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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_single.h"
  48 #include "kernelutil_x86_sse2_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJ_GeomP1P1_VF_sse2_single
  52  * Electrostatics interaction: None
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecNone_VdwLJ_GeomP1P1_VF_sse2_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 SSE, 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           tx,ty,tz,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     int              nvdwtype;
  89     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  90     int              *vdwtype;
  91     real             *vdwparam;
  92     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  93     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  94     __m128           dummy_mask,cutoff_mask;
  95     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  96     __m128           one     = _mm_set1_ps(1.0);
  97     __m128           two     = _mm_set1_ps(2.0);
  98     x                = xx[0];
  99     f                = ff[0];
 100
 101     nri              = nlist->nri;
 102     iinr             = nlist->iinr;
 103     jindex           = nlist->jindex;
 104     jjnr             = nlist->jjnr;
 105     shiftidx         = nlist->shift;
 106     gid              = nlist->gid;
 107     shiftvec         = fr->shift_vec[0];
 108     fshift           = fr->fshift[0];
 109     nvdwtype         = fr->ntype;
 110     vdwparam         = fr->nbfp;
 111     vdwtype          = mdatoms->typeA;
 112
 113     /* Avoid stupid compiler warnings */
 114     jnrA = jnrB = jnrC = jnrD = 0;
 115     j_coord_offsetA = 0;
 116     j_coord_offsetB = 0;
 117     j_coord_offsetC = 0;
 118     j_coord_offsetD = 0;
 119
 120     outeriter        = 0;
 121     inneriter        = 0;
 122
 123     for(iidx=0;iidx<4*DIM;iidx++)
 124     {
 125         scratch[iidx] = 0.0;
 126     }
 127
 128     /* Start outer loop over neighborlists */
 129     for(iidx=0; iidx<nri; iidx++)
 130     {
 131         /* Load shift vector for this list */
 132         i_shift_offset   = DIM*shiftidx[iidx];
 133
 134         /* Load limits for loop over neighbors */
 135         j_index_start    = jindex[iidx];
 136         j_index_end      = jindex[iidx+1];
 137
 138         /* Get outer coordinate index */
 139         inr              = iinr[iidx];
 140         i_coord_offset   = DIM*inr;
 141
 142         /* Load i particle coords and add shift vector */
 143         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 144
 145         fix0             = _mm_setzero_ps();
 146         fiy0             = _mm_setzero_ps();
 147         fiz0             = _mm_setzero_ps();
 148
 149         /* Load parameters for i particles */
 150         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 151
 152         /* Reset potential sums */
 153         vvdwsum          = _mm_setzero_ps();
 154
 155         /* Start inner kernel loop */
 156         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 157         {
 158
 159             /* Get j neighbor index, and coordinate index */
 160             jnrA             = jjnr[jidx];
 161             jnrB             = jjnr[jidx+1];
 162             jnrC             = jjnr[jidx+2];
 163             jnrD             = jjnr[jidx+3];
 164             j_coord_offsetA  = DIM*jnrA;
 165             j_coord_offsetB  = DIM*jnrB;
 166             j_coord_offsetC  = DIM*jnrC;
 167             j_coord_offsetD  = DIM*jnrD;
 168
 169             /* load j atom coordinates */
 170             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 171                                               x+j_coord_offsetC,x+j_coord_offsetD,
 172                                               &jx0,&jy0,&jz0);
 173
 174             /* Calculate displacement vector */
 175             dx00             = _mm_sub_ps(ix0,jx0);
 176             dy00             = _mm_sub_ps(iy0,jy0);
 177             dz00             = _mm_sub_ps(iz0,jz0);
 178
 179             /* Calculate squared distance and things based on it */
 180             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 181
 182             rinvsq00         = gmx_mm_inv_ps(rsq00);
 183
 184             /* Load parameters for j particles */
 185             vdwjidx0A        = 2*vdwtype[jnrA+0];
 186             vdwjidx0B        = 2*vdwtype[jnrB+0];
 187             vdwjidx0C        = 2*vdwtype[jnrC+0];
 188             vdwjidx0D        = 2*vdwtype[jnrD+0];
 189
 190             /**************************
 191              * CALCULATE INTERACTIONS *
 192              **************************/
 193
 194             /* Compute parameters for interactions between i and j atoms */
 195             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 196                                          vdwparam+vdwioffset0+vdwjidx0B,
 197                                          vdwparam+vdwioffset0+vdwjidx0C,
 198                                          vdwparam+vdwioffset0+vdwjidx0D,
 199                                          &c6_00,&c12_00);
 200
 201             /* LENNARD-JONES DISPERSION/REPULSION */
 202
 203             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 204             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 205             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 206             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 207             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 208
 209             /* Update potential sum for this i atom from the interaction with this j atom. */
 210             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 211
 212             fscal            = fvdw;
 213
 214             /* Calculate temporary vectorial force */
 215             tx               = _mm_mul_ps(fscal,dx00);
 216             ty               = _mm_mul_ps(fscal,dy00);
 217             tz               = _mm_mul_ps(fscal,dz00);
 218
 219             /* Update vectorial force */
 220             fix0             = _mm_add_ps(fix0,tx);
 221             fiy0             = _mm_add_ps(fiy0,ty);
 222             fiz0             = _mm_add_ps(fiz0,tz);
 223
 224             fjptrA             = f+j_coord_offsetA;
 225             fjptrB             = f+j_coord_offsetB;
 226             fjptrC             = f+j_coord_offsetC;
 227             fjptrD             = f+j_coord_offsetD;
 228             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 229
 230             /* Inner loop uses 32 flops */
 231         }
 232
 233         if(jidx<j_index_end)
 234         {
 235
 236             /* Get j neighbor index, and coordinate index */
 237             jnrlistA         = jjnr[jidx];
 238             jnrlistB         = jjnr[jidx+1];
 239             jnrlistC         = jjnr[jidx+2];
 240             jnrlistD         = jjnr[jidx+3];
 241             /* Sign of each element will be negative for non-real atoms.
 242              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 243              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 244              */
 245             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 246             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 247             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 248             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 249             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 250             j_coord_offsetA  = DIM*jnrA;
 251             j_coord_offsetB  = DIM*jnrB;
 252             j_coord_offsetC  = DIM*jnrC;
 253             j_coord_offsetD  = DIM*jnrD;
 254
 255             /* load j atom coordinates */
 256             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 257                                               x+j_coord_offsetC,x+j_coord_offsetD,
 258                                               &jx0,&jy0,&jz0);
 259
 260             /* Calculate displacement vector */
 261             dx00             = _mm_sub_ps(ix0,jx0);
 262             dy00             = _mm_sub_ps(iy0,jy0);
 263             dz00             = _mm_sub_ps(iz0,jz0);
 264
 265             /* Calculate squared distance and things based on it */
 266             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 267
 268             rinvsq00         = gmx_mm_inv_ps(rsq00);
 269
 270             /* Load parameters for j particles */
 271             vdwjidx0A        = 2*vdwtype[jnrA+0];
 272             vdwjidx0B        = 2*vdwtype[jnrB+0];
 273             vdwjidx0C        = 2*vdwtype[jnrC+0];
 274             vdwjidx0D        = 2*vdwtype[jnrD+0];
 275
 276             /**************************
 277              * CALCULATE INTERACTIONS *
 278              **************************/
 279
 280             /* Compute parameters for interactions between i and j atoms */
 281             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 282                                          vdwparam+vdwioffset0+vdwjidx0B,
 283                                          vdwparam+vdwioffset0+vdwjidx0C,
 284                                          vdwparam+vdwioffset0+vdwjidx0D,
 285                                          &c6_00,&c12_00);
 286
 287             /* LENNARD-JONES DISPERSION/REPULSION */
 288
 289             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 290             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 291             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 292             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 293             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 294
 295             /* Update potential sum for this i atom from the interaction with this j atom. */
 296             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 297             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 298
 299             fscal            = fvdw;
 300
 301             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 302
 303             /* Calculate temporary vectorial force */
 304             tx               = _mm_mul_ps(fscal,dx00);
 305             ty               = _mm_mul_ps(fscal,dy00);
 306             tz               = _mm_mul_ps(fscal,dz00);
 307
 308             /* Update vectorial force */
 309             fix0             = _mm_add_ps(fix0,tx);
 310             fiy0             = _mm_add_ps(fiy0,ty);
 311             fiz0             = _mm_add_ps(fiz0,tz);
 312
 313             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 314             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 315             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 316             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 317             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 318
 319             /* Inner loop uses 32 flops */
 320         }
 321
 322         /* End of innermost loop */
 323
 324         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 325                                               f+i_coord_offset,fshift+i_shift_offset);
 326
 327         ggid                        = gid[iidx];
 328         /* Update potential energies */
 329         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 330
 331         /* Increment number of inner iterations */
 332         inneriter                  += j_index_end - j_index_start;
 333
 334         /* Outer loop uses 7 flops */
 335     }
 336
 337     /* Increment number of outer iterations */
 338     outeriter        += nri;
 339
 340     /* Update outer/inner flops */
 341
 342     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*32);
 343 }
 344 /*
 345  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJ_GeomP1P1_F_sse2_single
 346  * Electrostatics interaction: None
 347  * VdW interaction:            LennardJones
 348  * Geometry:                   Particle-Particle
 349  * Calculate force/pot:        Force
 350  */
 351 void
 352 nb_kernel_ElecNone_VdwLJ_GeomP1P1_F_sse2_single
 353                     (t_nblist                    * gmx_restrict       nlist,
 354                      rvec                        * gmx_restrict          xx,
 355                      rvec                        * gmx_restrict          ff,
 356                      t_forcerec                  * gmx_restrict          fr,
 357                      t_mdatoms                   * gmx_restrict     mdatoms,
 358                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 359                      t_nrnb                      * gmx_restrict        nrnb)
 360 {
 361     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 362      * just 0 for non-waters.
 363      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 364      * jnr indices corresponding to data put in the four positions in the SIMD register.
 365      */
 366     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 367     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 368     int              jnrA,jnrB,jnrC,jnrD;
 369     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 370     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 371     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 372     real             rcutoff_scalar;
 373     real             *shiftvec,*fshift,*x,*f;
 374     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 375     real             scratch[4*DIM];
 376     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 377     int              vdwioffset0;
 378     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 379     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 380     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 381     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 382     int              nvdwtype;
 383     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 384     int              *vdwtype;
 385     real             *vdwparam;
 386     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 387     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 388     __m128           dummy_mask,cutoff_mask;
 389     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 390     __m128           one     = _mm_set1_ps(1.0);
 391     __m128           two     = _mm_set1_ps(2.0);
 392     x                = xx[0];
 393     f                = ff[0];
 394
 395     nri              = nlist->nri;
 396     iinr             = nlist->iinr;
 397     jindex           = nlist->jindex;
 398     jjnr             = nlist->jjnr;
 399     shiftidx         = nlist->shift;
 400     gid              = nlist->gid;
 401     shiftvec         = fr->shift_vec[0];
 402     fshift           = fr->fshift[0];
 403     nvdwtype         = fr->ntype;
 404     vdwparam         = fr->nbfp;
 405     vdwtype          = mdatoms->typeA;
 406
 407     /* Avoid stupid compiler warnings */
 408     jnrA = jnrB = jnrC = jnrD = 0;
 409     j_coord_offsetA = 0;
 410     j_coord_offsetB = 0;
 411     j_coord_offsetC = 0;
 412     j_coord_offsetD = 0;
 413
 414     outeriter        = 0;
 415     inneriter        = 0;
 416
 417     for(iidx=0;iidx<4*DIM;iidx++)
 418     {
 419         scratch[iidx] = 0.0;
 420     }
 421
 422     /* Start outer loop over neighborlists */
 423     for(iidx=0; iidx<nri; iidx++)
 424     {
 425         /* Load shift vector for this list */
 426         i_shift_offset   = DIM*shiftidx[iidx];
 427
 428         /* Load limits for loop over neighbors */
 429         j_index_start    = jindex[iidx];
 430         j_index_end      = jindex[iidx+1];
 431
 432         /* Get outer coordinate index */
 433         inr              = iinr[iidx];
 434         i_coord_offset   = DIM*inr;
 435
 436         /* Load i particle coords and add shift vector */
 437         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 438
 439         fix0             = _mm_setzero_ps();
 440         fiy0             = _mm_setzero_ps();
 441         fiz0             = _mm_setzero_ps();
 442
 443         /* Load parameters for i particles */
 444         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 445
 446         /* Start inner kernel loop */
 447         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 448         {
 449
 450             /* Get j neighbor index, and coordinate index */
 451             jnrA             = jjnr[jidx];
 452             jnrB             = jjnr[jidx+1];
 453             jnrC             = jjnr[jidx+2];
 454             jnrD             = jjnr[jidx+3];
 455             j_coord_offsetA  = DIM*jnrA;
 456             j_coord_offsetB  = DIM*jnrB;
 457             j_coord_offsetC  = DIM*jnrC;
 458             j_coord_offsetD  = DIM*jnrD;
 459
 460             /* load j atom coordinates */
 461             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 462                                               x+j_coord_offsetC,x+j_coord_offsetD,
 463                                               &jx0,&jy0,&jz0);
 464
 465             /* Calculate displacement vector */
 466             dx00             = _mm_sub_ps(ix0,jx0);
 467             dy00             = _mm_sub_ps(iy0,jy0);
 468             dz00             = _mm_sub_ps(iz0,jz0);
 469
 470             /* Calculate squared distance and things based on it */
 471             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 472
 473             rinvsq00         = gmx_mm_inv_ps(rsq00);
 474
 475             /* Load parameters for j particles */
 476             vdwjidx0A        = 2*vdwtype[jnrA+0];
 477             vdwjidx0B        = 2*vdwtype[jnrB+0];
 478             vdwjidx0C        = 2*vdwtype[jnrC+0];
 479             vdwjidx0D        = 2*vdwtype[jnrD+0];
 480
 481             /**************************
 482              * CALCULATE INTERACTIONS *
 483              **************************/
 484
 485             /* Compute parameters for interactions between i and j atoms */
 486             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 487                                          vdwparam+vdwioffset0+vdwjidx0B,
 488                                          vdwparam+vdwioffset0+vdwjidx0C,
 489                                          vdwparam+vdwioffset0+vdwjidx0D,
 490                                          &c6_00,&c12_00);
 491
 492             /* LENNARD-JONES DISPERSION/REPULSION */
 493
 494             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 495             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 496
 497             fscal            = fvdw;
 498
 499             /* Calculate temporary vectorial force */
 500             tx               = _mm_mul_ps(fscal,dx00);
 501             ty               = _mm_mul_ps(fscal,dy00);
 502             tz               = _mm_mul_ps(fscal,dz00);
 503
 504             /* Update vectorial force */
 505             fix0             = _mm_add_ps(fix0,tx);
 506             fiy0             = _mm_add_ps(fiy0,ty);
 507             fiz0             = _mm_add_ps(fiz0,tz);
 508
 509             fjptrA             = f+j_coord_offsetA;
 510             fjptrB             = f+j_coord_offsetB;
 511             fjptrC             = f+j_coord_offsetC;
 512             fjptrD             = f+j_coord_offsetD;
 513             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 514
 515             /* Inner loop uses 27 flops */
 516         }
 517
 518         if(jidx<j_index_end)
 519         {
 520
 521             /* Get j neighbor index, and coordinate index */
 522             jnrlistA         = jjnr[jidx];
 523             jnrlistB         = jjnr[jidx+1];
 524             jnrlistC         = jjnr[jidx+2];
 525             jnrlistD         = jjnr[jidx+3];
 526             /* Sign of each element will be negative for non-real atoms.
 527              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 528              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 529              */
 530             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 531             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 532             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 533             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 534             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 535             j_coord_offsetA  = DIM*jnrA;
 536             j_coord_offsetB  = DIM*jnrB;
 537             j_coord_offsetC  = DIM*jnrC;
 538             j_coord_offsetD  = DIM*jnrD;
 539
 540             /* load j atom coordinates */
 541             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 542                                               x+j_coord_offsetC,x+j_coord_offsetD,
 543                                               &jx0,&jy0,&jz0);
 544
 545             /* Calculate displacement vector */
 546             dx00             = _mm_sub_ps(ix0,jx0);
 547             dy00             = _mm_sub_ps(iy0,jy0);
 548             dz00             = _mm_sub_ps(iz0,jz0);
 549
 550             /* Calculate squared distance and things based on it */
 551             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 552
 553             rinvsq00         = gmx_mm_inv_ps(rsq00);
 554
 555             /* Load parameters for j particles */
 556             vdwjidx0A        = 2*vdwtype[jnrA+0];
 557             vdwjidx0B        = 2*vdwtype[jnrB+0];
 558             vdwjidx0C        = 2*vdwtype[jnrC+0];
 559             vdwjidx0D        = 2*vdwtype[jnrD+0];
 560
 561             /**************************
 562              * CALCULATE INTERACTIONS *
 563              **************************/
 564
 565             /* Compute parameters for interactions between i and j atoms */
 566             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 567                                          vdwparam+vdwioffset0+vdwjidx0B,
 568                                          vdwparam+vdwioffset0+vdwjidx0C,
 569                                          vdwparam+vdwioffset0+vdwjidx0D,
 570                                          &c6_00,&c12_00);
 571
 572             /* LENNARD-JONES DISPERSION/REPULSION */
 573
 574             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 575             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 576
 577             fscal            = fvdw;
 578
 579             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 580
 581             /* Calculate temporary vectorial force */
 582             tx               = _mm_mul_ps(fscal,dx00);
 583             ty               = _mm_mul_ps(fscal,dy00);
 584             tz               = _mm_mul_ps(fscal,dz00);
 585
 586             /* Update vectorial force */
 587             fix0             = _mm_add_ps(fix0,tx);
 588             fiy0             = _mm_add_ps(fiy0,ty);
 589             fiz0             = _mm_add_ps(fiz0,tz);
 590
 591             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 592             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 593             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 594             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 595             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 596
 597             /* Inner loop uses 27 flops */
 598         }
 599
 600         /* End of innermost loop */
 601
 602         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 603                                               f+i_coord_offset,fshift+i_shift_offset);
 604
 605         /* Increment number of inner iterations */
 606         inneriter                  += j_index_end - j_index_start;
 607
 608         /* Outer loop uses 6 flops */
 609     }
 610
 611     /* Increment number of outer iterations */
 612     outeriter        += nri;
 613
 614     /* Update outer/inner flops */
 615
 616     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*27);
 617 }