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 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomP1P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRF_VdwLJ_GeomP1P1_VF_avx_128_fma_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  86     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  89     real             *charge;
  90     int              nvdwtype;
  91     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  92     int              *vdwtype;
  93     real             *vdwparam;
  94     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  95     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  96     __m128           dummy_mask,cutoff_mask;
  97     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  98     __m128           one     = _mm_set1_ps(1.0);
  99     __m128           two     = _mm_set1_ps(2.0);
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     facel            = _mm_set1_ps(fr->epsfac);
 112     charge           = mdatoms->chargeA;
 113     krf              = _mm_set1_ps(fr->ic->k_rf);
 114     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 115     crf              = _mm_set1_ps(fr->ic->c_rf);
 116     nvdwtype         = fr->ntype;
 117     vdwparam         = fr->nbfp;
 118     vdwtype          = mdatoms->typeA;
 119
 120     /* Avoid stupid compiler warnings */
 121     jnrA = jnrB = jnrC = jnrD = 0;
 122     j_coord_offsetA = 0;
 123     j_coord_offsetB = 0;
 124     j_coord_offsetC = 0;
 125     j_coord_offsetD = 0;
 126
 127     outeriter        = 0;
 128     inneriter        = 0;
 129
 130     for(iidx=0;iidx<4*DIM;iidx++)
 131     {
 132         scratch[iidx] = 0.0;
 133     }
 134
 135     /* Start outer loop over neighborlists */
 136     for(iidx=0; iidx<nri; iidx++)
 137     {
 138         /* Load shift vector for this list */
 139         i_shift_offset   = DIM*shiftidx[iidx];
 140
 141         /* Load limits for loop over neighbors */
 142         j_index_start    = jindex[iidx];
 143         j_index_end      = jindex[iidx+1];
 144
 145         /* Get outer coordinate index */
 146         inr              = iinr[iidx];
 147         i_coord_offset   = DIM*inr;
 148
 149         /* Load i particle coords and add shift vector */
 150         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 151
 152         fix0             = _mm_setzero_ps();
 153         fiy0             = _mm_setzero_ps();
 154         fiz0             = _mm_setzero_ps();
 155
 156         /* Load parameters for i particles */
 157         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 158         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 159
 160         /* Reset potential sums */
 161         velecsum         = _mm_setzero_ps();
 162         vvdwsum          = _mm_setzero_ps();
 163
 164         /* Start inner kernel loop */
 165         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 166         {
 167
 168             /* Get j neighbor index, and coordinate index */
 169             jnrA             = jjnr[jidx];
 170             jnrB             = jjnr[jidx+1];
 171             jnrC             = jjnr[jidx+2];
 172             jnrD             = jjnr[jidx+3];
 173             j_coord_offsetA  = DIM*jnrA;
 174             j_coord_offsetB  = DIM*jnrB;
 175             j_coord_offsetC  = DIM*jnrC;
 176             j_coord_offsetD  = DIM*jnrD;
 177
 178             /* load j atom coordinates */
 179             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 180                                               x+j_coord_offsetC,x+j_coord_offsetD,
 181                                               &jx0,&jy0,&jz0);
 182
 183             /* Calculate displacement vector */
 184             dx00             = _mm_sub_ps(ix0,jx0);
 185             dy00             = _mm_sub_ps(iy0,jy0);
 186             dz00             = _mm_sub_ps(iz0,jz0);
 187
 188             /* Calculate squared distance and things based on it */
 189             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 190
 191             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 192
 193             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 194
 195             /* Load parameters for j particles */
 196             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 197                                                               charge+jnrC+0,charge+jnrD+0);
 198             vdwjidx0A        = 2*vdwtype[jnrA+0];
 199             vdwjidx0B        = 2*vdwtype[jnrB+0];
 200             vdwjidx0C        = 2*vdwtype[jnrC+0];
 201             vdwjidx0D        = 2*vdwtype[jnrD+0];
 202
 203             /**************************
 204              * CALCULATE INTERACTIONS *
 205              **************************/
 206
 207             /* Compute parameters for interactions between i and j atoms */
 208             qq00             = _mm_mul_ps(iq0,jq0);
 209             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 210                                          vdwparam+vdwioffset0+vdwjidx0B,
 211                                          vdwparam+vdwioffset0+vdwjidx0C,
 212                                          vdwparam+vdwioffset0+vdwjidx0D,
 213                                          &c6_00,&c12_00);
 214
 215             /* REACTION-FIELD ELECTROSTATICS */
 216             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 217             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 218
 219             /* LENNARD-JONES DISPERSION/REPULSION */
 220
 221             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 222             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 223             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 224             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 225             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 226
 227             /* Update potential sum for this i atom from the interaction with this j atom. */
 228             velecsum         = _mm_add_ps(velecsum,velec);
 229             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 230
 231             fscal            = _mm_add_ps(felec,fvdw);
 232
 233              /* Update vectorial force */
 234             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 235             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 236             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 237
 238             fjptrA             = f+j_coord_offsetA;
 239             fjptrB             = f+j_coord_offsetB;
 240             fjptrC             = f+j_coord_offsetC;
 241             fjptrD             = f+j_coord_offsetD;
 242             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 243                                                    _mm_mul_ps(dx00,fscal),
 244                                                    _mm_mul_ps(dy00,fscal),
 245                                                    _mm_mul_ps(dz00,fscal));
 246
 247             /* Inner loop uses 47 flops */
 248         }
 249
 250         if(jidx<j_index_end)
 251         {
 252
 253             /* Get j neighbor index, and coordinate index */
 254             jnrlistA         = jjnr[jidx];
 255             jnrlistB         = jjnr[jidx+1];
 256             jnrlistC         = jjnr[jidx+2];
 257             jnrlistD         = jjnr[jidx+3];
 258             /* Sign of each element will be negative for non-real atoms.
 259              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 260              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 261              */
 262             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 263             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 264             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 265             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 266             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 267             j_coord_offsetA  = DIM*jnrA;
 268             j_coord_offsetB  = DIM*jnrB;
 269             j_coord_offsetC  = DIM*jnrC;
 270             j_coord_offsetD  = DIM*jnrD;
 271
 272             /* load j atom coordinates */
 273             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 274                                               x+j_coord_offsetC,x+j_coord_offsetD,
 275                                               &jx0,&jy0,&jz0);
 276
 277             /* Calculate displacement vector */
 278             dx00             = _mm_sub_ps(ix0,jx0);
 279             dy00             = _mm_sub_ps(iy0,jy0);
 280             dz00             = _mm_sub_ps(iz0,jz0);
 281
 282             /* Calculate squared distance and things based on it */
 283             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 284
 285             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 286
 287             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 288
 289             /* Load parameters for j particles */
 290             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 291                                                               charge+jnrC+0,charge+jnrD+0);
 292             vdwjidx0A        = 2*vdwtype[jnrA+0];
 293             vdwjidx0B        = 2*vdwtype[jnrB+0];
 294             vdwjidx0C        = 2*vdwtype[jnrC+0];
 295             vdwjidx0D        = 2*vdwtype[jnrD+0];
 296
 297             /**************************
 298              * CALCULATE INTERACTIONS *
 299              **************************/
 300
 301             /* Compute parameters for interactions between i and j atoms */
 302             qq00             = _mm_mul_ps(iq0,jq0);
 303             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 304                                          vdwparam+vdwioffset0+vdwjidx0B,
 305                                          vdwparam+vdwioffset0+vdwjidx0C,
 306                                          vdwparam+vdwioffset0+vdwjidx0D,
 307                                          &c6_00,&c12_00);
 308
 309             /* REACTION-FIELD ELECTROSTATICS */
 310             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 311             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 312
 313             /* LENNARD-JONES DISPERSION/REPULSION */
 314
 315             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 316             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 317             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 318             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 319             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 320
 321             /* Update potential sum for this i atom from the interaction with this j atom. */
 322             velec            = _mm_andnot_ps(dummy_mask,velec);
 323             velecsum         = _mm_add_ps(velecsum,velec);
 324             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 325             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 326
 327             fscal            = _mm_add_ps(felec,fvdw);
 328
 329             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 330
 331              /* Update vectorial force */
 332             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 333             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 334             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 335
 336             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 337             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 338             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 339             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 340             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 341                                                    _mm_mul_ps(dx00,fscal),
 342                                                    _mm_mul_ps(dy00,fscal),
 343                                                    _mm_mul_ps(dz00,fscal));
 344
 345             /* Inner loop uses 47 flops */
 346         }
 347
 348         /* End of innermost loop */
 349
 350         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 351                                               f+i_coord_offset,fshift+i_shift_offset);
 352
 353         ggid                        = gid[iidx];
 354         /* Update potential energies */
 355         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 356         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 357
 358         /* Increment number of inner iterations */
 359         inneriter                  += j_index_end - j_index_start;
 360
 361         /* Outer loop uses 9 flops */
 362     }
 363
 364     /* Increment number of outer iterations */
 365     outeriter        += nri;
 366
 367     /* Update outer/inner flops */
 368
 369     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*47);
 370 }
 371 /*
 372  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_avx_128_fma_single
 373  * Electrostatics interaction: ReactionField
 374  * VdW interaction:            LennardJones
 375  * Geometry:                   Particle-Particle
 376  * Calculate force/pot:        Force
 377  */
 378 void
 379 nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_avx_128_fma_single
 380                     (t_nblist                    * gmx_restrict       nlist,
 381                      rvec                        * gmx_restrict          xx,
 382                      rvec                        * gmx_restrict          ff,
 383                      t_forcerec                  * gmx_restrict          fr,
 384                      t_mdatoms                   * gmx_restrict     mdatoms,
 385                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 386                      t_nrnb                      * gmx_restrict        nrnb)
 387 {
 388     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 389      * just 0 for non-waters.
 390      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 391      * jnr indices corresponding to data put in the four positions in the SIMD register.
 392      */
 393     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 394     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 395     int              jnrA,jnrB,jnrC,jnrD;
 396     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 397     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 398     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 399     real             rcutoff_scalar;
 400     real             *shiftvec,*fshift,*x,*f;
 401     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 402     real             scratch[4*DIM];
 403     __m128           fscal,rcutoff,rcutoff2,jidxall;
 404     int              vdwioffset0;
 405     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 406     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 407     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 408     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 409     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 410     real             *charge;
 411     int              nvdwtype;
 412     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 413     int              *vdwtype;
 414     real             *vdwparam;
 415     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 416     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 417     __m128           dummy_mask,cutoff_mask;
 418     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 419     __m128           one     = _mm_set1_ps(1.0);
 420     __m128           two     = _mm_set1_ps(2.0);
 421     x                = xx[0];
 422     f                = ff[0];
 423
 424     nri              = nlist->nri;
 425     iinr             = nlist->iinr;
 426     jindex           = nlist->jindex;
 427     jjnr             = nlist->jjnr;
 428     shiftidx         = nlist->shift;
 429     gid              = nlist->gid;
 430     shiftvec         = fr->shift_vec[0];
 431     fshift           = fr->fshift[0];
 432     facel            = _mm_set1_ps(fr->epsfac);
 433     charge           = mdatoms->chargeA;
 434     krf              = _mm_set1_ps(fr->ic->k_rf);
 435     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 436     crf              = _mm_set1_ps(fr->ic->c_rf);
 437     nvdwtype         = fr->ntype;
 438     vdwparam         = fr->nbfp;
 439     vdwtype          = mdatoms->typeA;
 440
 441     /* Avoid stupid compiler warnings */
 442     jnrA = jnrB = jnrC = jnrD = 0;
 443     j_coord_offsetA = 0;
 444     j_coord_offsetB = 0;
 445     j_coord_offsetC = 0;
 446     j_coord_offsetD = 0;
 447
 448     outeriter        = 0;
 449     inneriter        = 0;
 450
 451     for(iidx=0;iidx<4*DIM;iidx++)
 452     {
 453         scratch[iidx] = 0.0;
 454     }
 455
 456     /* Start outer loop over neighborlists */
 457     for(iidx=0; iidx<nri; iidx++)
 458     {
 459         /* Load shift vector for this list */
 460         i_shift_offset   = DIM*shiftidx[iidx];
 461
 462         /* Load limits for loop over neighbors */
 463         j_index_start    = jindex[iidx];
 464         j_index_end      = jindex[iidx+1];
 465
 466         /* Get outer coordinate index */
 467         inr              = iinr[iidx];
 468         i_coord_offset   = DIM*inr;
 469
 470         /* Load i particle coords and add shift vector */
 471         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 472
 473         fix0             = _mm_setzero_ps();
 474         fiy0             = _mm_setzero_ps();
 475         fiz0             = _mm_setzero_ps();
 476
 477         /* Load parameters for i particles */
 478         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 479         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 480
 481         /* Start inner kernel loop */
 482         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 483         {
 484
 485             /* Get j neighbor index, and coordinate index */
 486             jnrA             = jjnr[jidx];
 487             jnrB             = jjnr[jidx+1];
 488             jnrC             = jjnr[jidx+2];
 489             jnrD             = jjnr[jidx+3];
 490             j_coord_offsetA  = DIM*jnrA;
 491             j_coord_offsetB  = DIM*jnrB;
 492             j_coord_offsetC  = DIM*jnrC;
 493             j_coord_offsetD  = DIM*jnrD;
 494
 495             /* load j atom coordinates */
 496             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 497                                               x+j_coord_offsetC,x+j_coord_offsetD,
 498                                               &jx0,&jy0,&jz0);
 499
 500             /* Calculate displacement vector */
 501             dx00             = _mm_sub_ps(ix0,jx0);
 502             dy00             = _mm_sub_ps(iy0,jy0);
 503             dz00             = _mm_sub_ps(iz0,jz0);
 504
 505             /* Calculate squared distance and things based on it */
 506             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 507
 508             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 509
 510             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 511
 512             /* Load parameters for j particles */
 513             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 514                                                               charge+jnrC+0,charge+jnrD+0);
 515             vdwjidx0A        = 2*vdwtype[jnrA+0];
 516             vdwjidx0B        = 2*vdwtype[jnrB+0];
 517             vdwjidx0C        = 2*vdwtype[jnrC+0];
 518             vdwjidx0D        = 2*vdwtype[jnrD+0];
 519
 520             /**************************
 521              * CALCULATE INTERACTIONS *
 522              **************************/
 523
 524             /* Compute parameters for interactions between i and j atoms */
 525             qq00             = _mm_mul_ps(iq0,jq0);
 526             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 527                                          vdwparam+vdwioffset0+vdwjidx0B,
 528                                          vdwparam+vdwioffset0+vdwjidx0C,
 529                                          vdwparam+vdwioffset0+vdwjidx0D,
 530                                          &c6_00,&c12_00);
 531
 532             /* REACTION-FIELD ELECTROSTATICS */
 533             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 534
 535             /* LENNARD-JONES DISPERSION/REPULSION */
 536
 537             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 538             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 539
 540             fscal            = _mm_add_ps(felec,fvdw);
 541
 542              /* Update vectorial force */
 543             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 544             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 545             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 546
 547             fjptrA             = f+j_coord_offsetA;
 548             fjptrB             = f+j_coord_offsetB;
 549             fjptrC             = f+j_coord_offsetC;
 550             fjptrD             = f+j_coord_offsetD;
 551             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 552                                                    _mm_mul_ps(dx00,fscal),
 553                                                    _mm_mul_ps(dy00,fscal),
 554                                                    _mm_mul_ps(dz00,fscal));
 555
 556             /* Inner loop uses 37 flops */
 557         }
 558
 559         if(jidx<j_index_end)
 560         {
 561
 562             /* Get j neighbor index, and coordinate index */
 563             jnrlistA         = jjnr[jidx];
 564             jnrlistB         = jjnr[jidx+1];
 565             jnrlistC         = jjnr[jidx+2];
 566             jnrlistD         = jjnr[jidx+3];
 567             /* Sign of each element will be negative for non-real atoms.
 568              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 569              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 570              */
 571             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 572             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 573             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 574             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 575             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 576             j_coord_offsetA  = DIM*jnrA;
 577             j_coord_offsetB  = DIM*jnrB;
 578             j_coord_offsetC  = DIM*jnrC;
 579             j_coord_offsetD  = DIM*jnrD;
 580
 581             /* load j atom coordinates */
 582             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 583                                               x+j_coord_offsetC,x+j_coord_offsetD,
 584                                               &jx0,&jy0,&jz0);
 585
 586             /* Calculate displacement vector */
 587             dx00             = _mm_sub_ps(ix0,jx0);
 588             dy00             = _mm_sub_ps(iy0,jy0);
 589             dz00             = _mm_sub_ps(iz0,jz0);
 590
 591             /* Calculate squared distance and things based on it */
 592             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 593
 594             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 595
 596             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 597
 598             /* Load parameters for j particles */
 599             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 600                                                               charge+jnrC+0,charge+jnrD+0);
 601             vdwjidx0A        = 2*vdwtype[jnrA+0];
 602             vdwjidx0B        = 2*vdwtype[jnrB+0];
 603             vdwjidx0C        = 2*vdwtype[jnrC+0];
 604             vdwjidx0D        = 2*vdwtype[jnrD+0];
 605
 606             /**************************
 607              * CALCULATE INTERACTIONS *
 608              **************************/
 609
 610             /* Compute parameters for interactions between i and j atoms */
 611             qq00             = _mm_mul_ps(iq0,jq0);
 612             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 613                                          vdwparam+vdwioffset0+vdwjidx0B,
 614                                          vdwparam+vdwioffset0+vdwjidx0C,
 615                                          vdwparam+vdwioffset0+vdwjidx0D,
 616                                          &c6_00,&c12_00);
 617
 618             /* REACTION-FIELD ELECTROSTATICS */
 619             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 620
 621             /* LENNARD-JONES DISPERSION/REPULSION */
 622
 623             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 624             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 625
 626             fscal            = _mm_add_ps(felec,fvdw);
 627
 628             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 629
 630              /* Update vectorial force */
 631             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 632             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 633             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 634
 635             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 636             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 637             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 638             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 639             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 640                                                    _mm_mul_ps(dx00,fscal),
 641                                                    _mm_mul_ps(dy00,fscal),
 642                                                    _mm_mul_ps(dz00,fscal));
 643
 644             /* Inner loop uses 37 flops */
 645         }
 646
 647         /* End of innermost loop */
 648
 649         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 650                                               f+i_coord_offset,fshift+i_shift_offset);
 651
 652         /* Increment number of inner iterations */
 653         inneriter                  += j_index_end - j_index_start;
 654
 655         /* Outer loop uses 7 flops */
 656     }
 657
 658     /* Increment number of outer iterations */
 659     outeriter        += nri;
 660
 661     /* Update outer/inner flops */
 662
 663     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*37);
 664 }