src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecNone_VdwLJ_GeomP1P1_avx_128_fma_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 avx_128_fma_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJ_GeomP1P1_VF_avx_128_fma_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_avx_128_fma_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 AVX_128, 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           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_msub_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              /* Update vectorial force */
 215             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 216             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 217             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 218
 219             fjptrA             = f+j_coord_offsetA;
 220             fjptrB             = f+j_coord_offsetB;
 221             fjptrC             = f+j_coord_offsetC;
 222             fjptrD             = f+j_coord_offsetD;
 223             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 224                                                    _mm_mul_ps(dx00,fscal),
 225                                                    _mm_mul_ps(dy00,fscal),
 226                                                    _mm_mul_ps(dz00,fscal));
 227
 228             /* Inner loop uses 35 flops */
 229         }
 230
 231         if(jidx<j_index_end)
 232         {
 233
 234             /* Get j neighbor index, and coordinate index */
 235             jnrlistA         = jjnr[jidx];
 236             jnrlistB         = jjnr[jidx+1];
 237             jnrlistC         = jjnr[jidx+2];
 238             jnrlistD         = jjnr[jidx+3];
 239             /* Sign of each element will be negative for non-real atoms.
 240              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 241              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 242              */
 243             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 244             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 245             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 246             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 247             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 248             j_coord_offsetA  = DIM*jnrA;
 249             j_coord_offsetB  = DIM*jnrB;
 250             j_coord_offsetC  = DIM*jnrC;
 251             j_coord_offsetD  = DIM*jnrD;
 252
 253             /* load j atom coordinates */
 254             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 255                                               x+j_coord_offsetC,x+j_coord_offsetD,
 256                                               &jx0,&jy0,&jz0);
 257
 258             /* Calculate displacement vector */
 259             dx00             = _mm_sub_ps(ix0,jx0);
 260             dy00             = _mm_sub_ps(iy0,jy0);
 261             dz00             = _mm_sub_ps(iz0,jz0);
 262
 263             /* Calculate squared distance and things based on it */
 264             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 265
 266             rinvsq00         = gmx_mm_inv_ps(rsq00);
 267
 268             /* Load parameters for j particles */
 269             vdwjidx0A        = 2*vdwtype[jnrA+0];
 270             vdwjidx0B        = 2*vdwtype[jnrB+0];
 271             vdwjidx0C        = 2*vdwtype[jnrC+0];
 272             vdwjidx0D        = 2*vdwtype[jnrD+0];
 273
 274             /**************************
 275              * CALCULATE INTERACTIONS *
 276              **************************/
 277
 278             /* Compute parameters for interactions between i and j atoms */
 279             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 280                                          vdwparam+vdwioffset0+vdwjidx0B,
 281                                          vdwparam+vdwioffset0+vdwjidx0C,
 282                                          vdwparam+vdwioffset0+vdwjidx0D,
 283                                          &c6_00,&c12_00);
 284
 285             /* LENNARD-JONES DISPERSION/REPULSION */
 286
 287             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 288             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 289             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 290             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 291             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 292
 293             /* Update potential sum for this i atom from the interaction with this j atom. */
 294             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 295             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 296
 297             fscal            = fvdw;
 298
 299             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 300
 301              /* Update vectorial force */
 302             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 303             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 304             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 305
 306             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 307             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 308             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 309             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 310             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 311                                                    _mm_mul_ps(dx00,fscal),
 312                                                    _mm_mul_ps(dy00,fscal),
 313                                                    _mm_mul_ps(dz00,fscal));
 314
 315             /* Inner loop uses 35 flops */
 316         }
 317
 318         /* End of innermost loop */
 319
 320         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 321                                               f+i_coord_offset,fshift+i_shift_offset);
 322
 323         ggid                        = gid[iidx];
 324         /* Update potential energies */
 325         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 326
 327         /* Increment number of inner iterations */
 328         inneriter                  += j_index_end - j_index_start;
 329
 330         /* Outer loop uses 7 flops */
 331     }
 332
 333     /* Increment number of outer iterations */
 334     outeriter        += nri;
 335
 336     /* Update outer/inner flops */
 337
 338     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*35);
 339 }
 340 /*
 341  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJ_GeomP1P1_F_avx_128_fma_single
 342  * Electrostatics interaction: None
 343  * VdW interaction:            LennardJones
 344  * Geometry:                   Particle-Particle
 345  * Calculate force/pot:        Force
 346  */
 347 void
 348 nb_kernel_ElecNone_VdwLJ_GeomP1P1_F_avx_128_fma_single
 349                     (t_nblist                    * gmx_restrict       nlist,
 350                      rvec                        * gmx_restrict          xx,
 351                      rvec                        * gmx_restrict          ff,
 352                      t_forcerec                  * gmx_restrict          fr,
 353                      t_mdatoms                   * gmx_restrict     mdatoms,
 354                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 355                      t_nrnb                      * gmx_restrict        nrnb)
 356 {
 357     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 358      * just 0 for non-waters.
 359      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 360      * jnr indices corresponding to data put in the four positions in the SIMD register.
 361      */
 362     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 363     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 364     int              jnrA,jnrB,jnrC,jnrD;
 365     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 366     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 367     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 368     real             rcutoff_scalar;
 369     real             *shiftvec,*fshift,*x,*f;
 370     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 371     real             scratch[4*DIM];
 372     __m128           fscal,rcutoff,rcutoff2,jidxall;
 373     int              vdwioffset0;
 374     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 375     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 376     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 377     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 378     int              nvdwtype;
 379     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 380     int              *vdwtype;
 381     real             *vdwparam;
 382     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 383     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 384     __m128           dummy_mask,cutoff_mask;
 385     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 386     __m128           one     = _mm_set1_ps(1.0);
 387     __m128           two     = _mm_set1_ps(2.0);
 388     x                = xx[0];
 389     f                = ff[0];
 390
 391     nri              = nlist->nri;
 392     iinr             = nlist->iinr;
 393     jindex           = nlist->jindex;
 394     jjnr             = nlist->jjnr;
 395     shiftidx         = nlist->shift;
 396     gid              = nlist->gid;
 397     shiftvec         = fr->shift_vec[0];
 398     fshift           = fr->fshift[0];
 399     nvdwtype         = fr->ntype;
 400     vdwparam         = fr->nbfp;
 401     vdwtype          = mdatoms->typeA;
 402
 403     /* Avoid stupid compiler warnings */
 404     jnrA = jnrB = jnrC = jnrD = 0;
 405     j_coord_offsetA = 0;
 406     j_coord_offsetB = 0;
 407     j_coord_offsetC = 0;
 408     j_coord_offsetD = 0;
 409
 410     outeriter        = 0;
 411     inneriter        = 0;
 412
 413     for(iidx=0;iidx<4*DIM;iidx++)
 414     {
 415         scratch[iidx] = 0.0;
 416     }
 417
 418     /* Start outer loop over neighborlists */
 419     for(iidx=0; iidx<nri; iidx++)
 420     {
 421         /* Load shift vector for this list */
 422         i_shift_offset   = DIM*shiftidx[iidx];
 423
 424         /* Load limits for loop over neighbors */
 425         j_index_start    = jindex[iidx];
 426         j_index_end      = jindex[iidx+1];
 427
 428         /* Get outer coordinate index */
 429         inr              = iinr[iidx];
 430         i_coord_offset   = DIM*inr;
 431
 432         /* Load i particle coords and add shift vector */
 433         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 434
 435         fix0             = _mm_setzero_ps();
 436         fiy0             = _mm_setzero_ps();
 437         fiz0             = _mm_setzero_ps();
 438
 439         /* Load parameters for i particles */
 440         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 441
 442         /* Start inner kernel loop */
 443         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 444         {
 445
 446             /* Get j neighbor index, and coordinate index */
 447             jnrA             = jjnr[jidx];
 448             jnrB             = jjnr[jidx+1];
 449             jnrC             = jjnr[jidx+2];
 450             jnrD             = jjnr[jidx+3];
 451             j_coord_offsetA  = DIM*jnrA;
 452             j_coord_offsetB  = DIM*jnrB;
 453             j_coord_offsetC  = DIM*jnrC;
 454             j_coord_offsetD  = DIM*jnrD;
 455
 456             /* load j atom coordinates */
 457             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 458                                               x+j_coord_offsetC,x+j_coord_offsetD,
 459                                               &jx0,&jy0,&jz0);
 460
 461             /* Calculate displacement vector */
 462             dx00             = _mm_sub_ps(ix0,jx0);
 463             dy00             = _mm_sub_ps(iy0,jy0);
 464             dz00             = _mm_sub_ps(iz0,jz0);
 465
 466             /* Calculate squared distance and things based on it */
 467             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 468
 469             rinvsq00         = gmx_mm_inv_ps(rsq00);
 470
 471             /* Load parameters for j particles */
 472             vdwjidx0A        = 2*vdwtype[jnrA+0];
 473             vdwjidx0B        = 2*vdwtype[jnrB+0];
 474             vdwjidx0C        = 2*vdwtype[jnrC+0];
 475             vdwjidx0D        = 2*vdwtype[jnrD+0];
 476
 477             /**************************
 478              * CALCULATE INTERACTIONS *
 479              **************************/
 480
 481             /* Compute parameters for interactions between i and j atoms */
 482             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 483                                          vdwparam+vdwioffset0+vdwjidx0B,
 484                                          vdwparam+vdwioffset0+vdwjidx0C,
 485                                          vdwparam+vdwioffset0+vdwjidx0D,
 486                                          &c6_00,&c12_00);
 487
 488             /* LENNARD-JONES DISPERSION/REPULSION */
 489
 490             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 491             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 492
 493             fscal            = fvdw;
 494
 495              /* Update vectorial force */
 496             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 497             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 498             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 499
 500             fjptrA             = f+j_coord_offsetA;
 501             fjptrB             = f+j_coord_offsetB;
 502             fjptrC             = f+j_coord_offsetC;
 503             fjptrD             = f+j_coord_offsetD;
 504             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 505                                                    _mm_mul_ps(dx00,fscal),
 506                                                    _mm_mul_ps(dy00,fscal),
 507                                                    _mm_mul_ps(dz00,fscal));
 508
 509             /* Inner loop uses 30 flops */
 510         }
 511
 512         if(jidx<j_index_end)
 513         {
 514
 515             /* Get j neighbor index, and coordinate index */
 516             jnrlistA         = jjnr[jidx];
 517             jnrlistB         = jjnr[jidx+1];
 518             jnrlistC         = jjnr[jidx+2];
 519             jnrlistD         = jjnr[jidx+3];
 520             /* Sign of each element will be negative for non-real atoms.
 521              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 522              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 523              */
 524             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 525             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 526             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 527             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 528             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 529             j_coord_offsetA  = DIM*jnrA;
 530             j_coord_offsetB  = DIM*jnrB;
 531             j_coord_offsetC  = DIM*jnrC;
 532             j_coord_offsetD  = DIM*jnrD;
 533
 534             /* load j atom coordinates */
 535             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 536                                               x+j_coord_offsetC,x+j_coord_offsetD,
 537                                               &jx0,&jy0,&jz0);
 538
 539             /* Calculate displacement vector */
 540             dx00             = _mm_sub_ps(ix0,jx0);
 541             dy00             = _mm_sub_ps(iy0,jy0);
 542             dz00             = _mm_sub_ps(iz0,jz0);
 543
 544             /* Calculate squared distance and things based on it */
 545             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 546
 547             rinvsq00         = gmx_mm_inv_ps(rsq00);
 548
 549             /* Load parameters for j particles */
 550             vdwjidx0A        = 2*vdwtype[jnrA+0];
 551             vdwjidx0B        = 2*vdwtype[jnrB+0];
 552             vdwjidx0C        = 2*vdwtype[jnrC+0];
 553             vdwjidx0D        = 2*vdwtype[jnrD+0];
 554
 555             /**************************
 556              * CALCULATE INTERACTIONS *
 557              **************************/
 558
 559             /* Compute parameters for interactions between i and j atoms */
 560             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 561                                          vdwparam+vdwioffset0+vdwjidx0B,
 562                                          vdwparam+vdwioffset0+vdwjidx0C,
 563                                          vdwparam+vdwioffset0+vdwjidx0D,
 564                                          &c6_00,&c12_00);
 565
 566             /* LENNARD-JONES DISPERSION/REPULSION */
 567
 568             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 569             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 570
 571             fscal            = fvdw;
 572
 573             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 574
 575              /* Update vectorial force */
 576             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 577             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 578             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 579
 580             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 581             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 582             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 583             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 584             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 585                                                    _mm_mul_ps(dx00,fscal),
 586                                                    _mm_mul_ps(dy00,fscal),
 587                                                    _mm_mul_ps(dz00,fscal));
 588
 589             /* Inner loop uses 30 flops */
 590         }
 591
 592         /* End of innermost loop */
 593
 594         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 595                                               f+i_coord_offset,fshift+i_shift_offset);
 596
 597         /* Increment number of inner iterations */
 598         inneriter                  += j_index_end - j_index_start;
 599
 600         /* Outer loop uses 6 flops */
 601     }
 602
 603     /* Increment number of outer iterations */
 604     outeriter        += nri;
 605
 606     /* Update outer/inner flops */
 607
 608     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*30);
 609 }