252         }
 253
 254         if(jidx<j_index_end)
 255         {
 256
 257             /* Get j neighbor index, and coordinate index */
 258             jnrlistA         = jjnr[jidx];
 259             jnrlistB         = jjnr[jidx+1];
 260             jnrlistC         = jjnr[jidx+2];
 261             jnrlistD         = jjnr[jidx+3];
 262             /* Sign of each element will be negative for non-real atoms.
 263              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 264              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 265              */
 266             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 267             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 268             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 269             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 270             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 271             j_coord_offsetA  = DIM*jnrA;
 272             j_coord_offsetB  = DIM*jnrB;
 273             j_coord_offsetC  = DIM*jnrC;
 274             j_coord_offsetD  = DIM*jnrD;
 275
 276             /* load j atom coordinates */
 277             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 278                                               x+j_coord_offsetC,x+j_coord_offsetD,
 279                                               &jx0,&jy0,&jz0);
 280
 281             /* Calculate displacement vector */
 282             dx00             = _mm_sub_ps(ix0,jx0);
 283             dy00             = _mm_sub_ps(iy0,jy0);
 284             dz00             = _mm_sub_ps(iz0,jz0);
 285
 286             /* Calculate squared distance and things based on it */
 287             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 288
 289             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 290
 291             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 292
 293             /* Load parameters for j particles */
 294             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 295                                                               charge+jnrC+0,charge+jnrD+0);
 296             vdwjidx0A        = 2*vdwtype[jnrA+0];
 297             vdwjidx0B        = 2*vdwtype[jnrB+0];
 298             vdwjidx0C        = 2*vdwtype[jnrC+0];
 299             vdwjidx0D        = 2*vdwtype[jnrD+0];
 300
 301             /**************************
 302              * CALCULATE INTERACTIONS *
 303              **************************/
 304
 305             /* Compute parameters for interactions between i and j atoms */
 306             qq00             = _mm_mul_ps(iq0,jq0);
 307             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 308                                          vdwparam+vdwioffset0+vdwjidx0B,
 309                                          vdwparam+vdwioffset0+vdwjidx0C,
 310                                          vdwparam+vdwioffset0+vdwjidx0D,
 311                                          &c6_00,&c12_00);
 312
 313             /* REACTION-FIELD ELECTROSTATICS */
 314             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 315             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 316
 317             /* LENNARD-JONES DISPERSION/REPULSION */
 318
 319             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 320             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 321             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 322             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 323             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 324
 325             /* Update potential sum for this i atom from the interaction with this j atom. */
 326             velec            = _mm_andnot_ps(dummy_mask,velec);
 327             velecsum         = _mm_add_ps(velecsum,velec);
 328             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 329             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 330
 331             fscal            = _mm_add_ps(felec,fvdw);
 332
 333             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 334
 335             /* Calculate temporary vectorial force */
 336             tx               = _mm_mul_ps(fscal,dx00);
 337             ty               = _mm_mul_ps(fscal,dy00);
 338             tz               = _mm_mul_ps(fscal,dz00);
 339
 340             /* Update vectorial force */
 341             fix0             = _mm_add_ps(fix0,tx);
 342             fiy0             = _mm_add_ps(fiy0,ty);
 343             fiz0             = _mm_add_ps(fiz0,tz);
 344
 345             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 346             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 347             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 348             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 349             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 350
 351             /* Inner loop uses 44 flops */
 352         }
 353
 354         /* End of innermost loop */
 355
 356         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 357                                               f+i_coord_offset,fshift+i_shift_offset);
 358
 359         ggid                        = gid[iidx];
 360         /* Update potential energies */
 361         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 362         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 363
 364         /* Increment number of inner iterations */
 365         inneriter                  += j_index_end - j_index_start;
 366
 367         /* Outer loop uses 9 flops */
 368     }
 369
 370     /* Increment number of outer iterations */
 371     outeriter        += nri;
 372
 373     /* Update outer/inner flops */
 374
 375     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*44);
 376 }
 377 /*
 378  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_sse2_single
 379  * Electrostatics interaction: ReactionField
 380  * VdW interaction:            LennardJones
 381  * Geometry:                   Particle-Particle
 382  * Calculate force/pot:        Force
 383  */
 384 void
 385 nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_sse2_single
 386                     (t_nblist                    * gmx_restrict       nlist,
 387                      rvec                        * gmx_restrict          xx,
 388                      rvec                        * gmx_restrict          ff,
 389                      t_forcerec                  * gmx_restrict          fr,
 390                      t_mdatoms                   * gmx_restrict     mdatoms,
 391                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 392                      t_nrnb                      * gmx_restrict        nrnb)
 393 {
 394     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 395      * just 0 for non-waters.
 396      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 397      * jnr indices corresponding to data put in the four positions in the SIMD register.
 398      */
 399     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 400     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 401     int              jnrA,jnrB,jnrC,jnrD;
 402     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 403     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 404     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 405     real             rcutoff_scalar;
 406     real             *shiftvec,*fshift,*x,*f;
 407     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 408     real             scratch[4*DIM];
 409     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 410     int              vdwioffset0;
 411     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 412     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 413     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 414     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 415     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 416     real             *charge;
 417     int              nvdwtype;
 418     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 419     int              *vdwtype;
 420     real             *vdwparam;
 421     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 422     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 423     __m128           dummy_mask,cutoff_mask;
 424     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 425     __m128           one     = _mm_set1_ps(1.0);
 426     __m128           two     = _mm_set1_ps(2.0);
 427     x                = xx[0];
 428     f                = ff[0];
 429
 430     nri              = nlist->nri;
 431     iinr             = nlist->iinr;
 432     jindex           = nlist->jindex;
 433     jjnr             = nlist->jjnr;
 434     shiftidx         = nlist->shift;
 435     gid              = nlist->gid;
 436     shiftvec         = fr->shift_vec[0];
 437     fshift           = fr->fshift[0];
 438     facel            = _mm_set1_ps(fr->epsfac);
 439     charge           = mdatoms->chargeA;
 440     krf              = _mm_set1_ps(fr->ic->k_rf);
 441     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 442     crf              = _mm_set1_ps(fr->ic->c_rf);
 443     nvdwtype         = fr->ntype;
 444     vdwparam         = fr->nbfp;
 445     vdwtype          = mdatoms->typeA;
 446
 447     /* Avoid stupid compiler warnings */
 448     jnrA = jnrB = jnrC = jnrD = 0;
 449     j_coord_offsetA = 0;
 450     j_coord_offsetB = 0;
 451     j_coord_offsetC = 0;
 452     j_coord_offsetD = 0;
 453
 454     outeriter        = 0;
 455     inneriter        = 0;
 456
 457     for(iidx=0;iidx<4*DIM;iidx++)
 458     {
 459         scratch[iidx] = 0.0;
 460     }
 461
 462     /* Start outer loop over neighborlists */
 463     for(iidx=0; iidx<nri; iidx++)
 464     {
 465         /* Load shift vector for this list */
 466         i_shift_offset   = DIM*shiftidx[iidx];
 467
 468         /* Load limits for loop over neighbors */
 469         j_index_start    = jindex[iidx];
 470         j_index_end      = jindex[iidx+1];
 471
 472         /* Get outer coordinate index */
 473         inr              = iinr[iidx];
 474         i_coord_offset   = DIM*inr;
 475
 476         /* Load i particle coords and add shift vector */
 477         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 478
 479         fix0             = _mm_setzero_ps();
 480         fiy0             = _mm_setzero_ps();
 481         fiz0             = _mm_setzero_ps();
 482
 483         /* Load parameters for i particles */
 484         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 485         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 486
 487         /* Start inner kernel loop */
 488         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 489         {
 490
 491             /* Get j neighbor index, and coordinate index */
 492             jnrA             = jjnr[jidx];
 493             jnrB             = jjnr[jidx+1];
 494             jnrC             = jjnr[jidx+2];
 495             jnrD             = jjnr[jidx+3];
 496             j_coord_offsetA  = DIM*jnrA;
 497             j_coord_offsetB  = DIM*jnrB;
 498             j_coord_offsetC  = DIM*jnrC;
 499             j_coord_offsetD  = DIM*jnrD;
 500
 501             /* load j atom coordinates */
 502             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 503                                               x+j_coord_offsetC,x+j_coord_offsetD,
 504                                               &jx0,&jy0,&jz0);
 505
 506             /* Calculate displacement vector */
 507             dx00             = _mm_sub_ps(ix0,jx0);
 508             dy00             = _mm_sub_ps(iy0,jy0);
 509             dz00             = _mm_sub_ps(iz0,jz0);
 510
 511             /* Calculate squared distance and things based on it */
 512             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 513
 514             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 515
 516             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 517
 518             /* Load parameters for j particles */
 519             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 520                                                               charge+jnrC+0,charge+jnrD+0);
 521             vdwjidx0A        = 2*vdwtype[jnrA+0];
 522             vdwjidx0B        = 2*vdwtype[jnrB+0];
 523             vdwjidx0C        = 2*vdwtype[jnrC+0];
 524             vdwjidx0D        = 2*vdwtype[jnrD+0];
 525
 526             /**************************
 527              * CALCULATE INTERACTIONS *
 528              **************************/
 529
 530             /* Compute parameters for interactions between i and j atoms */
 531             qq00             = _mm_mul_ps(iq0,jq0);
 532             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 533                                          vdwparam+vdwioffset0+vdwjidx0B,
 534                                          vdwparam+vdwioffset0+vdwjidx0C,
 535                                          vdwparam+vdwioffset0+vdwjidx0D,
 536                                          &c6_00,&c12_00);
 537
 538             /* REACTION-FIELD ELECTROSTATICS */
 539             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 540
 541             /* LENNARD-JONES DISPERSION/REPULSION */
 542
 543             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 544             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 545
 546             fscal            = _mm_add_ps(felec,fvdw);
 547
 548             /* Calculate temporary vectorial force */
 549             tx               = _mm_mul_ps(fscal,dx00);
 550             ty               = _mm_mul_ps(fscal,dy00);
 551             tz               = _mm_mul_ps(fscal,dz00);
 552
 553             /* Update vectorial force */
 554             fix0             = _mm_add_ps(fix0,tx);
 555             fiy0             = _mm_add_ps(fiy0,ty);
 556             fiz0             = _mm_add_ps(fiz0,tz);
 557
 558             fjptrA             = f+j_coord_offsetA;
 559             fjptrB             = f+j_coord_offsetB;
 560             fjptrC             = f+j_coord_offsetC;
 561             fjptrD             = f+j_coord_offsetD;
 562             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 563
 564             /* Inner loop uses 34 flops */
 565         }
 566
 567         if(jidx<j_index_end)
 568         {
 569
 570             /* Get j neighbor index, and coordinate index */
 571             jnrlistA         = jjnr[jidx];
 572             jnrlistB         = jjnr[jidx+1];
 573             jnrlistC         = jjnr[jidx+2];
 574             jnrlistD         = jjnr[jidx+3];
 575             /* Sign of each element will be negative for non-real atoms.
 576              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 577              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 578              */
 579             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 580             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 581             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 582             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 583             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 584             j_coord_offsetA  = DIM*jnrA;
 585             j_coord_offsetB  = DIM*jnrB;
 586             j_coord_offsetC  = DIM*jnrC;
 587             j_coord_offsetD  = DIM*jnrD;
 588
 589             /* load j atom coordinates */
 590             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 591                                               x+j_coord_offsetC,x+j_coord_offsetD,
 592                                               &jx0,&jy0,&jz0);
 593
 594             /* Calculate displacement vector */
 595             dx00             = _mm_sub_ps(ix0,jx0);
 596             dy00             = _mm_sub_ps(iy0,jy0);
 597             dz00             = _mm_sub_ps(iz0,jz0);
 598
 599             /* Calculate squared distance and things based on it */
 600             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 601
 602             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 603
 604             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 605
 606             /* Load parameters for j particles */
 607             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 608                                                               charge+jnrC+0,charge+jnrD+0);
 609             vdwjidx0A        = 2*vdwtype[jnrA+0];
 610             vdwjidx0B        = 2*vdwtype[jnrB+0];
 611             vdwjidx0C        = 2*vdwtype[jnrC+0];
 612             vdwjidx0D        = 2*vdwtype[jnrD+0];
 613
 614             /**************************
 615              * CALCULATE INTERACTIONS *
 616              **************************/
 617
 618             /* Compute parameters for interactions between i and j atoms */
 619             qq00             = _mm_mul_ps(iq0,jq0);
 620             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 621                                          vdwparam+vdwioffset0+vdwjidx0B,
 622                                          vdwparam+vdwioffset0+vdwjidx0C,
 623                                          vdwparam+vdwioffset0+vdwjidx0D,
 624                                          &c6_00,&c12_00);
 625
 626             /* REACTION-FIELD ELECTROSTATICS */
 627             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 628
 629             /* LENNARD-JONES DISPERSION/REPULSION */
 630
 631             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 632             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 633
 634             fscal            = _mm_add_ps(felec,fvdw);
 635
 636             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 637
 638             /* Calculate temporary vectorial force */
 639             tx               = _mm_mul_ps(fscal,dx00);
 640             ty               = _mm_mul_ps(fscal,dy00);
 641             tz               = _mm_mul_ps(fscal,dz00);
 642
 643             /* Update vectorial force */
 644             fix0             = _mm_add_ps(fix0,tx);
 645             fiy0             = _mm_add_ps(fiy0,ty);
 646             fiz0             = _mm_add_ps(fiz0,tz);
 647
 648             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 649             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 650             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 651             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 652             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 653
 654             /* Inner loop uses 34 flops */
 655         }
 656
 657         /* End of innermost loop */
 658
 659         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 660                                               f+i_coord_offset,fshift+i_shift_offset);
 661
 662         /* Increment number of inner iterations */
 663         inneriter                  += j_index_end - j_index_start;
 664
 665         /* Outer loop uses 7 flops */
 666     }
 667
 668     /* Increment number of outer iterations */
 669     outeriter        += nri;
 670
 671     /* Update outer/inner flops */
 672
 673     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*34);
 674 }