src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecNone_VdwLJ_GeomP1P1_sse4_1_single.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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_VdwLJ_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_VdwLJ_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     /* Avoid stupid compiler warnings */
 113     jnrA = jnrB = jnrC = jnrD = 0;
 114     j_coord_offsetA = 0;
 115     j_coord_offsetB = 0;
 116     j_coord_offsetC = 0;
 117     j_coord_offsetD = 0;
 118
 119     outeriter        = 0;
 120     inneriter        = 0;
 121
 122     for(iidx=0;iidx<4*DIM;iidx++)
 123     {
 124         scratch[iidx] = 0.0;
 125     }
 126
 127     /* Start outer loop over neighborlists */
 128     for(iidx=0; iidx<nri; iidx++)
 129     {
 130         /* Load shift vector for this list */
 131         i_shift_offset   = DIM*shiftidx[iidx];
 132
 133         /* Load limits for loop over neighbors */
 134         j_index_start    = jindex[iidx];
 135         j_index_end      = jindex[iidx+1];
 136
 137         /* Get outer coordinate index */
 138         inr              = iinr[iidx];
 139         i_coord_offset   = DIM*inr;
 140
 141         /* Load i particle coords and add shift vector */
 142         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 143
 144         fix0             = _mm_setzero_ps();
 145         fiy0             = _mm_setzero_ps();
 146         fiz0             = _mm_setzero_ps();
 147
 148         /* Load parameters for i particles */
 149         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 150
 151         /* Reset potential sums */
 152         vvdwsum          = _mm_setzero_ps();
 153
 154         /* Start inner kernel loop */
 155         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 156         {
 157
 158             /* Get j neighbor index, and coordinate index */
 159             jnrA             = jjnr[jidx];
 160             jnrB             = jjnr[jidx+1];
 161             jnrC             = jjnr[jidx+2];
 162             jnrD             = jjnr[jidx+3];
 163             j_coord_offsetA  = DIM*jnrA;
 164             j_coord_offsetB  = DIM*jnrB;
 165             j_coord_offsetC  = DIM*jnrC;
 166             j_coord_offsetD  = DIM*jnrD;
 167
 168             /* load j atom coordinates */
 169             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 170                                               x+j_coord_offsetC,x+j_coord_offsetD,
 171                                               &jx0,&jy0,&jz0);
 172
 173             /* Calculate displacement vector */
 174             dx00             = _mm_sub_ps(ix0,jx0);
 175             dy00             = _mm_sub_ps(iy0,jy0);
 176             dz00             = _mm_sub_ps(iz0,jz0);
 177
 178             /* Calculate squared distance and things based on it */
 179             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 180
 181             rinvsq00         = sse41_inv_f(rsq00);
 182
 183             /* Load parameters for j particles */
 184             vdwjidx0A        = 2*vdwtype[jnrA+0];
 185             vdwjidx0B        = 2*vdwtype[jnrB+0];
 186             vdwjidx0C        = 2*vdwtype[jnrC+0];
 187             vdwjidx0D        = 2*vdwtype[jnrD+0];
 188
 189             /**************************
 190              * CALCULATE INTERACTIONS *
 191              **************************/
 192
 193             /* Compute parameters for interactions between i and j atoms */
 194             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 195                                          vdwparam+vdwioffset0+vdwjidx0B,
 196                                          vdwparam+vdwioffset0+vdwjidx0C,
 197                                          vdwparam+vdwioffset0+vdwjidx0D,
 198                                          &c6_00,&c12_00);
 199
 200             /* LENNARD-JONES DISPERSION/REPULSION */
 201
 202             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 203             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 204             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 205             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 206             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 207
 208             /* Update potential sum for this i atom from the interaction with this j atom. */
 209             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 210
 211             fscal            = fvdw;
 212
 213             /* Calculate temporary vectorial force */
 214             tx               = _mm_mul_ps(fscal,dx00);
 215             ty               = _mm_mul_ps(fscal,dy00);
 216             tz               = _mm_mul_ps(fscal,dz00);
 217
 218             /* Update vectorial force */
 219             fix0             = _mm_add_ps(fix0,tx);
 220             fiy0             = _mm_add_ps(fiy0,ty);
 221             fiz0             = _mm_add_ps(fiz0,tz);
 222
 223             fjptrA             = f+j_coord_offsetA;
 224             fjptrB             = f+j_coord_offsetB;
 225             fjptrC             = f+j_coord_offsetC;
 226             fjptrD             = f+j_coord_offsetD;
 227             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 228
 229             /* Inner loop uses 32 flops */
 230         }
 231
 232         if(jidx<j_index_end)
 233         {
 234
 235             /* Get j neighbor index, and coordinate index */
 236             jnrlistA         = jjnr[jidx];
 237             jnrlistB         = jjnr[jidx+1];
 238             jnrlistC         = jjnr[jidx+2];
 239             jnrlistD         = jjnr[jidx+3];
 240             /* Sign of each element will be negative for non-real atoms.
 241              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 242              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 243              */
 244             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 245             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 246             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 247             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 248             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 249             j_coord_offsetA  = DIM*jnrA;
 250             j_coord_offsetB  = DIM*jnrB;
 251             j_coord_offsetC  = DIM*jnrC;
 252             j_coord_offsetD  = DIM*jnrD;
 253
 254             /* load j atom coordinates */
 255             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 256                                               x+j_coord_offsetC,x+j_coord_offsetD,
 257                                               &jx0,&jy0,&jz0);
 258
 259             /* Calculate displacement vector */
 260             dx00             = _mm_sub_ps(ix0,jx0);
 261             dy00             = _mm_sub_ps(iy0,jy0);
 262             dz00             = _mm_sub_ps(iz0,jz0);
 263
 264             /* Calculate squared distance and things based on it */
 265             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 266
 267             rinvsq00         = sse41_inv_f(rsq00);
 268
 269             /* Load parameters for j particles */
 270             vdwjidx0A        = 2*vdwtype[jnrA+0];
 271             vdwjidx0B        = 2*vdwtype[jnrB+0];
 272             vdwjidx0C        = 2*vdwtype[jnrC+0];
 273             vdwjidx0D        = 2*vdwtype[jnrD+0];
 274
 275             /**************************
 276              * CALCULATE INTERACTIONS *
 277              **************************/
 278
 279             /* Compute parameters for interactions between i and j atoms */
 280             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 281                                          vdwparam+vdwioffset0+vdwjidx0B,
 282                                          vdwparam+vdwioffset0+vdwjidx0C,
 283                                          vdwparam+vdwioffset0+vdwjidx0D,
 284                                          &c6_00,&c12_00);
 285
 286             /* LENNARD-JONES DISPERSION/REPULSION */
 287
 288             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 289             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 290             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 291             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 292             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 293
 294             /* Update potential sum for this i atom from the interaction with this j atom. */
 295             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 296             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 297
 298             fscal            = fvdw;
 299
 300             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 301
 302             /* Calculate temporary vectorial force */
 303             tx               = _mm_mul_ps(fscal,dx00);
 304             ty               = _mm_mul_ps(fscal,dy00);
 305             tz               = _mm_mul_ps(fscal,dz00);
 306
 307             /* Update vectorial force */
 308             fix0             = _mm_add_ps(fix0,tx);
 309             fiy0             = _mm_add_ps(fiy0,ty);
 310             fiz0             = _mm_add_ps(fiz0,tz);
 311
 312             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 313             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 314             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 315             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 316             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 317
 318             /* Inner loop uses 32 flops */
 319         }
 320
 321         /* End of innermost loop */
 322
 323         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 324                                               f+i_coord_offset,fshift+i_shift_offset);
 325
 326         ggid                        = gid[iidx];
 327         /* Update potential energies */
 328         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 329
 330         /* Increment number of inner iterations */
 331         inneriter                  += j_index_end - j_index_start;
 332
 333         /* Outer loop uses 7 flops */
 334     }
 335
 336     /* Increment number of outer iterations */
 337     outeriter        += nri;
 338
 339     /* Update outer/inner flops */
 340
 341     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*32);
 342 }
 343 /*
 344  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJ_GeomP1P1_F_sse4_1_single
 345  * Electrostatics interaction: None
 346  * VdW interaction:            LennardJones
 347  * Geometry:                   Particle-Particle
 348  * Calculate force/pot:        Force
 349  */
 350 void
 351 nb_kernel_ElecNone_VdwLJ_GeomP1P1_F_sse4_1_single
 352                     (t_nblist                    * gmx_restrict       nlist,
 353                      rvec                        * gmx_restrict          xx,
 354                      rvec                        * gmx_restrict          ff,
 355                      struct t_forcerec           * gmx_restrict          fr,
 356                      t_mdatoms                   * gmx_restrict     mdatoms,
 357                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 358                      t_nrnb                      * gmx_restrict        nrnb)
 359 {
 360     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 361      * just 0 for non-waters.
 362      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 363      * jnr indices corresponding to data put in the four positions in the SIMD register.
 364      */
 365     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 366     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 367     int              jnrA,jnrB,jnrC,jnrD;
 368     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 369     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 370     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 371     real             rcutoff_scalar;
 372     real             *shiftvec,*fshift,*x,*f;
 373     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 374     real             scratch[4*DIM];
 375     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 376     int              vdwioffset0;
 377     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 378     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 379     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 380     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 381     int              nvdwtype;
 382     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 383     int              *vdwtype;
 384     real             *vdwparam;
 385     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 386     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 387     __m128           dummy_mask,cutoff_mask;
 388     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 389     __m128           one     = _mm_set1_ps(1.0);
 390     __m128           two     = _mm_set1_ps(2.0);
 391     x                = xx[0];
 392     f                = ff[0];
 393
 394     nri              = nlist->nri;
 395     iinr             = nlist->iinr;
 396     jindex           = nlist->jindex;
 397     jjnr             = nlist->jjnr;
 398     shiftidx         = nlist->shift;
 399     gid              = nlist->gid;
 400     shiftvec         = fr->shift_vec[0];
 401     fshift           = fr->fshift[0];
 402     nvdwtype         = fr->ntype;
 403     vdwparam         = fr->nbfp;
 404     vdwtype          = mdatoms->typeA;
 405
 406     /* Avoid stupid compiler warnings */
 407     jnrA = jnrB = jnrC = jnrD = 0;
 408     j_coord_offsetA = 0;
 409     j_coord_offsetB = 0;
 410     j_coord_offsetC = 0;
 411     j_coord_offsetD = 0;
 412
 413     outeriter        = 0;
 414     inneriter        = 0;
 415
 416     for(iidx=0;iidx<4*DIM;iidx++)
 417     {
 418         scratch[iidx] = 0.0;
 419     }
 420
 421     /* Start outer loop over neighborlists */
 422     for(iidx=0; iidx<nri; iidx++)
 423     {
 424         /* Load shift vector for this list */
 425         i_shift_offset   = DIM*shiftidx[iidx];
 426
 427         /* Load limits for loop over neighbors */
 428         j_index_start    = jindex[iidx];
 429         j_index_end      = jindex[iidx+1];
 430
 431         /* Get outer coordinate index */
 432         inr              = iinr[iidx];
 433         i_coord_offset   = DIM*inr;
 434
 435         /* Load i particle coords and add shift vector */
 436         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 437
 438         fix0             = _mm_setzero_ps();
 439         fiy0             = _mm_setzero_ps();
 440         fiz0             = _mm_setzero_ps();
 441
 442         /* Load parameters for i particles */
 443         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 444
 445         /* Start inner kernel loop */
 446         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 447         {
 448
 449             /* Get j neighbor index, and coordinate index */
 450             jnrA             = jjnr[jidx];
 451             jnrB             = jjnr[jidx+1];
 452             jnrC             = jjnr[jidx+2];
 453             jnrD             = jjnr[jidx+3];
 454             j_coord_offsetA  = DIM*jnrA;
 455             j_coord_offsetB  = DIM*jnrB;
 456             j_coord_offsetC  = DIM*jnrC;
 457             j_coord_offsetD  = DIM*jnrD;
 458
 459             /* load j atom coordinates */
 460             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 461                                               x+j_coord_offsetC,x+j_coord_offsetD,
 462                                               &jx0,&jy0,&jz0);
 463
 464             /* Calculate displacement vector */
 465             dx00             = _mm_sub_ps(ix0,jx0);
 466             dy00             = _mm_sub_ps(iy0,jy0);
 467             dz00             = _mm_sub_ps(iz0,jz0);
 468
 469             /* Calculate squared distance and things based on it */
 470             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 471
 472             rinvsq00         = sse41_inv_f(rsq00);
 473
 474             /* Load parameters for j particles */
 475             vdwjidx0A        = 2*vdwtype[jnrA+0];
 476             vdwjidx0B        = 2*vdwtype[jnrB+0];
 477             vdwjidx0C        = 2*vdwtype[jnrC+0];
 478             vdwjidx0D        = 2*vdwtype[jnrD+0];
 479
 480             /**************************
 481              * CALCULATE INTERACTIONS *
 482              **************************/
 483
 484             /* Compute parameters for interactions between i and j atoms */
 485             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 486                                          vdwparam+vdwioffset0+vdwjidx0B,
 487                                          vdwparam+vdwioffset0+vdwjidx0C,
 488                                          vdwparam+vdwioffset0+vdwjidx0D,
 489                                          &c6_00,&c12_00);
 490
 491             /* LENNARD-JONES DISPERSION/REPULSION */
 492
 493             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 494             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 495
 496             fscal            = fvdw;
 497
 498             /* Calculate temporary vectorial force */
 499             tx               = _mm_mul_ps(fscal,dx00);
 500             ty               = _mm_mul_ps(fscal,dy00);
 501             tz               = _mm_mul_ps(fscal,dz00);
 502
 503             /* Update vectorial force */
 504             fix0             = _mm_add_ps(fix0,tx);
 505             fiy0             = _mm_add_ps(fiy0,ty);
 506             fiz0             = _mm_add_ps(fiz0,tz);
 507
 508             fjptrA             = f+j_coord_offsetA;
 509             fjptrB             = f+j_coord_offsetB;
 510             fjptrC             = f+j_coord_offsetC;
 511             fjptrD             = f+j_coord_offsetD;
 512             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 513
 514             /* Inner loop uses 27 flops */
 515         }
 516
 517         if(jidx<j_index_end)
 518         {
 519
 520             /* Get j neighbor index, and coordinate index */
 521             jnrlistA         = jjnr[jidx];
 522             jnrlistB         = jjnr[jidx+1];
 523             jnrlistC         = jjnr[jidx+2];
 524             jnrlistD         = jjnr[jidx+3];
 525             /* Sign of each element will be negative for non-real atoms.
 526              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 527              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 528              */
 529             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 530             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 531             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 532             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 533             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 534             j_coord_offsetA  = DIM*jnrA;
 535             j_coord_offsetB  = DIM*jnrB;
 536             j_coord_offsetC  = DIM*jnrC;
 537             j_coord_offsetD  = DIM*jnrD;
 538
 539             /* load j atom coordinates */
 540             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 541                                               x+j_coord_offsetC,x+j_coord_offsetD,
 542                                               &jx0,&jy0,&jz0);
 543
 544             /* Calculate displacement vector */
 545             dx00             = _mm_sub_ps(ix0,jx0);
 546             dy00             = _mm_sub_ps(iy0,jy0);
 547             dz00             = _mm_sub_ps(iz0,jz0);
 548
 549             /* Calculate squared distance and things based on it */
 550             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 551
 552             rinvsq00         = sse41_inv_f(rsq00);
 553
 554             /* Load parameters for j particles */
 555             vdwjidx0A        = 2*vdwtype[jnrA+0];
 556             vdwjidx0B        = 2*vdwtype[jnrB+0];
 557             vdwjidx0C        = 2*vdwtype[jnrC+0];
 558             vdwjidx0D        = 2*vdwtype[jnrD+0];
 559
 560             /**************************
 561              * CALCULATE INTERACTIONS *
 562              **************************/
 563
 564             /* Compute parameters for interactions between i and j atoms */
 565             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 566                                          vdwparam+vdwioffset0+vdwjidx0B,
 567                                          vdwparam+vdwioffset0+vdwjidx0C,
 568                                          vdwparam+vdwioffset0+vdwjidx0D,
 569                                          &c6_00,&c12_00);
 570
 571             /* LENNARD-JONES DISPERSION/REPULSION */
 572
 573             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 574             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 575
 576             fscal            = fvdw;
 577
 578             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 579
 580             /* Calculate temporary vectorial force */
 581             tx               = _mm_mul_ps(fscal,dx00);
 582             ty               = _mm_mul_ps(fscal,dy00);
 583             tz               = _mm_mul_ps(fscal,dz00);
 584
 585             /* Update vectorial force */
 586             fix0             = _mm_add_ps(fix0,tx);
 587             fiy0             = _mm_add_ps(fiy0,ty);
 588             fiz0             = _mm_add_ps(fiz0,tz);
 589
 590             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 591             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 592             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 593             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 594             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 595
 596             /* Inner loop uses 27 flops */
 597         }
 598
 599         /* End of innermost loop */
 600
 601         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 602                                               f+i_coord_offset,fshift+i_shift_offset);
 603
 604         /* Increment number of inner iterations */
 605         inneriter                  += j_index_end - j_index_start;
 606
 607         /* Outer loop uses 6 flops */
 608     }
 609
 610     /* Increment number of outer iterations */
 611     outeriter        += nri;
 612
 613     /* Update outer/inner flops */
 614
 615     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*27);
 616 }