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, 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/math/vec.h"
  46 #include "gromacs/legacyheaders/nrnb.h"
  47
  48 #include "gromacs/simd/math_x86_sse2_single.h"
  49 #include "kernelutil_x86_sse2_single.h"
  50
  51 /*
  52  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomP1P1_VF_sse2_single
  53  * Electrostatics interaction: ReactionField
  54  * VdW interaction:            LennardJones
  55  * Geometry:                   Particle-Particle
  56  * Calculate force/pot:        PotentialAndForce
  57  */
  58 void
  59 nb_kernel_ElecRF_VdwLJ_GeomP1P1_VF_sse2_single
  60                     (t_nblist                    * gmx_restrict       nlist,
  61                      rvec                        * gmx_restrict          xx,
  62                      rvec                        * gmx_restrict          ff,
  63                      t_forcerec                  * gmx_restrict          fr,
  64                      t_mdatoms                   * gmx_restrict     mdatoms,
  65                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  66                      t_nrnb                      * gmx_restrict        nrnb)
  67 {
  68     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  69      * just 0 for non-waters.
  70      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  71      * jnr indices corresponding to data put in the four positions in the SIMD register.
  72      */
  73     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  74     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  75     int              jnrA,jnrB,jnrC,jnrD;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  84     int              vdwioffset0;
  85     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  86     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  87     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  90     real             *charge;
  91     int              nvdwtype;
  92     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  93     int              *vdwtype;
  94     real             *vdwparam;
  95     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  96     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  97     __m128           dummy_mask,cutoff_mask;
  98     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  99     __m128           one     = _mm_set1_ps(1.0);
 100     __m128           two     = _mm_set1_ps(2.0);
 101     x                = xx[0];
 102     f                = ff[0];
 103
 104     nri              = nlist->nri;
 105     iinr             = nlist->iinr;
 106     jindex           = nlist->jindex;
 107     jjnr             = nlist->jjnr;
 108     shiftidx         = nlist->shift;
 109     gid              = nlist->gid;
 110     shiftvec         = fr->shift_vec[0];
 111     fshift           = fr->fshift[0];
 112     facel            = _mm_set1_ps(fr->epsfac);
 113     charge           = mdatoms->chargeA;
 114     krf              = _mm_set1_ps(fr->ic->k_rf);
 115     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 116     crf              = _mm_set1_ps(fr->ic->c_rf);
 117     nvdwtype         = fr->ntype;
 118     vdwparam         = fr->nbfp;
 119     vdwtype          = mdatoms->typeA;
 120
 121     /* Avoid stupid compiler warnings */
 122     jnrA = jnrB = jnrC = jnrD = 0;
 123     j_coord_offsetA = 0;
 124     j_coord_offsetB = 0;
 125     j_coord_offsetC = 0;
 126     j_coord_offsetD = 0;
 127
 128     outeriter        = 0;
 129     inneriter        = 0;
 130
 131     for(iidx=0;iidx<4*DIM;iidx++)
 132     {
 133         scratch[iidx] = 0.0;
 134     }
 135
 136     /* Start outer loop over neighborlists */
 137     for(iidx=0; iidx<nri; iidx++)
 138     {
 139         /* Load shift vector for this list */
 140         i_shift_offset   = DIM*shiftidx[iidx];
 141
 142         /* Load limits for loop over neighbors */
 143         j_index_start    = jindex[iidx];
 144         j_index_end      = jindex[iidx+1];
 145
 146         /* Get outer coordinate index */
 147         inr              = iinr[iidx];
 148         i_coord_offset   = DIM*inr;
 149
 150         /* Load i particle coords and add shift vector */
 151         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 152
 153         fix0             = _mm_setzero_ps();
 154         fiy0             = _mm_setzero_ps();
 155         fiz0             = _mm_setzero_ps();
 156
 157         /* Load parameters for i particles */
 158         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 159         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 160
 161         /* Reset potential sums */
 162         velecsum         = _mm_setzero_ps();
 163         vvdwsum          = _mm_setzero_ps();
 164
 165         /* Start inner kernel loop */
 166         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 167         {
 168
 169             /* Get j neighbor index, and coordinate index */
 170             jnrA             = jjnr[jidx];
 171             jnrB             = jjnr[jidx+1];
 172             jnrC             = jjnr[jidx+2];
 173             jnrD             = jjnr[jidx+3];
 174             j_coord_offsetA  = DIM*jnrA;
 175             j_coord_offsetB  = DIM*jnrB;
 176             j_coord_offsetC  = DIM*jnrC;
 177             j_coord_offsetD  = DIM*jnrD;
 178
 179             /* load j atom coordinates */
 180             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 181                                               x+j_coord_offsetC,x+j_coord_offsetD,
 182                                               &jx0,&jy0,&jz0);
 183
 184             /* Calculate displacement vector */
 185             dx00             = _mm_sub_ps(ix0,jx0);
 186             dy00             = _mm_sub_ps(iy0,jy0);
 187             dz00             = _mm_sub_ps(iz0,jz0);
 188
 189             /* Calculate squared distance and things based on it */
 190             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 191
 192             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 193
 194             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 195
 196             /* Load parameters for j particles */
 197             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 198                                                               charge+jnrC+0,charge+jnrD+0);
 199             vdwjidx0A        = 2*vdwtype[jnrA+0];
 200             vdwjidx0B        = 2*vdwtype[jnrB+0];
 201             vdwjidx0C        = 2*vdwtype[jnrC+0];
 202             vdwjidx0D        = 2*vdwtype[jnrD+0];
 203
 204             /**************************
 205              * CALCULATE INTERACTIONS *
 206              **************************/
 207
 208             /* Compute parameters for interactions between i and j atoms */
 209             qq00             = _mm_mul_ps(iq0,jq0);
 210             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 211                                          vdwparam+vdwioffset0+vdwjidx0B,
 212                                          vdwparam+vdwioffset0+vdwjidx0C,
 213                                          vdwparam+vdwioffset0+vdwjidx0D,
 214                                          &c6_00,&c12_00);
 215
 216             /* REACTION-FIELD ELECTROSTATICS */
 217             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 218             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 219
 220             /* LENNARD-JONES DISPERSION/REPULSION */
 221
 222             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 223             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 224             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 225             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 226             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 227
 228             /* Update potential sum for this i atom from the interaction with this j atom. */
 229             velecsum         = _mm_add_ps(velecsum,velec);
 230             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 231
 232             fscal            = _mm_add_ps(felec,fvdw);
 233
 234             /* Calculate temporary vectorial force */
 235             tx               = _mm_mul_ps(fscal,dx00);
 236             ty               = _mm_mul_ps(fscal,dy00);
 237             tz               = _mm_mul_ps(fscal,dz00);
 238
 239             /* Update vectorial force */
 240             fix0             = _mm_add_ps(fix0,tx);
 241             fiy0             = _mm_add_ps(fiy0,ty);
 242             fiz0             = _mm_add_ps(fiz0,tz);
 243
 244             fjptrA             = f+j_coord_offsetA;
 245             fjptrB             = f+j_coord_offsetB;
 246             fjptrC             = f+j_coord_offsetC;
 247             fjptrD             = f+j_coord_offsetD;
 248             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 249
 250             /* Inner loop uses 44 flops */
 251         }
 252
 253         if(jidx<j_index_end)
 254         {
 255
 256             /* Get j neighbor index, and coordinate index */
 257             jnrlistA         = jjnr[jidx];
 258             jnrlistB         = jjnr[jidx+1];
 259             jnrlistC         = jjnr[jidx+2];
 260             jnrlistD         = jjnr[jidx+3];
 261             /* Sign of each element will be negative for non-real atoms.
 262              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 263              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 264              */
 265             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 266             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 267             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 268             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 269             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 270             j_coord_offsetA  = DIM*jnrA;
 271             j_coord_offsetB  = DIM*jnrB;
 272             j_coord_offsetC  = DIM*jnrC;
 273             j_coord_offsetD  = DIM*jnrD;
 274
 275             /* load j atom coordinates */
 276             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 277                                               x+j_coord_offsetC,x+j_coord_offsetD,
 278                                               &jx0,&jy0,&jz0);
 279
 280             /* Calculate displacement vector */
 281             dx00             = _mm_sub_ps(ix0,jx0);
 282             dy00             = _mm_sub_ps(iy0,jy0);
 283             dz00             = _mm_sub_ps(iz0,jz0);
 284
 285             /* Calculate squared distance and things based on it */
 286             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 287
 288             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 289
 290             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 291
 292             /* Load parameters for j particles */
 293             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 294                                                               charge+jnrC+0,charge+jnrD+0);
 295             vdwjidx0A        = 2*vdwtype[jnrA+0];
 296             vdwjidx0B        = 2*vdwtype[jnrB+0];
 297             vdwjidx0C        = 2*vdwtype[jnrC+0];
 298             vdwjidx0D        = 2*vdwtype[jnrD+0];
 299
 300             /**************************
 301              * CALCULATE INTERACTIONS *
 302              **************************/
 303
 304             /* Compute parameters for interactions between i and j atoms */
 305             qq00             = _mm_mul_ps(iq0,jq0);
 306             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 307                                          vdwparam+vdwioffset0+vdwjidx0B,
 308                                          vdwparam+vdwioffset0+vdwjidx0C,
 309                                          vdwparam+vdwioffset0+vdwjidx0D,
 310                                          &c6_00,&c12_00);
 311
 312             /* REACTION-FIELD ELECTROSTATICS */
 313             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 314             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 315
 316             /* LENNARD-JONES DISPERSION/REPULSION */
 317
 318             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 319             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 320             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 321             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 322             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 323
 324             /* Update potential sum for this i atom from the interaction with this j atom. */
 325             velec            = _mm_andnot_ps(dummy_mask,velec);
 326             velecsum         = _mm_add_ps(velecsum,velec);
 327             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 328             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 329
 330             fscal            = _mm_add_ps(felec,fvdw);
 331
 332             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 333
 334             /* Calculate temporary vectorial force */
 335             tx               = _mm_mul_ps(fscal,dx00);
 336             ty               = _mm_mul_ps(fscal,dy00);
 337             tz               = _mm_mul_ps(fscal,dz00);
 338
 339             /* Update vectorial force */
 340             fix0             = _mm_add_ps(fix0,tx);
 341             fiy0             = _mm_add_ps(fiy0,ty);
 342             fiz0             = _mm_add_ps(fiz0,tz);
 343
 344             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 345             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 346             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 347             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 348             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 349
 350             /* Inner loop uses 44 flops */
 351         }
 352
 353         /* End of innermost loop */
 354
 355         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 356                                               f+i_coord_offset,fshift+i_shift_offset);
 357
 358         ggid                        = gid[iidx];
 359         /* Update potential energies */
 360         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 361         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 362
 363         /* Increment number of inner iterations */
 364         inneriter                  += j_index_end - j_index_start;
 365
 366         /* Outer loop uses 9 flops */
 367     }
 368
 369     /* Increment number of outer iterations */
 370     outeriter        += nri;
 371
 372     /* Update outer/inner flops */
 373
 374     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*44);
 375 }
 376 /*
 377  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_sse2_single
 378  * Electrostatics interaction: ReactionField
 379  * VdW interaction:            LennardJones
 380  * Geometry:                   Particle-Particle
 381  * Calculate force/pot:        Force
 382  */
 383 void
 384 nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_sse2_single
 385                     (t_nblist                    * gmx_restrict       nlist,
 386                      rvec                        * gmx_restrict          xx,
 387                      rvec                        * gmx_restrict          ff,
 388                      t_forcerec                  * gmx_restrict          fr,
 389                      t_mdatoms                   * gmx_restrict     mdatoms,
 390                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 391                      t_nrnb                      * gmx_restrict        nrnb)
 392 {
 393     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 394      * just 0 for non-waters.
 395      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 396      * jnr indices corresponding to data put in the four positions in the SIMD register.
 397      */
 398     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 399     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 400     int              jnrA,jnrB,jnrC,jnrD;
 401     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 402     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 403     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 404     real             rcutoff_scalar;
 405     real             *shiftvec,*fshift,*x,*f;
 406     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 407     real             scratch[4*DIM];
 408     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 409     int              vdwioffset0;
 410     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 411     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 412     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 413     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 414     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 415     real             *charge;
 416     int              nvdwtype;
 417     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 418     int              *vdwtype;
 419     real             *vdwparam;
 420     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 421     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 422     __m128           dummy_mask,cutoff_mask;
 423     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 424     __m128           one     = _mm_set1_ps(1.0);
 425     __m128           two     = _mm_set1_ps(2.0);
 426     x                = xx[0];
 427     f                = ff[0];
 428
 429     nri              = nlist->nri;
 430     iinr             = nlist->iinr;
 431     jindex           = nlist->jindex;
 432     jjnr             = nlist->jjnr;
 433     shiftidx         = nlist->shift;
 434     gid              = nlist->gid;
 435     shiftvec         = fr->shift_vec[0];
 436     fshift           = fr->fshift[0];
 437     facel            = _mm_set1_ps(fr->epsfac);
 438     charge           = mdatoms->chargeA;
 439     krf              = _mm_set1_ps(fr->ic->k_rf);
 440     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 441     crf              = _mm_set1_ps(fr->ic->c_rf);
 442     nvdwtype         = fr->ntype;
 443     vdwparam         = fr->nbfp;
 444     vdwtype          = mdatoms->typeA;
 445
 446     /* Avoid stupid compiler warnings */
 447     jnrA = jnrB = jnrC = jnrD = 0;
 448     j_coord_offsetA = 0;
 449     j_coord_offsetB = 0;
 450     j_coord_offsetC = 0;
 451     j_coord_offsetD = 0;
 452
 453     outeriter        = 0;
 454     inneriter        = 0;
 455
 456     for(iidx=0;iidx<4*DIM;iidx++)
 457     {
 458         scratch[iidx] = 0.0;
 459     }
 460
 461     /* Start outer loop over neighborlists */
 462     for(iidx=0; iidx<nri; iidx++)
 463     {
 464         /* Load shift vector for this list */
 465         i_shift_offset   = DIM*shiftidx[iidx];
 466
 467         /* Load limits for loop over neighbors */
 468         j_index_start    = jindex[iidx];
 469         j_index_end      = jindex[iidx+1];
 470
 471         /* Get outer coordinate index */
 472         inr              = iinr[iidx];
 473         i_coord_offset   = DIM*inr;
 474
 475         /* Load i particle coords and add shift vector */
 476         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 477
 478         fix0             = _mm_setzero_ps();
 479         fiy0             = _mm_setzero_ps();
 480         fiz0             = _mm_setzero_ps();
 481
 482         /* Load parameters for i particles */
 483         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 484         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 485
 486         /* Start inner kernel loop */
 487         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 488         {
 489
 490             /* Get j neighbor index, and coordinate index */
 491             jnrA             = jjnr[jidx];
 492             jnrB             = jjnr[jidx+1];
 493             jnrC             = jjnr[jidx+2];
 494             jnrD             = jjnr[jidx+3];
 495             j_coord_offsetA  = DIM*jnrA;
 496             j_coord_offsetB  = DIM*jnrB;
 497             j_coord_offsetC  = DIM*jnrC;
 498             j_coord_offsetD  = DIM*jnrD;
 499
 500             /* load j atom coordinates */
 501             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 502                                               x+j_coord_offsetC,x+j_coord_offsetD,
 503                                               &jx0,&jy0,&jz0);
 504
 505             /* Calculate displacement vector */
 506             dx00             = _mm_sub_ps(ix0,jx0);
 507             dy00             = _mm_sub_ps(iy0,jy0);
 508             dz00             = _mm_sub_ps(iz0,jz0);
 509
 510             /* Calculate squared distance and things based on it */
 511             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 512
 513             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 514
 515             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 516
 517             /* Load parameters for j particles */
 518             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 519                                                               charge+jnrC+0,charge+jnrD+0);
 520             vdwjidx0A        = 2*vdwtype[jnrA+0];
 521             vdwjidx0B        = 2*vdwtype[jnrB+0];
 522             vdwjidx0C        = 2*vdwtype[jnrC+0];
 523             vdwjidx0D        = 2*vdwtype[jnrD+0];
 524
 525             /**************************
 526              * CALCULATE INTERACTIONS *
 527              **************************/
 528
 529             /* Compute parameters for interactions between i and j atoms */
 530             qq00             = _mm_mul_ps(iq0,jq0);
 531             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 532                                          vdwparam+vdwioffset0+vdwjidx0B,
 533                                          vdwparam+vdwioffset0+vdwjidx0C,
 534                                          vdwparam+vdwioffset0+vdwjidx0D,
 535                                          &c6_00,&c12_00);
 536
 537             /* REACTION-FIELD ELECTROSTATICS */
 538             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 539
 540             /* LENNARD-JONES DISPERSION/REPULSION */
 541
 542             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 543             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 544
 545             fscal            = _mm_add_ps(felec,fvdw);
 546
 547             /* Calculate temporary vectorial force */
 548             tx               = _mm_mul_ps(fscal,dx00);
 549             ty               = _mm_mul_ps(fscal,dy00);
 550             tz               = _mm_mul_ps(fscal,dz00);
 551
 552             /* Update vectorial force */
 553             fix0             = _mm_add_ps(fix0,tx);
 554             fiy0             = _mm_add_ps(fiy0,ty);
 555             fiz0             = _mm_add_ps(fiz0,tz);
 556
 557             fjptrA             = f+j_coord_offsetA;
 558             fjptrB             = f+j_coord_offsetB;
 559             fjptrC             = f+j_coord_offsetC;
 560             fjptrD             = f+j_coord_offsetD;
 561             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 562
 563             /* Inner loop uses 34 flops */
 564         }
 565
 566         if(jidx<j_index_end)
 567         {
 568
 569             /* Get j neighbor index, and coordinate index */
 570             jnrlistA         = jjnr[jidx];
 571             jnrlistB         = jjnr[jidx+1];
 572             jnrlistC         = jjnr[jidx+2];
 573             jnrlistD         = jjnr[jidx+3];
 574             /* Sign of each element will be negative for non-real atoms.
 575              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 576              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 577              */
 578             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 579             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 580             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 581             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 582             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 583             j_coord_offsetA  = DIM*jnrA;
 584             j_coord_offsetB  = DIM*jnrB;
 585             j_coord_offsetC  = DIM*jnrC;
 586             j_coord_offsetD  = DIM*jnrD;
 587
 588             /* load j atom coordinates */
 589             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 590                                               x+j_coord_offsetC,x+j_coord_offsetD,
 591                                               &jx0,&jy0,&jz0);
 592
 593             /* Calculate displacement vector */
 594             dx00             = _mm_sub_ps(ix0,jx0);
 595             dy00             = _mm_sub_ps(iy0,jy0);
 596             dz00             = _mm_sub_ps(iz0,jz0);
 597
 598             /* Calculate squared distance and things based on it */
 599             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 600
 601             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 602
 603             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 604
 605             /* Load parameters for j particles */
 606             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 607                                                               charge+jnrC+0,charge+jnrD+0);
 608             vdwjidx0A        = 2*vdwtype[jnrA+0];
 609             vdwjidx0B        = 2*vdwtype[jnrB+0];
 610             vdwjidx0C        = 2*vdwtype[jnrC+0];
 611             vdwjidx0D        = 2*vdwtype[jnrD+0];
 612
 613             /**************************
 614              * CALCULATE INTERACTIONS *
 615              **************************/
 616
 617             /* Compute parameters for interactions between i and j atoms */
 618             qq00             = _mm_mul_ps(iq0,jq0);
 619             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 620                                          vdwparam+vdwioffset0+vdwjidx0B,
 621                                          vdwparam+vdwioffset0+vdwjidx0C,
 622                                          vdwparam+vdwioffset0+vdwjidx0D,
 623                                          &c6_00,&c12_00);
 624
 625             /* REACTION-FIELD ELECTROSTATICS */
 626             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 627
 628             /* LENNARD-JONES DISPERSION/REPULSION */
 629
 630             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 631             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 632
 633             fscal            = _mm_add_ps(felec,fvdw);
 634
 635             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 636
 637             /* Calculate temporary vectorial force */
 638             tx               = _mm_mul_ps(fscal,dx00);
 639             ty               = _mm_mul_ps(fscal,dy00);
 640             tz               = _mm_mul_ps(fscal,dz00);
 641
 642             /* Update vectorial force */
 643             fix0             = _mm_add_ps(fix0,tx);
 644             fiy0             = _mm_add_ps(fiy0,ty);
 645             fiz0             = _mm_add_ps(fiz0,tz);
 646
 647             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 648             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 649             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 650             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 651             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 652
 653             /* Inner loop uses 34 flops */
 654         }
 655
 656         /* End of innermost loop */
 657
 658         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 659                                               f+i_coord_offset,fshift+i_shift_offset);
 660
 661         /* Increment number of inner iterations */
 662         inneriter                  += j_index_end - j_index_start;
 663
 664         /* Outer loop uses 7 flops */
 665     }
 666
 667     /* Increment number of outer iterations */
 668     outeriter        += nri;
 669
 670     /* Update outer/inner flops */
 671
 672     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*34);
 673 }