src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecNone_VdwLJ_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_VdwLJ_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_VdwLJ_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     /* Avoid stupid compiler warnings */
 116     jnrA = jnrB = jnrC = jnrD = 0;
 117     j_coord_offsetA = 0;
 118     j_coord_offsetB = 0;
 119     j_coord_offsetC = 0;
 120     j_coord_offsetD = 0;
 121
 122     outeriter        = 0;
 123     inneriter        = 0;
 124
 125     for(iidx=0;iidx<4*DIM;iidx++)
 126     {
 127         scratch[iidx] = 0.0;
 128     }
 129
 130     /* Start outer loop over neighborlists */
 131     for(iidx=0; iidx<nri; iidx++)
 132     {
 133         /* Load shift vector for this list */
 134         i_shift_offset   = DIM*shiftidx[iidx];
 135
 136         /* Load limits for loop over neighbors */
 137         j_index_start    = jindex[iidx];
 138         j_index_end      = jindex[iidx+1];
 139
 140         /* Get outer coordinate index */
 141         inr              = iinr[iidx];
 142         i_coord_offset   = DIM*inr;
 143
 144         /* Load i particle coords and add shift vector */
 145         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 146
 147         fix0             = _mm_setzero_ps();
 148         fiy0             = _mm_setzero_ps();
 149         fiz0             = _mm_setzero_ps();
 150
 151         /* Load parameters for i particles */
 152         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 153
 154         /* Reset potential sums */
 155         vvdwsum          = _mm_setzero_ps();
 156
 157         /* Start inner kernel loop */
 158         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 159         {
 160
 161             /* Get j neighbor index, and coordinate index */
 162             jnrA             = jjnr[jidx];
 163             jnrB             = jjnr[jidx+1];
 164             jnrC             = jjnr[jidx+2];
 165             jnrD             = jjnr[jidx+3];
 166             j_coord_offsetA  = DIM*jnrA;
 167             j_coord_offsetB  = DIM*jnrB;
 168             j_coord_offsetC  = DIM*jnrC;
 169             j_coord_offsetD  = DIM*jnrD;
 170
 171             /* load j atom coordinates */
 172             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 173                                               x+j_coord_offsetC,x+j_coord_offsetD,
 174                                               &jx0,&jy0,&jz0);
 175
 176             /* Calculate displacement vector */
 177             dx00             = _mm_sub_ps(ix0,jx0);
 178             dy00             = _mm_sub_ps(iy0,jy0);
 179             dz00             = _mm_sub_ps(iz0,jz0);
 180
 181             /* Calculate squared distance and things based on it */
 182             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 183
 184             rinvsq00         = gmx_mm_inv_ps(rsq00);
 185
 186             /* Load parameters for j particles */
 187             vdwjidx0A        = 2*vdwtype[jnrA+0];
 188             vdwjidx0B        = 2*vdwtype[jnrB+0];
 189             vdwjidx0C        = 2*vdwtype[jnrC+0];
 190             vdwjidx0D        = 2*vdwtype[jnrD+0];
 191
 192             /**************************
 193              * CALCULATE INTERACTIONS *
 194              **************************/
 195
 196             /* Compute parameters for interactions between i and j atoms */
 197             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 198                                          vdwparam+vdwioffset0+vdwjidx0B,
 199                                          vdwparam+vdwioffset0+vdwjidx0C,
 200                                          vdwparam+vdwioffset0+vdwjidx0D,
 201                                          &c6_00,&c12_00);
 202
 203             /* LENNARD-JONES DISPERSION/REPULSION */
 204
 205             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 206             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 207             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 208             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 209             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 210
 211             /* Update potential sum for this i atom from the interaction with this j atom. */
 212             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 213
 214             fscal            = fvdw;
 215
 216             /* Calculate temporary vectorial force */
 217             tx               = _mm_mul_ps(fscal,dx00);
 218             ty               = _mm_mul_ps(fscal,dy00);
 219             tz               = _mm_mul_ps(fscal,dz00);
 220
 221             /* Update vectorial force */
 222             fix0             = _mm_add_ps(fix0,tx);
 223             fiy0             = _mm_add_ps(fiy0,ty);
 224             fiz0             = _mm_add_ps(fiz0,tz);
 225
 226             fjptrA             = f+j_coord_offsetA;
 227             fjptrB             = f+j_coord_offsetB;
 228             fjptrC             = f+j_coord_offsetC;
 229             fjptrD             = f+j_coord_offsetD;
 230             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 231
 232             /* Inner loop uses 32 flops */
 233         }
 234
 235         if(jidx<j_index_end)
 236         {
 237
 238             /* Get j neighbor index, and coordinate index */
 239             jnrlistA         = jjnr[jidx];
 240             jnrlistB         = jjnr[jidx+1];
 241             jnrlistC         = jjnr[jidx+2];
 242             jnrlistD         = jjnr[jidx+3];
 243             /* Sign of each element will be negative for non-real atoms.
 244              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 245              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 246              */
 247             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 248             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 249             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 250             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 251             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 252             j_coord_offsetA  = DIM*jnrA;
 253             j_coord_offsetB  = DIM*jnrB;
 254             j_coord_offsetC  = DIM*jnrC;
 255             j_coord_offsetD  = DIM*jnrD;
 256
 257             /* load j atom coordinates */
 258             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 259                                               x+j_coord_offsetC,x+j_coord_offsetD,
 260                                               &jx0,&jy0,&jz0);
 261
 262             /* Calculate displacement vector */
 263             dx00             = _mm_sub_ps(ix0,jx0);
 264             dy00             = _mm_sub_ps(iy0,jy0);
 265             dz00             = _mm_sub_ps(iz0,jz0);
 266
 267             /* Calculate squared distance and things based on it */
 268             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 269
 270             rinvsq00         = gmx_mm_inv_ps(rsq00);
 271
 272             /* Load parameters for j particles */
 273             vdwjidx0A        = 2*vdwtype[jnrA+0];
 274             vdwjidx0B        = 2*vdwtype[jnrB+0];
 275             vdwjidx0C        = 2*vdwtype[jnrC+0];
 276             vdwjidx0D        = 2*vdwtype[jnrD+0];
 277
 278             /**************************
 279              * CALCULATE INTERACTIONS *
 280              **************************/
 281
 282             /* Compute parameters for interactions between i and j atoms */
 283             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 284                                          vdwparam+vdwioffset0+vdwjidx0B,
 285                                          vdwparam+vdwioffset0+vdwjidx0C,
 286                                          vdwparam+vdwioffset0+vdwjidx0D,
 287                                          &c6_00,&c12_00);
 288
 289             /* LENNARD-JONES DISPERSION/REPULSION */
 290
 291             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 292             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 293             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 294             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 295             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 296
 297             /* Update potential sum for this i atom from the interaction with this j atom. */
 298             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 299             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 300
 301             fscal            = fvdw;
 302
 303             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 304
 305             /* Calculate temporary vectorial force */
 306             tx               = _mm_mul_ps(fscal,dx00);
 307             ty               = _mm_mul_ps(fscal,dy00);
 308             tz               = _mm_mul_ps(fscal,dz00);
 309
 310             /* Update vectorial force */
 311             fix0             = _mm_add_ps(fix0,tx);
 312             fiy0             = _mm_add_ps(fiy0,ty);
 313             fiz0             = _mm_add_ps(fiz0,tz);
 314
 315             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 316             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 317             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 318             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 319             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 320
 321             /* Inner loop uses 32 flops */
 322         }
 323
 324         /* End of innermost loop */
 325
 326         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 327                                               f+i_coord_offset,fshift+i_shift_offset);
 328
 329         ggid                        = gid[iidx];
 330         /* Update potential energies */
 331         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 332
 333         /* Increment number of inner iterations */
 334         inneriter                  += j_index_end - j_index_start;
 335
 336         /* Outer loop uses 7 flops */
 337     }
 338
 339     /* Increment number of outer iterations */
 340     outeriter        += nri;
 341
 342     /* Update outer/inner flops */
 343
 344     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*32);
 345 }
 346 /*
 347  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJ_GeomP1P1_F_sse2_single
 348  * Electrostatics interaction: None
 349  * VdW interaction:            LennardJones
 350  * Geometry:                   Particle-Particle
 351  * Calculate force/pot:        Force
 352  */
 353 void
 354 nb_kernel_ElecNone_VdwLJ_GeomP1P1_F_sse2_single
 355                     (t_nblist                    * gmx_restrict       nlist,
 356                      rvec                        * gmx_restrict          xx,
 357                      rvec                        * gmx_restrict          ff,
 358                      t_forcerec                  * gmx_restrict          fr,
 359                      t_mdatoms                   * gmx_restrict     mdatoms,
 360                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 361                      t_nrnb                      * gmx_restrict        nrnb)
 362 {
 363     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 364      * just 0 for non-waters.
 365      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 366      * jnr indices corresponding to data put in the four positions in the SIMD register.
 367      */
 368     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 369     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 370     int              jnrA,jnrB,jnrC,jnrD;
 371     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 372     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 373     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 374     real             rcutoff_scalar;
 375     real             *shiftvec,*fshift,*x,*f;
 376     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 377     real             scratch[4*DIM];
 378     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 379     int              vdwioffset0;
 380     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 381     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 382     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 383     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 384     int              nvdwtype;
 385     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 386     int              *vdwtype;
 387     real             *vdwparam;
 388     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 389     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 390     __m128           dummy_mask,cutoff_mask;
 391     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 392     __m128           one     = _mm_set1_ps(1.0);
 393     __m128           two     = _mm_set1_ps(2.0);
 394     x                = xx[0];
 395     f                = ff[0];
 396
 397     nri              = nlist->nri;
 398     iinr             = nlist->iinr;
 399     jindex           = nlist->jindex;
 400     jjnr             = nlist->jjnr;
 401     shiftidx         = nlist->shift;
 402     gid              = nlist->gid;
 403     shiftvec         = fr->shift_vec[0];
 404     fshift           = fr->fshift[0];
 405     nvdwtype         = fr->ntype;
 406     vdwparam         = fr->nbfp;
 407     vdwtype          = mdatoms->typeA;
 408
 409     /* Avoid stupid compiler warnings */
 410     jnrA = jnrB = jnrC = jnrD = 0;
 411     j_coord_offsetA = 0;
 412     j_coord_offsetB = 0;
 413     j_coord_offsetC = 0;
 414     j_coord_offsetD = 0;
 415
 416     outeriter        = 0;
 417     inneriter        = 0;
 418
 419     for(iidx=0;iidx<4*DIM;iidx++)
 420     {
 421         scratch[iidx] = 0.0;
 422     }
 423
 424     /* Start outer loop over neighborlists */
 425     for(iidx=0; iidx<nri; iidx++)
 426     {
 427         /* Load shift vector for this list */
 428         i_shift_offset   = DIM*shiftidx[iidx];
 429
 430         /* Load limits for loop over neighbors */
 431         j_index_start    = jindex[iidx];
 432         j_index_end      = jindex[iidx+1];
 433
 434         /* Get outer coordinate index */
 435         inr              = iinr[iidx];
 436         i_coord_offset   = DIM*inr;
 437
 438         /* Load i particle coords and add shift vector */
 439         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 440
 441         fix0             = _mm_setzero_ps();
 442         fiy0             = _mm_setzero_ps();
 443         fiz0             = _mm_setzero_ps();
 444
 445         /* Load parameters for i particles */
 446         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 447
 448         /* Start inner kernel loop */
 449         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 450         {
 451
 452             /* Get j neighbor index, and coordinate index */
 453             jnrA             = jjnr[jidx];
 454             jnrB             = jjnr[jidx+1];
 455             jnrC             = jjnr[jidx+2];
 456             jnrD             = jjnr[jidx+3];
 457             j_coord_offsetA  = DIM*jnrA;
 458             j_coord_offsetB  = DIM*jnrB;
 459             j_coord_offsetC  = DIM*jnrC;
 460             j_coord_offsetD  = DIM*jnrD;
 461
 462             /* load j atom coordinates */
 463             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 464                                               x+j_coord_offsetC,x+j_coord_offsetD,
 465                                               &jx0,&jy0,&jz0);
 466
 467             /* Calculate displacement vector */
 468             dx00             = _mm_sub_ps(ix0,jx0);
 469             dy00             = _mm_sub_ps(iy0,jy0);
 470             dz00             = _mm_sub_ps(iz0,jz0);
 471
 472             /* Calculate squared distance and things based on it */
 473             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 474
 475             rinvsq00         = gmx_mm_inv_ps(rsq00);
 476
 477             /* Load parameters for j particles */
 478             vdwjidx0A        = 2*vdwtype[jnrA+0];
 479             vdwjidx0B        = 2*vdwtype[jnrB+0];
 480             vdwjidx0C        = 2*vdwtype[jnrC+0];
 481             vdwjidx0D        = 2*vdwtype[jnrD+0];
 482
 483             /**************************
 484              * CALCULATE INTERACTIONS *
 485              **************************/
 486
 487             /* Compute parameters for interactions between i and j atoms */
 488             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 489                                          vdwparam+vdwioffset0+vdwjidx0B,
 490                                          vdwparam+vdwioffset0+vdwjidx0C,
 491                                          vdwparam+vdwioffset0+vdwjidx0D,
 492                                          &c6_00,&c12_00);
 493
 494             /* LENNARD-JONES DISPERSION/REPULSION */
 495
 496             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 497             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 498
 499             fscal            = fvdw;
 500
 501             /* Calculate temporary vectorial force */
 502             tx               = _mm_mul_ps(fscal,dx00);
 503             ty               = _mm_mul_ps(fscal,dy00);
 504             tz               = _mm_mul_ps(fscal,dz00);
 505
 506             /* Update vectorial force */
 507             fix0             = _mm_add_ps(fix0,tx);
 508             fiy0             = _mm_add_ps(fiy0,ty);
 509             fiz0             = _mm_add_ps(fiz0,tz);
 510
 511             fjptrA             = f+j_coord_offsetA;
 512             fjptrB             = f+j_coord_offsetB;
 513             fjptrC             = f+j_coord_offsetC;
 514             fjptrD             = f+j_coord_offsetD;
 515             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 516
 517             /* Inner loop uses 27 flops */
 518         }
 519
 520         if(jidx<j_index_end)
 521         {
 522
 523             /* Get j neighbor index, and coordinate index */
 524             jnrlistA         = jjnr[jidx];
 525             jnrlistB         = jjnr[jidx+1];
 526             jnrlistC         = jjnr[jidx+2];
 527             jnrlistD         = jjnr[jidx+3];
 528             /* Sign of each element will be negative for non-real atoms.
 529              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 530              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 531              */
 532             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 533             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 534             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 535             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 536             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 537             j_coord_offsetA  = DIM*jnrA;
 538             j_coord_offsetB  = DIM*jnrB;
 539             j_coord_offsetC  = DIM*jnrC;
 540             j_coord_offsetD  = DIM*jnrD;
 541
 542             /* load j atom coordinates */
 543             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 544                                               x+j_coord_offsetC,x+j_coord_offsetD,
 545                                               &jx0,&jy0,&jz0);
 546
 547             /* Calculate displacement vector */
 548             dx00             = _mm_sub_ps(ix0,jx0);
 549             dy00             = _mm_sub_ps(iy0,jy0);
 550             dz00             = _mm_sub_ps(iz0,jz0);
 551
 552             /* Calculate squared distance and things based on it */
 553             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 554
 555             rinvsq00         = gmx_mm_inv_ps(rsq00);
 556
 557             /* Load parameters for j particles */
 558             vdwjidx0A        = 2*vdwtype[jnrA+0];
 559             vdwjidx0B        = 2*vdwtype[jnrB+0];
 560             vdwjidx0C        = 2*vdwtype[jnrC+0];
 561             vdwjidx0D        = 2*vdwtype[jnrD+0];
 562
 563             /**************************
 564              * CALCULATE INTERACTIONS *
 565              **************************/
 566
 567             /* Compute parameters for interactions between i and j atoms */
 568             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 569                                          vdwparam+vdwioffset0+vdwjidx0B,
 570                                          vdwparam+vdwioffset0+vdwjidx0C,
 571                                          vdwparam+vdwioffset0+vdwjidx0D,
 572                                          &c6_00,&c12_00);
 573
 574             /* LENNARD-JONES DISPERSION/REPULSION */
 575
 576             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 577             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 578
 579             fscal            = fvdw;
 580
 581             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 582
 583             /* Calculate temporary vectorial force */
 584             tx               = _mm_mul_ps(fscal,dx00);
 585             ty               = _mm_mul_ps(fscal,dy00);
 586             tz               = _mm_mul_ps(fscal,dz00);
 587
 588             /* Update vectorial force */
 589             fix0             = _mm_add_ps(fix0,tx);
 590             fiy0             = _mm_add_ps(fiy0,ty);
 591             fiz0             = _mm_add_ps(fiz0,tz);
 592
 593             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 594             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 595             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 596             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 597             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 598
 599             /* Inner loop uses 27 flops */
 600         }
 601
 602         /* End of innermost loop */
 603
 604         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 605                                               f+i_coord_offset,fshift+i_shift_offset);
 606
 607         /* Increment number of inner iterations */
 608         inneriter                  += j_index_end - j_index_start;
 609
 610         /* Outer loop uses 6 flops */
 611     }
 612
 613     /* Increment number of outer iterations */
 614     outeriter        += nri;
 615
 616     /* Update outer/inner flops */
 617
 618     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*27);
 619 }