src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecNone_VdwLJSh_GeomP1P1_sse4_1_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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 sse4_1_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/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse4_1_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSh_GeomP1P1_VF_sse4_1_single
  51  * Electrostatics interaction: None
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecNone_VdwLJSh_GeomP1P1_VF_sse4_1_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  85     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     int              nvdwtype;
  88     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  89     int              *vdwtype;
  90     real             *vdwparam;
  91     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  92     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  93     __m128           dummy_mask,cutoff_mask;
  94     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  95     __m128           one     = _mm_set1_ps(1.0);
  96     __m128           two     = _mm_set1_ps(2.0);
  97     x                = xx[0];
  98     f                = ff[0];
  99
 100     nri              = nlist->nri;
 101     iinr             = nlist->iinr;
 102     jindex           = nlist->jindex;
 103     jjnr             = nlist->jjnr;
 104     shiftidx         = nlist->shift;
 105     gid              = nlist->gid;
 106     shiftvec         = fr->shift_vec[0];
 107     fshift           = fr->fshift[0];
 108     nvdwtype         = fr->ntype;
 109     vdwparam         = fr->nbfp;
 110     vdwtype          = mdatoms->typeA;
 111
 112     rcutoff_scalar   = fr->ic->rvdw;
 113     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 114     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 115
 116     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 117     rvdw             = _mm_set1_ps(fr->ic->rvdw);
 118
 119     /* Avoid stupid compiler warnings */
 120     jnrA = jnrB = jnrC = jnrD = 0;
 121     j_coord_offsetA = 0;
 122     j_coord_offsetB = 0;
 123     j_coord_offsetC = 0;
 124     j_coord_offsetD = 0;
 125
 126     outeriter        = 0;
 127     inneriter        = 0;
 128
 129     for(iidx=0;iidx<4*DIM;iidx++)
 130     {
 131         scratch[iidx] = 0.0;
 132     }
 133
 134     /* Start outer loop over neighborlists */
 135     for(iidx=0; iidx<nri; iidx++)
 136     {
 137         /* Load shift vector for this list */
 138         i_shift_offset   = DIM*shiftidx[iidx];
 139
 140         /* Load limits for loop over neighbors */
 141         j_index_start    = jindex[iidx];
 142         j_index_end      = jindex[iidx+1];
 143
 144         /* Get outer coordinate index */
 145         inr              = iinr[iidx];
 146         i_coord_offset   = DIM*inr;
 147
 148         /* Load i particle coords and add shift vector */
 149         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 150
 151         fix0             = _mm_setzero_ps();
 152         fiy0             = _mm_setzero_ps();
 153         fiz0             = _mm_setzero_ps();
 154
 155         /* Load parameters for i particles */
 156         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 157
 158         /* Reset potential sums */
 159         vvdwsum          = _mm_setzero_ps();
 160
 161         /* Start inner kernel loop */
 162         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 163         {
 164
 165             /* Get j neighbor index, and coordinate index */
 166             jnrA             = jjnr[jidx];
 167             jnrB             = jjnr[jidx+1];
 168             jnrC             = jjnr[jidx+2];
 169             jnrD             = jjnr[jidx+3];
 170             j_coord_offsetA  = DIM*jnrA;
 171             j_coord_offsetB  = DIM*jnrB;
 172             j_coord_offsetC  = DIM*jnrC;
 173             j_coord_offsetD  = DIM*jnrD;
 174
 175             /* load j atom coordinates */
 176             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 177                                               x+j_coord_offsetC,x+j_coord_offsetD,
 178                                               &jx0,&jy0,&jz0);
 179
 180             /* Calculate displacement vector */
 181             dx00             = _mm_sub_ps(ix0,jx0);
 182             dy00             = _mm_sub_ps(iy0,jy0);
 183             dz00             = _mm_sub_ps(iz0,jz0);
 184
 185             /* Calculate squared distance and things based on it */
 186             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 187
 188             rinvsq00         = sse41_inv_f(rsq00);
 189
 190             /* Load parameters for j particles */
 191             vdwjidx0A        = 2*vdwtype[jnrA+0];
 192             vdwjidx0B        = 2*vdwtype[jnrB+0];
 193             vdwjidx0C        = 2*vdwtype[jnrC+0];
 194             vdwjidx0D        = 2*vdwtype[jnrD+0];
 195
 196             /**************************
 197              * CALCULATE INTERACTIONS *
 198              **************************/
 199
 200             if (gmx_mm_any_lt(rsq00,rcutoff2))
 201             {
 202
 203             /* Compute parameters for interactions between i and j atoms */
 204             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 205                                          vdwparam+vdwioffset0+vdwjidx0B,
 206                                          vdwparam+vdwioffset0+vdwjidx0C,
 207                                          vdwparam+vdwioffset0+vdwjidx0D,
 208                                          &c6_00,&c12_00);
 209
 210             /* LENNARD-JONES DISPERSION/REPULSION */
 211
 212             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 213             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 214             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 215             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 216                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 217             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 218
 219             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 220
 221             /* Update potential sum for this i atom from the interaction with this j atom. */
 222             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 223             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 224
 225             fscal            = fvdw;
 226
 227             fscal            = _mm_and_ps(fscal,cutoff_mask);
 228
 229             /* Calculate temporary vectorial force */
 230             tx               = _mm_mul_ps(fscal,dx00);
 231             ty               = _mm_mul_ps(fscal,dy00);
 232             tz               = _mm_mul_ps(fscal,dz00);
 233
 234             /* Update vectorial force */
 235             fix0             = _mm_add_ps(fix0,tx);
 236             fiy0             = _mm_add_ps(fiy0,ty);
 237             fiz0             = _mm_add_ps(fiz0,tz);
 238
 239             fjptrA             = f+j_coord_offsetA;
 240             fjptrB             = f+j_coord_offsetB;
 241             fjptrC             = f+j_coord_offsetC;
 242             fjptrD             = f+j_coord_offsetD;
 243             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 244
 245             }
 246
 247             /* Inner loop uses 41 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             rinvsq00         = sse41_inv_f(rsq00);
 286
 287             /* Load parameters for j particles */
 288             vdwjidx0A        = 2*vdwtype[jnrA+0];
 289             vdwjidx0B        = 2*vdwtype[jnrB+0];
 290             vdwjidx0C        = 2*vdwtype[jnrC+0];
 291             vdwjidx0D        = 2*vdwtype[jnrD+0];
 292
 293             /**************************
 294              * CALCULATE INTERACTIONS *
 295              **************************/
 296
 297             if (gmx_mm_any_lt(rsq00,rcutoff2))
 298             {
 299
 300             /* Compute parameters for interactions between i and j atoms */
 301             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 302                                          vdwparam+vdwioffset0+vdwjidx0B,
 303                                          vdwparam+vdwioffset0+vdwjidx0C,
 304                                          vdwparam+vdwioffset0+vdwjidx0D,
 305                                          &c6_00,&c12_00);
 306
 307             /* LENNARD-JONES DISPERSION/REPULSION */
 308
 309             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 310             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 311             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 312             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 313                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 314             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 315
 316             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 317
 318             /* Update potential sum for this i atom from the interaction with this j atom. */
 319             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 320             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 321             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 322
 323             fscal            = fvdw;
 324
 325             fscal            = _mm_and_ps(fscal,cutoff_mask);
 326
 327             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 328
 329             /* Calculate temporary vectorial force */
 330             tx               = _mm_mul_ps(fscal,dx00);
 331             ty               = _mm_mul_ps(fscal,dy00);
 332             tz               = _mm_mul_ps(fscal,dz00);
 333
 334             /* Update vectorial force */
 335             fix0             = _mm_add_ps(fix0,tx);
 336             fiy0             = _mm_add_ps(fiy0,ty);
 337             fiz0             = _mm_add_ps(fiz0,tz);
 338
 339             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 340             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 341             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 342             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 343             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 344
 345             }
 346
 347             /* Inner loop uses 41 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(vvdwsum,kernel_data->energygrp_vdw+ggid);
 358
 359         /* Increment number of inner iterations */
 360         inneriter                  += j_index_end - j_index_start;
 361
 362         /* Outer loop uses 7 flops */
 363     }
 364
 365     /* Increment number of outer iterations */
 366     outeriter        += nri;
 367
 368     /* Update outer/inner flops */
 369
 370     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*41);
 371 }
 372 /*
 373  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSh_GeomP1P1_F_sse4_1_single
 374  * Electrostatics interaction: None
 375  * VdW interaction:            LennardJones
 376  * Geometry:                   Particle-Particle
 377  * Calculate force/pot:        Force
 378  */
 379 void
 380 nb_kernel_ElecNone_VdwLJSh_GeomP1P1_F_sse4_1_single
 381                     (t_nblist                    * gmx_restrict       nlist,
 382                      rvec                        * gmx_restrict          xx,
 383                      rvec                        * gmx_restrict          ff,
 384                      struct t_forcerec           * gmx_restrict          fr,
 385                      t_mdatoms                   * gmx_restrict     mdatoms,
 386                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 387                      t_nrnb                      * gmx_restrict        nrnb)
 388 {
 389     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 390      * just 0 for non-waters.
 391      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 392      * jnr indices corresponding to data put in the four positions in the SIMD register.
 393      */
 394     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 395     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 396     int              jnrA,jnrB,jnrC,jnrD;
 397     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 398     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 399     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 400     real             rcutoff_scalar;
 401     real             *shiftvec,*fshift,*x,*f;
 402     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 403     real             scratch[4*DIM];
 404     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 405     int              vdwioffset0;
 406     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 407     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 408     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 409     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 410     int              nvdwtype;
 411     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 412     int              *vdwtype;
 413     real             *vdwparam;
 414     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 415     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 416     __m128           dummy_mask,cutoff_mask;
 417     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 418     __m128           one     = _mm_set1_ps(1.0);
 419     __m128           two     = _mm_set1_ps(2.0);
 420     x                = xx[0];
 421     f                = ff[0];
 422
 423     nri              = nlist->nri;
 424     iinr             = nlist->iinr;
 425     jindex           = nlist->jindex;
 426     jjnr             = nlist->jjnr;
 427     shiftidx         = nlist->shift;
 428     gid              = nlist->gid;
 429     shiftvec         = fr->shift_vec[0];
 430     fshift           = fr->fshift[0];
 431     nvdwtype         = fr->ntype;
 432     vdwparam         = fr->nbfp;
 433     vdwtype          = mdatoms->typeA;
 434
 435     rcutoff_scalar   = fr->ic->rvdw;
 436     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 437     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 438
 439     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 440     rvdw             = _mm_set1_ps(fr->ic->rvdw);
 441
 442     /* Avoid stupid compiler warnings */
 443     jnrA = jnrB = jnrC = jnrD = 0;
 444     j_coord_offsetA = 0;
 445     j_coord_offsetB = 0;
 446     j_coord_offsetC = 0;
 447     j_coord_offsetD = 0;
 448
 449     outeriter        = 0;
 450     inneriter        = 0;
 451
 452     for(iidx=0;iidx<4*DIM;iidx++)
 453     {
 454         scratch[iidx] = 0.0;
 455     }
 456
 457     /* Start outer loop over neighborlists */
 458     for(iidx=0; iidx<nri; iidx++)
 459     {
 460         /* Load shift vector for this list */
 461         i_shift_offset   = DIM*shiftidx[iidx];
 462
 463         /* Load limits for loop over neighbors */
 464         j_index_start    = jindex[iidx];
 465         j_index_end      = jindex[iidx+1];
 466
 467         /* Get outer coordinate index */
 468         inr              = iinr[iidx];
 469         i_coord_offset   = DIM*inr;
 470
 471         /* Load i particle coords and add shift vector */
 472         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 473
 474         fix0             = _mm_setzero_ps();
 475         fiy0             = _mm_setzero_ps();
 476         fiz0             = _mm_setzero_ps();
 477
 478         /* Load parameters for i particles */
 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             rinvsq00         = sse41_inv_f(rsq00);
 509
 510             /* Load parameters for j particles */
 511             vdwjidx0A        = 2*vdwtype[jnrA+0];
 512             vdwjidx0B        = 2*vdwtype[jnrB+0];
 513             vdwjidx0C        = 2*vdwtype[jnrC+0];
 514             vdwjidx0D        = 2*vdwtype[jnrD+0];
 515
 516             /**************************
 517              * CALCULATE INTERACTIONS *
 518              **************************/
 519
 520             if (gmx_mm_any_lt(rsq00,rcutoff2))
 521             {
 522
 523             /* Compute parameters for interactions between i and j atoms */
 524             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 525                                          vdwparam+vdwioffset0+vdwjidx0B,
 526                                          vdwparam+vdwioffset0+vdwjidx0C,
 527                                          vdwparam+vdwioffset0+vdwjidx0D,
 528                                          &c6_00,&c12_00);
 529
 530             /* LENNARD-JONES DISPERSION/REPULSION */
 531
 532             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 533             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 534
 535             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 536
 537             fscal            = fvdw;
 538
 539             fscal            = _mm_and_ps(fscal,cutoff_mask);
 540
 541             /* Calculate temporary vectorial force */
 542             tx               = _mm_mul_ps(fscal,dx00);
 543             ty               = _mm_mul_ps(fscal,dy00);
 544             tz               = _mm_mul_ps(fscal,dz00);
 545
 546             /* Update vectorial force */
 547             fix0             = _mm_add_ps(fix0,tx);
 548             fiy0             = _mm_add_ps(fiy0,ty);
 549             fiz0             = _mm_add_ps(fiz0,tz);
 550
 551             fjptrA             = f+j_coord_offsetA;
 552             fjptrB             = f+j_coord_offsetB;
 553             fjptrC             = f+j_coord_offsetC;
 554             fjptrD             = f+j_coord_offsetD;
 555             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 556
 557             }
 558
 559             /* Inner loop uses 30 flops */
 560         }
 561
 562         if(jidx<j_index_end)
 563         {
 564
 565             /* Get j neighbor index, and coordinate index */
 566             jnrlistA         = jjnr[jidx];
 567             jnrlistB         = jjnr[jidx+1];
 568             jnrlistC         = jjnr[jidx+2];
 569             jnrlistD         = jjnr[jidx+3];
 570             /* Sign of each element will be negative for non-real atoms.
 571              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 572              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 573              */
 574             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 575             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 576             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 577             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 578             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 579             j_coord_offsetA  = DIM*jnrA;
 580             j_coord_offsetB  = DIM*jnrB;
 581             j_coord_offsetC  = DIM*jnrC;
 582             j_coord_offsetD  = DIM*jnrD;
 583
 584             /* load j atom coordinates */
 585             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 586                                               x+j_coord_offsetC,x+j_coord_offsetD,
 587                                               &jx0,&jy0,&jz0);
 588
 589             /* Calculate displacement vector */
 590             dx00             = _mm_sub_ps(ix0,jx0);
 591             dy00             = _mm_sub_ps(iy0,jy0);
 592             dz00             = _mm_sub_ps(iz0,jz0);
 593
 594             /* Calculate squared distance and things based on it */
 595             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 596
 597             rinvsq00         = sse41_inv_f(rsq00);
 598
 599             /* Load parameters for j particles */
 600             vdwjidx0A        = 2*vdwtype[jnrA+0];
 601             vdwjidx0B        = 2*vdwtype[jnrB+0];
 602             vdwjidx0C        = 2*vdwtype[jnrC+0];
 603             vdwjidx0D        = 2*vdwtype[jnrD+0];
 604
 605             /**************************
 606              * CALCULATE INTERACTIONS *
 607              **************************/
 608
 609             if (gmx_mm_any_lt(rsq00,rcutoff2))
 610             {
 611
 612             /* Compute parameters for interactions between i and j atoms */
 613             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 614                                          vdwparam+vdwioffset0+vdwjidx0B,
 615                                          vdwparam+vdwioffset0+vdwjidx0C,
 616                                          vdwparam+vdwioffset0+vdwjidx0D,
 617                                          &c6_00,&c12_00);
 618
 619             /* LENNARD-JONES DISPERSION/REPULSION */
 620
 621             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 622             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 623
 624             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 625
 626             fscal            = fvdw;
 627
 628             fscal            = _mm_and_ps(fscal,cutoff_mask);
 629
 630             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 631
 632             /* Calculate temporary vectorial force */
 633             tx               = _mm_mul_ps(fscal,dx00);
 634             ty               = _mm_mul_ps(fscal,dy00);
 635             tz               = _mm_mul_ps(fscal,dz00);
 636
 637             /* Update vectorial force */
 638             fix0             = _mm_add_ps(fix0,tx);
 639             fiy0             = _mm_add_ps(fiy0,ty);
 640             fiz0             = _mm_add_ps(fiz0,tz);
 641
 642             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 643             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 644             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 645             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 646             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 647
 648             }
 649
 650             /* Inner loop uses 30 flops */
 651         }
 652
 653         /* End of innermost loop */
 654
 655         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 656                                               f+i_coord_offset,fshift+i_shift_offset);
 657
 658         /* Increment number of inner iterations */
 659         inneriter                  += j_index_end - j_index_start;
 660
 661         /* Outer loop uses 6 flops */
 662     }
 663
 664     /* Increment number of outer iterations */
 665     outeriter        += nri;
 666
 667     /* Update outer/inner flops */
 668
 669     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*30);
 670 }