src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRF_VdwLJ_GeomP1P1_avx_128_fma_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 avx_128_fma_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_avx_128_fma_single.h"
  50 #include "kernelutil_x86_avx_128_fma_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomP1P1_VF_avx_128_fma_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_avx_128_fma_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 AVX_128, 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           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_macc_ps(krf,rsq00,rinv00),crf));
 219             felec            = _mm_mul_ps(qq00,_mm_msub_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_msub_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              /* Update vectorial force */
 236             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 237             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 238             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 239
 240             fjptrA             = f+j_coord_offsetA;
 241             fjptrB             = f+j_coord_offsetB;
 242             fjptrC             = f+j_coord_offsetC;
 243             fjptrD             = f+j_coord_offsetD;
 244             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 245                                                    _mm_mul_ps(dx00,fscal),
 246                                                    _mm_mul_ps(dy00,fscal),
 247                                                    _mm_mul_ps(dz00,fscal));
 248
 249             /* Inner loop uses 47 flops */
 250         }
 251
 252         if(jidx<j_index_end)
 253         {
 254
 255             /* Get j neighbor index, and coordinate index */
 256             jnrlistA         = jjnr[jidx];
 257             jnrlistB         = jjnr[jidx+1];
 258             jnrlistC         = jjnr[jidx+2];
 259             jnrlistD         = jjnr[jidx+3];
 260             /* Sign of each element will be negative for non-real atoms.
 261              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 262              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 263              */
 264             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 265             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 266             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 267             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 268             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 269             j_coord_offsetA  = DIM*jnrA;
 270             j_coord_offsetB  = DIM*jnrB;
 271             j_coord_offsetC  = DIM*jnrC;
 272             j_coord_offsetD  = DIM*jnrD;
 273
 274             /* load j atom coordinates */
 275             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 276                                               x+j_coord_offsetC,x+j_coord_offsetD,
 277                                               &jx0,&jy0,&jz0);
 278
 279             /* Calculate displacement vector */
 280             dx00             = _mm_sub_ps(ix0,jx0);
 281             dy00             = _mm_sub_ps(iy0,jy0);
 282             dz00             = _mm_sub_ps(iz0,jz0);
 283
 284             /* Calculate squared distance and things based on it */
 285             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 286
 287             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 288
 289             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 290
 291             /* Load parameters for j particles */
 292             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 293                                                               charge+jnrC+0,charge+jnrD+0);
 294             vdwjidx0A        = 2*vdwtype[jnrA+0];
 295             vdwjidx0B        = 2*vdwtype[jnrB+0];
 296             vdwjidx0C        = 2*vdwtype[jnrC+0];
 297             vdwjidx0D        = 2*vdwtype[jnrD+0];
 298
 299             /**************************
 300              * CALCULATE INTERACTIONS *
 301              **************************/
 302
 303             /* Compute parameters for interactions between i and j atoms */
 304             qq00             = _mm_mul_ps(iq0,jq0);
 305             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 306                                          vdwparam+vdwioffset0+vdwjidx0B,
 307                                          vdwparam+vdwioffset0+vdwjidx0C,
 308                                          vdwparam+vdwioffset0+vdwjidx0D,
 309                                          &c6_00,&c12_00);
 310
 311             /* REACTION-FIELD ELECTROSTATICS */
 312             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 313             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 314
 315             /* LENNARD-JONES DISPERSION/REPULSION */
 316
 317             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 318             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 319             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 320             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 321             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 322
 323             /* Update potential sum for this i atom from the interaction with this j atom. */
 324             velec            = _mm_andnot_ps(dummy_mask,velec);
 325             velecsum         = _mm_add_ps(velecsum,velec);
 326             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 327             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 328
 329             fscal            = _mm_add_ps(felec,fvdw);
 330
 331             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 332
 333              /* Update vectorial force */
 334             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 335             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 336             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 337
 338             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 339             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 340             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 341             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 342             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 343                                                    _mm_mul_ps(dx00,fscal),
 344                                                    _mm_mul_ps(dy00,fscal),
 345                                                    _mm_mul_ps(dz00,fscal));
 346
 347             /* Inner loop uses 47 flops */
 348         }
 349
 350         /* End of innermost loop */
 351
 352         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 353                                               f+i_coord_offset,fshift+i_shift_offset);
 354
 355         ggid                        = gid[iidx];
 356         /* Update potential energies */
 357         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 358         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 359
 360         /* Increment number of inner iterations */
 361         inneriter                  += j_index_end - j_index_start;
 362
 363         /* Outer loop uses 9 flops */
 364     }
 365
 366     /* Increment number of outer iterations */
 367     outeriter        += nri;
 368
 369     /* Update outer/inner flops */
 370
 371     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*47);
 372 }
 373 /*
 374  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_avx_128_fma_single
 375  * Electrostatics interaction: ReactionField
 376  * VdW interaction:            LennardJones
 377  * Geometry:                   Particle-Particle
 378  * Calculate force/pot:        Force
 379  */
 380 void
 381 nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_avx_128_fma_single
 382                     (t_nblist                    * gmx_restrict       nlist,
 383                      rvec                        * gmx_restrict          xx,
 384                      rvec                        * gmx_restrict          ff,
 385                      t_forcerec                  * gmx_restrict          fr,
 386                      t_mdatoms                   * gmx_restrict     mdatoms,
 387                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 388                      t_nrnb                      * gmx_restrict        nrnb)
 389 {
 390     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 391      * just 0 for non-waters.
 392      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 393      * jnr indices corresponding to data put in the four positions in the SIMD register.
 394      */
 395     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 396     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 397     int              jnrA,jnrB,jnrC,jnrD;
 398     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 399     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 400     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 401     real             rcutoff_scalar;
 402     real             *shiftvec,*fshift,*x,*f;
 403     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 404     real             scratch[4*DIM];
 405     __m128           fscal,rcutoff,rcutoff2,jidxall;
 406     int              vdwioffset0;
 407     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 408     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 409     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 410     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 411     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 412     real             *charge;
 413     int              nvdwtype;
 414     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 415     int              *vdwtype;
 416     real             *vdwparam;
 417     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 418     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 419     __m128           dummy_mask,cutoff_mask;
 420     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 421     __m128           one     = _mm_set1_ps(1.0);
 422     __m128           two     = _mm_set1_ps(2.0);
 423     x                = xx[0];
 424     f                = ff[0];
 425
 426     nri              = nlist->nri;
 427     iinr             = nlist->iinr;
 428     jindex           = nlist->jindex;
 429     jjnr             = nlist->jjnr;
 430     shiftidx         = nlist->shift;
 431     gid              = nlist->gid;
 432     shiftvec         = fr->shift_vec[0];
 433     fshift           = fr->fshift[0];
 434     facel            = _mm_set1_ps(fr->epsfac);
 435     charge           = mdatoms->chargeA;
 436     krf              = _mm_set1_ps(fr->ic->k_rf);
 437     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 438     crf              = _mm_set1_ps(fr->ic->c_rf);
 439     nvdwtype         = fr->ntype;
 440     vdwparam         = fr->nbfp;
 441     vdwtype          = mdatoms->typeA;
 442
 443     /* Avoid stupid compiler warnings */
 444     jnrA = jnrB = jnrC = jnrD = 0;
 445     j_coord_offsetA = 0;
 446     j_coord_offsetB = 0;
 447     j_coord_offsetC = 0;
 448     j_coord_offsetD = 0;
 449
 450     outeriter        = 0;
 451     inneriter        = 0;
 452
 453     for(iidx=0;iidx<4*DIM;iidx++)
 454     {
 455         scratch[iidx] = 0.0;
 456     }
 457
 458     /* Start outer loop over neighborlists */
 459     for(iidx=0; iidx<nri; iidx++)
 460     {
 461         /* Load shift vector for this list */
 462         i_shift_offset   = DIM*shiftidx[iidx];
 463
 464         /* Load limits for loop over neighbors */
 465         j_index_start    = jindex[iidx];
 466         j_index_end      = jindex[iidx+1];
 467
 468         /* Get outer coordinate index */
 469         inr              = iinr[iidx];
 470         i_coord_offset   = DIM*inr;
 471
 472         /* Load i particle coords and add shift vector */
 473         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 474
 475         fix0             = _mm_setzero_ps();
 476         fiy0             = _mm_setzero_ps();
 477         fiz0             = _mm_setzero_ps();
 478
 479         /* Load parameters for i particles */
 480         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 481         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 482
 483         /* Start inner kernel loop */
 484         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 485         {
 486
 487             /* Get j neighbor index, and coordinate index */
 488             jnrA             = jjnr[jidx];
 489             jnrB             = jjnr[jidx+1];
 490             jnrC             = jjnr[jidx+2];
 491             jnrD             = jjnr[jidx+3];
 492             j_coord_offsetA  = DIM*jnrA;
 493             j_coord_offsetB  = DIM*jnrB;
 494             j_coord_offsetC  = DIM*jnrC;
 495             j_coord_offsetD  = DIM*jnrD;
 496
 497             /* load j atom coordinates */
 498             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 499                                               x+j_coord_offsetC,x+j_coord_offsetD,
 500                                               &jx0,&jy0,&jz0);
 501
 502             /* Calculate displacement vector */
 503             dx00             = _mm_sub_ps(ix0,jx0);
 504             dy00             = _mm_sub_ps(iy0,jy0);
 505             dz00             = _mm_sub_ps(iz0,jz0);
 506
 507             /* Calculate squared distance and things based on it */
 508             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 509
 510             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 511
 512             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 513
 514             /* Load parameters for j particles */
 515             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 516                                                               charge+jnrC+0,charge+jnrD+0);
 517             vdwjidx0A        = 2*vdwtype[jnrA+0];
 518             vdwjidx0B        = 2*vdwtype[jnrB+0];
 519             vdwjidx0C        = 2*vdwtype[jnrC+0];
 520             vdwjidx0D        = 2*vdwtype[jnrD+0];
 521
 522             /**************************
 523              * CALCULATE INTERACTIONS *
 524              **************************/
 525
 526             /* Compute parameters for interactions between i and j atoms */
 527             qq00             = _mm_mul_ps(iq0,jq0);
 528             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 529                                          vdwparam+vdwioffset0+vdwjidx0B,
 530                                          vdwparam+vdwioffset0+vdwjidx0C,
 531                                          vdwparam+vdwioffset0+vdwjidx0D,
 532                                          &c6_00,&c12_00);
 533
 534             /* REACTION-FIELD ELECTROSTATICS */
 535             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 536
 537             /* LENNARD-JONES DISPERSION/REPULSION */
 538
 539             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 540             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 541
 542             fscal            = _mm_add_ps(felec,fvdw);
 543
 544              /* Update vectorial force */
 545             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 546             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 547             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 548
 549             fjptrA             = f+j_coord_offsetA;
 550             fjptrB             = f+j_coord_offsetB;
 551             fjptrC             = f+j_coord_offsetC;
 552             fjptrD             = f+j_coord_offsetD;
 553             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 554                                                    _mm_mul_ps(dx00,fscal),
 555                                                    _mm_mul_ps(dy00,fscal),
 556                                                    _mm_mul_ps(dz00,fscal));
 557
 558             /* Inner loop uses 37 flops */
 559         }
 560
 561         if(jidx<j_index_end)
 562         {
 563
 564             /* Get j neighbor index, and coordinate index */
 565             jnrlistA         = jjnr[jidx];
 566             jnrlistB         = jjnr[jidx+1];
 567             jnrlistC         = jjnr[jidx+2];
 568             jnrlistD         = jjnr[jidx+3];
 569             /* Sign of each element will be negative for non-real atoms.
 570              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 571              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 572              */
 573             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 574             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 575             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 576             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 577             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 578             j_coord_offsetA  = DIM*jnrA;
 579             j_coord_offsetB  = DIM*jnrB;
 580             j_coord_offsetC  = DIM*jnrC;
 581             j_coord_offsetD  = DIM*jnrD;
 582
 583             /* load j atom coordinates */
 584             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 585                                               x+j_coord_offsetC,x+j_coord_offsetD,
 586                                               &jx0,&jy0,&jz0);
 587
 588             /* Calculate displacement vector */
 589             dx00             = _mm_sub_ps(ix0,jx0);
 590             dy00             = _mm_sub_ps(iy0,jy0);
 591             dz00             = _mm_sub_ps(iz0,jz0);
 592
 593             /* Calculate squared distance and things based on it */
 594             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 595
 596             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 597
 598             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 599
 600             /* Load parameters for j particles */
 601             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 602                                                               charge+jnrC+0,charge+jnrD+0);
 603             vdwjidx0A        = 2*vdwtype[jnrA+0];
 604             vdwjidx0B        = 2*vdwtype[jnrB+0];
 605             vdwjidx0C        = 2*vdwtype[jnrC+0];
 606             vdwjidx0D        = 2*vdwtype[jnrD+0];
 607
 608             /**************************
 609              * CALCULATE INTERACTIONS *
 610              **************************/
 611
 612             /* Compute parameters for interactions between i and j atoms */
 613             qq00             = _mm_mul_ps(iq0,jq0);
 614             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 615                                          vdwparam+vdwioffset0+vdwjidx0B,
 616                                          vdwparam+vdwioffset0+vdwjidx0C,
 617                                          vdwparam+vdwioffset0+vdwjidx0D,
 618                                          &c6_00,&c12_00);
 619
 620             /* REACTION-FIELD ELECTROSTATICS */
 621             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 622
 623             /* LENNARD-JONES DISPERSION/REPULSION */
 624
 625             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 626             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 627
 628             fscal            = _mm_add_ps(felec,fvdw);
 629
 630             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 631
 632              /* Update vectorial force */
 633             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 634             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 635             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 636
 637             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 638             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 639             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 640             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 641             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 642                                                    _mm_mul_ps(dx00,fscal),
 643                                                    _mm_mul_ps(dy00,fscal),
 644                                                    _mm_mul_ps(dz00,fscal));
 645
 646             /* Inner loop uses 37 flops */
 647         }
 648
 649         /* End of innermost loop */
 650
 651         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 652                                               f+i_coord_offset,fshift+i_shift_offset);
 653
 654         /* Increment number of inner iterations */
 655         inneriter                  += j_index_end - j_index_start;
 656
 657         /* Outer loop uses 7 flops */
 658     }
 659
 660     /* Increment number of outer iterations */
 661     outeriter        += nri;
 662
 663     /* Update outer/inner flops */
 664
 665     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*37);
 666 }