src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecNone_VdwLJSh_GeomP1P1_sse2_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 sse2_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_single.h"
  50 #include "kernelutil_x86_sse2_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSh_GeomP1P1_VF_sse2_single
  54  * Electrostatics interaction: None
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecNone_VdwLJSh_GeomP1P1_VF_sse2_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  88     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     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     nvdwtype         = fr->ntype;
 112     vdwparam         = fr->nbfp;
 113     vdwtype          = mdatoms->typeA;
 114
 115     rcutoff_scalar   = fr->rvdw;
 116     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 117     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 118
 119     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 120     rvdw             = _mm_set1_ps(fr->rvdw);
 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         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 160
 161         /* Reset potential sums */
 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             rinvsq00         = gmx_mm_inv_ps(rsq00);
 192
 193             /* Load parameters for j particles */
 194             vdwjidx0A        = 2*vdwtype[jnrA+0];
 195             vdwjidx0B        = 2*vdwtype[jnrB+0];
 196             vdwjidx0C        = 2*vdwtype[jnrC+0];
 197             vdwjidx0D        = 2*vdwtype[jnrD+0];
 198
 199             /**************************
 200              * CALCULATE INTERACTIONS *
 201              **************************/
 202
 203             if (gmx_mm_any_lt(rsq00,rcutoff2))
 204             {
 205
 206             /* Compute parameters for interactions between i and j atoms */
 207             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 208                                          vdwparam+vdwioffset0+vdwjidx0B,
 209                                          vdwparam+vdwioffset0+vdwjidx0C,
 210                                          vdwparam+vdwioffset0+vdwjidx0D,
 211                                          &c6_00,&c12_00);
 212
 213             /* LENNARD-JONES DISPERSION/REPULSION */
 214
 215             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 216             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 217             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 218             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) ,
 219                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 220             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 221
 222             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 223
 224             /* Update potential sum for this i atom from the interaction with this j atom. */
 225             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 226             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 227
 228             fscal            = fvdw;
 229
 230             fscal            = _mm_and_ps(fscal,cutoff_mask);
 231
 232             /* Calculate temporary vectorial force */
 233             tx               = _mm_mul_ps(fscal,dx00);
 234             ty               = _mm_mul_ps(fscal,dy00);
 235             tz               = _mm_mul_ps(fscal,dz00);
 236
 237             /* Update vectorial force */
 238             fix0             = _mm_add_ps(fix0,tx);
 239             fiy0             = _mm_add_ps(fiy0,ty);
 240             fiz0             = _mm_add_ps(fiz0,tz);
 241
 242             fjptrA             = f+j_coord_offsetA;
 243             fjptrB             = f+j_coord_offsetB;
 244             fjptrC             = f+j_coord_offsetC;
 245             fjptrD             = f+j_coord_offsetD;
 246             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 247
 248             }
 249
 250             /* Inner loop uses 41 flops */
 251         }
 252
 253         if(jidx<j_index_end)
 254         {
 255
 256             /* Get j neighbor index, and coordinate index */
 257             jnrlistA         = jjnr[jidx];
 258             jnrlistB         = jjnr[jidx+1];
 259             jnrlistC         = jjnr[jidx+2];
 260             jnrlistD         = jjnr[jidx+3];
 261             /* Sign of each element will be negative for non-real atoms.
 262              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 263              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 264              */
 265             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 266             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 267             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 268             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 269             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 270             j_coord_offsetA  = DIM*jnrA;
 271             j_coord_offsetB  = DIM*jnrB;
 272             j_coord_offsetC  = DIM*jnrC;
 273             j_coord_offsetD  = DIM*jnrD;
 274
 275             /* load j atom coordinates */
 276             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 277                                               x+j_coord_offsetC,x+j_coord_offsetD,
 278                                               &jx0,&jy0,&jz0);
 279
 280             /* Calculate displacement vector */
 281             dx00             = _mm_sub_ps(ix0,jx0);
 282             dy00             = _mm_sub_ps(iy0,jy0);
 283             dz00             = _mm_sub_ps(iz0,jz0);
 284
 285             /* Calculate squared distance and things based on it */
 286             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 287
 288             rinvsq00         = gmx_mm_inv_ps(rsq00);
 289
 290             /* Load parameters for j particles */
 291             vdwjidx0A        = 2*vdwtype[jnrA+0];
 292             vdwjidx0B        = 2*vdwtype[jnrB+0];
 293             vdwjidx0C        = 2*vdwtype[jnrC+0];
 294             vdwjidx0D        = 2*vdwtype[jnrD+0];
 295
 296             /**************************
 297              * CALCULATE INTERACTIONS *
 298              **************************/
 299
 300             if (gmx_mm_any_lt(rsq00,rcutoff2))
 301             {
 302
 303             /* Compute parameters for interactions between i and j atoms */
 304             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 305                                          vdwparam+vdwioffset0+vdwjidx0B,
 306                                          vdwparam+vdwioffset0+vdwjidx0C,
 307                                          vdwparam+vdwioffset0+vdwjidx0D,
 308                                          &c6_00,&c12_00);
 309
 310             /* LENNARD-JONES DISPERSION/REPULSION */
 311
 312             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 313             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 314             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 315             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) ,
 316                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 317             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 318
 319             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 320
 321             /* Update potential sum for this i atom from the interaction with this j atom. */
 322             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 323             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 324             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 325
 326             fscal            = fvdw;
 327
 328             fscal            = _mm_and_ps(fscal,cutoff_mask);
 329
 330             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 331
 332             /* Calculate temporary vectorial force */
 333             tx               = _mm_mul_ps(fscal,dx00);
 334             ty               = _mm_mul_ps(fscal,dy00);
 335             tz               = _mm_mul_ps(fscal,dz00);
 336
 337             /* Update vectorial force */
 338             fix0             = _mm_add_ps(fix0,tx);
 339             fiy0             = _mm_add_ps(fiy0,ty);
 340             fiz0             = _mm_add_ps(fiz0,tz);
 341
 342             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 343             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 344             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 345             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 346             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 347
 348             }
 349
 350             /* Inner loop uses 41 flops */
 351         }
 352
 353         /* End of innermost loop */
 354
 355         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 356                                               f+i_coord_offset,fshift+i_shift_offset);
 357
 358         ggid                        = gid[iidx];
 359         /* Update potential energies */
 360         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 361
 362         /* Increment number of inner iterations */
 363         inneriter                  += j_index_end - j_index_start;
 364
 365         /* Outer loop uses 7 flops */
 366     }
 367
 368     /* Increment number of outer iterations */
 369     outeriter        += nri;
 370
 371     /* Update outer/inner flops */
 372
 373     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*41);
 374 }
 375 /*
 376  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSh_GeomP1P1_F_sse2_single
 377  * Electrostatics interaction: None
 378  * VdW interaction:            LennardJones
 379  * Geometry:                   Particle-Particle
 380  * Calculate force/pot:        Force
 381  */
 382 void
 383 nb_kernel_ElecNone_VdwLJSh_GeomP1P1_F_sse2_single
 384                     (t_nblist                    * gmx_restrict       nlist,
 385                      rvec                        * gmx_restrict          xx,
 386                      rvec                        * gmx_restrict          ff,
 387                      t_forcerec                  * gmx_restrict          fr,
 388                      t_mdatoms                   * gmx_restrict     mdatoms,
 389                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 390                      t_nrnb                      * gmx_restrict        nrnb)
 391 {
 392     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 393      * just 0 for non-waters.
 394      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 395      * jnr indices corresponding to data put in the four positions in the SIMD register.
 396      */
 397     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 398     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 399     int              jnrA,jnrB,jnrC,jnrD;
 400     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 401     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 402     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 403     real             rcutoff_scalar;
 404     real             *shiftvec,*fshift,*x,*f;
 405     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 406     real             scratch[4*DIM];
 407     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 408     int              vdwioffset0;
 409     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 410     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 411     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 412     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 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     nvdwtype         = fr->ntype;
 435     vdwparam         = fr->nbfp;
 436     vdwtype          = mdatoms->typeA;
 437
 438     rcutoff_scalar   = fr->rvdw;
 439     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 440     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 441
 442     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 443     rvdw             = _mm_set1_ps(fr->rvdw);
 444
 445     /* Avoid stupid compiler warnings */
 446     jnrA = jnrB = jnrC = jnrD = 0;
 447     j_coord_offsetA = 0;
 448     j_coord_offsetB = 0;
 449     j_coord_offsetC = 0;
 450     j_coord_offsetD = 0;
 451
 452     outeriter        = 0;
 453     inneriter        = 0;
 454
 455     for(iidx=0;iidx<4*DIM;iidx++)
 456     {
 457         scratch[iidx] = 0.0;
 458     }
 459
 460     /* Start outer loop over neighborlists */
 461     for(iidx=0; iidx<nri; iidx++)
 462     {
 463         /* Load shift vector for this list */
 464         i_shift_offset   = DIM*shiftidx[iidx];
 465
 466         /* Load limits for loop over neighbors */
 467         j_index_start    = jindex[iidx];
 468         j_index_end      = jindex[iidx+1];
 469
 470         /* Get outer coordinate index */
 471         inr              = iinr[iidx];
 472         i_coord_offset   = DIM*inr;
 473
 474         /* Load i particle coords and add shift vector */
 475         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 476
 477         fix0             = _mm_setzero_ps();
 478         fiy0             = _mm_setzero_ps();
 479         fiz0             = _mm_setzero_ps();
 480
 481         /* Load parameters for i particles */
 482         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 483
 484         /* Start inner kernel loop */
 485         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 486         {
 487
 488             /* Get j neighbor index, and coordinate index */
 489             jnrA             = jjnr[jidx];
 490             jnrB             = jjnr[jidx+1];
 491             jnrC             = jjnr[jidx+2];
 492             jnrD             = jjnr[jidx+3];
 493             j_coord_offsetA  = DIM*jnrA;
 494             j_coord_offsetB  = DIM*jnrB;
 495             j_coord_offsetC  = DIM*jnrC;
 496             j_coord_offsetD  = DIM*jnrD;
 497
 498             /* load j atom coordinates */
 499             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 500                                               x+j_coord_offsetC,x+j_coord_offsetD,
 501                                               &jx0,&jy0,&jz0);
 502
 503             /* Calculate displacement vector */
 504             dx00             = _mm_sub_ps(ix0,jx0);
 505             dy00             = _mm_sub_ps(iy0,jy0);
 506             dz00             = _mm_sub_ps(iz0,jz0);
 507
 508             /* Calculate squared distance and things based on it */
 509             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 510
 511             rinvsq00         = gmx_mm_inv_ps(rsq00);
 512
 513             /* Load parameters for j particles */
 514             vdwjidx0A        = 2*vdwtype[jnrA+0];
 515             vdwjidx0B        = 2*vdwtype[jnrB+0];
 516             vdwjidx0C        = 2*vdwtype[jnrC+0];
 517             vdwjidx0D        = 2*vdwtype[jnrD+0];
 518
 519             /**************************
 520              * CALCULATE INTERACTIONS *
 521              **************************/
 522
 523             if (gmx_mm_any_lt(rsq00,rcutoff2))
 524             {
 525
 526             /* Compute parameters for interactions between i and j atoms */
 527             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 528                                          vdwparam+vdwioffset0+vdwjidx0B,
 529                                          vdwparam+vdwioffset0+vdwjidx0C,
 530                                          vdwparam+vdwioffset0+vdwjidx0D,
 531                                          &c6_00,&c12_00);
 532
 533             /* LENNARD-JONES DISPERSION/REPULSION */
 534
 535             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 536             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 537
 538             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 539
 540             fscal            = fvdw;
 541
 542             fscal            = _mm_and_ps(fscal,cutoff_mask);
 543
 544             /* Calculate temporary vectorial force */
 545             tx               = _mm_mul_ps(fscal,dx00);
 546             ty               = _mm_mul_ps(fscal,dy00);
 547             tz               = _mm_mul_ps(fscal,dz00);
 548
 549             /* Update vectorial force */
 550             fix0             = _mm_add_ps(fix0,tx);
 551             fiy0             = _mm_add_ps(fiy0,ty);
 552             fiz0             = _mm_add_ps(fiz0,tz);
 553
 554             fjptrA             = f+j_coord_offsetA;
 555             fjptrB             = f+j_coord_offsetB;
 556             fjptrC             = f+j_coord_offsetC;
 557             fjptrD             = f+j_coord_offsetD;
 558             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 559
 560             }
 561
 562             /* Inner loop uses 30 flops */
 563         }
 564
 565         if(jidx<j_index_end)
 566         {
 567
 568             /* Get j neighbor index, and coordinate index */
 569             jnrlistA         = jjnr[jidx];
 570             jnrlistB         = jjnr[jidx+1];
 571             jnrlistC         = jjnr[jidx+2];
 572             jnrlistD         = jjnr[jidx+3];
 573             /* Sign of each element will be negative for non-real atoms.
 574              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 575              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 576              */
 577             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 578             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 579             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 580             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 581             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 582             j_coord_offsetA  = DIM*jnrA;
 583             j_coord_offsetB  = DIM*jnrB;
 584             j_coord_offsetC  = DIM*jnrC;
 585             j_coord_offsetD  = DIM*jnrD;
 586
 587             /* load j atom coordinates */
 588             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 589                                               x+j_coord_offsetC,x+j_coord_offsetD,
 590                                               &jx0,&jy0,&jz0);
 591
 592             /* Calculate displacement vector */
 593             dx00             = _mm_sub_ps(ix0,jx0);
 594             dy00             = _mm_sub_ps(iy0,jy0);
 595             dz00             = _mm_sub_ps(iz0,jz0);
 596
 597             /* Calculate squared distance and things based on it */
 598             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 599
 600             rinvsq00         = gmx_mm_inv_ps(rsq00);
 601
 602             /* Load parameters for j particles */
 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             if (gmx_mm_any_lt(rsq00,rcutoff2))
 613             {
 614
 615             /* Compute parameters for interactions between i and j atoms */
 616             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 617                                          vdwparam+vdwioffset0+vdwjidx0B,
 618                                          vdwparam+vdwioffset0+vdwjidx0C,
 619                                          vdwparam+vdwioffset0+vdwjidx0D,
 620                                          &c6_00,&c12_00);
 621
 622             /* LENNARD-JONES DISPERSION/REPULSION */
 623
 624             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 625             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 626
 627             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 628
 629             fscal            = fvdw;
 630
 631             fscal            = _mm_and_ps(fscal,cutoff_mask);
 632
 633             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 634
 635             /* Calculate temporary vectorial force */
 636             tx               = _mm_mul_ps(fscal,dx00);
 637             ty               = _mm_mul_ps(fscal,dy00);
 638             tz               = _mm_mul_ps(fscal,dz00);
 639
 640             /* Update vectorial force */
 641             fix0             = _mm_add_ps(fix0,tx);
 642             fiy0             = _mm_add_ps(fiy0,ty);
 643             fiz0             = _mm_add_ps(fiz0,tz);
 644
 645             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 646             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 647             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 648             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 649             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 650
 651             }
 652
 653             /* Inner loop uses 30 flops */
 654         }
 655
 656         /* End of innermost loop */
 657
 658         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 659                                               f+i_coord_offset,fshift+i_shift_offset);
 660
 661         /* Increment number of inner iterations */
 662         inneriter                  += j_index_end - j_index_start;
 663
 664         /* Outer loop uses 6 flops */
 665     }
 666
 667     /* Increment number of outer iterations */
 668     outeriter        += nri;
 669
 670     /* Update outer/inner flops */
 671
 672     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*30);
 673 }