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 "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_avx_128_fma_single.h"
  50 #include "kernelutil_x86_avx_128_fma_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJ_GeomP1P1_VF_avx_128_fma_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_avx_128_fma_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 AVX_128, 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           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_msub_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              /* Update vectorial force */
 217             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 218             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 219             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 220
 221             fjptrA             = f+j_coord_offsetA;
 222             fjptrB             = f+j_coord_offsetB;
 223             fjptrC             = f+j_coord_offsetC;
 224             fjptrD             = f+j_coord_offsetD;
 225             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 226                                                    _mm_mul_ps(dx00,fscal),
 227                                                    _mm_mul_ps(dy00,fscal),
 228                                                    _mm_mul_ps(dz00,fscal));
 229
 230             /* Inner loop uses 35 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_msub_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              /* Update vectorial force */
 304             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 305             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 306             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 307
 308             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 309             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 310             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 311             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 312             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 313                                                    _mm_mul_ps(dx00,fscal),
 314                                                    _mm_mul_ps(dy00,fscal),
 315                                                    _mm_mul_ps(dz00,fscal));
 316
 317             /* Inner loop uses 35 flops */
 318         }
 319
 320         /* End of innermost loop */
 321
 322         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 323                                               f+i_coord_offset,fshift+i_shift_offset);
 324
 325         ggid                        = gid[iidx];
 326         /* Update potential energies */
 327         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 328
 329         /* Increment number of inner iterations */
 330         inneriter                  += j_index_end - j_index_start;
 331
 332         /* Outer loop uses 7 flops */
 333     }
 334
 335     /* Increment number of outer iterations */
 336     outeriter        += nri;
 337
 338     /* Update outer/inner flops */
 339
 340     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*35);
 341 }
 342 /*
 343  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJ_GeomP1P1_F_avx_128_fma_single
 344  * Electrostatics interaction: None
 345  * VdW interaction:            LennardJones
 346  * Geometry:                   Particle-Particle
 347  * Calculate force/pot:        Force
 348  */
 349 void
 350 nb_kernel_ElecNone_VdwLJ_GeomP1P1_F_avx_128_fma_single
 351                     (t_nblist                    * gmx_restrict       nlist,
 352                      rvec                        * gmx_restrict          xx,
 353                      rvec                        * gmx_restrict          ff,
 354                      t_forcerec                  * gmx_restrict          fr,
 355                      t_mdatoms                   * gmx_restrict     mdatoms,
 356                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 357                      t_nrnb                      * gmx_restrict        nrnb)
 358 {
 359     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 360      * just 0 for non-waters.
 361      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 362      * jnr indices corresponding to data put in the four positions in the SIMD register.
 363      */
 364     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 365     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 366     int              jnrA,jnrB,jnrC,jnrD;
 367     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 368     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 369     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 370     real             rcutoff_scalar;
 371     real             *shiftvec,*fshift,*x,*f;
 372     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 373     real             scratch[4*DIM];
 374     __m128           fscal,rcutoff,rcutoff2,jidxall;
 375     int              vdwioffset0;
 376     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 377     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 378     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 379     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 380     int              nvdwtype;
 381     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 382     int              *vdwtype;
 383     real             *vdwparam;
 384     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 385     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 386     __m128           dummy_mask,cutoff_mask;
 387     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 388     __m128           one     = _mm_set1_ps(1.0);
 389     __m128           two     = _mm_set1_ps(2.0);
 390     x                = xx[0];
 391     f                = ff[0];
 392
 393     nri              = nlist->nri;
 394     iinr             = nlist->iinr;
 395     jindex           = nlist->jindex;
 396     jjnr             = nlist->jjnr;
 397     shiftidx         = nlist->shift;
 398     gid              = nlist->gid;
 399     shiftvec         = fr->shift_vec[0];
 400     fshift           = fr->fshift[0];
 401     nvdwtype         = fr->ntype;
 402     vdwparam         = fr->nbfp;
 403     vdwtype          = mdatoms->typeA;
 404
 405     /* Avoid stupid compiler warnings */
 406     jnrA = jnrB = jnrC = jnrD = 0;
 407     j_coord_offsetA = 0;
 408     j_coord_offsetB = 0;
 409     j_coord_offsetC = 0;
 410     j_coord_offsetD = 0;
 411
 412     outeriter        = 0;
 413     inneriter        = 0;
 414
 415     for(iidx=0;iidx<4*DIM;iidx++)
 416     {
 417         scratch[iidx] = 0.0;
 418     }
 419
 420     /* Start outer loop over neighborlists */
 421     for(iidx=0; iidx<nri; iidx++)
 422     {
 423         /* Load shift vector for this list */
 424         i_shift_offset   = DIM*shiftidx[iidx];
 425
 426         /* Load limits for loop over neighbors */
 427         j_index_start    = jindex[iidx];
 428         j_index_end      = jindex[iidx+1];
 429
 430         /* Get outer coordinate index */
 431         inr              = iinr[iidx];
 432         i_coord_offset   = DIM*inr;
 433
 434         /* Load i particle coords and add shift vector */
 435         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 436
 437         fix0             = _mm_setzero_ps();
 438         fiy0             = _mm_setzero_ps();
 439         fiz0             = _mm_setzero_ps();
 440
 441         /* Load parameters for i particles */
 442         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 443
 444         /* Start inner kernel loop */
 445         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 446         {
 447
 448             /* Get j neighbor index, and coordinate index */
 449             jnrA             = jjnr[jidx];
 450             jnrB             = jjnr[jidx+1];
 451             jnrC             = jjnr[jidx+2];
 452             jnrD             = jjnr[jidx+3];
 453             j_coord_offsetA  = DIM*jnrA;
 454             j_coord_offsetB  = DIM*jnrB;
 455             j_coord_offsetC  = DIM*jnrC;
 456             j_coord_offsetD  = DIM*jnrD;
 457
 458             /* load j atom coordinates */
 459             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 460                                               x+j_coord_offsetC,x+j_coord_offsetD,
 461                                               &jx0,&jy0,&jz0);
 462
 463             /* Calculate displacement vector */
 464             dx00             = _mm_sub_ps(ix0,jx0);
 465             dy00             = _mm_sub_ps(iy0,jy0);
 466             dz00             = _mm_sub_ps(iz0,jz0);
 467
 468             /* Calculate squared distance and things based on it */
 469             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 470
 471             rinvsq00         = gmx_mm_inv_ps(rsq00);
 472
 473             /* Load parameters for j particles */
 474             vdwjidx0A        = 2*vdwtype[jnrA+0];
 475             vdwjidx0B        = 2*vdwtype[jnrB+0];
 476             vdwjidx0C        = 2*vdwtype[jnrC+0];
 477             vdwjidx0D        = 2*vdwtype[jnrD+0];
 478
 479             /**************************
 480              * CALCULATE INTERACTIONS *
 481              **************************/
 482
 483             /* Compute parameters for interactions between i and j atoms */
 484             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 485                                          vdwparam+vdwioffset0+vdwjidx0B,
 486                                          vdwparam+vdwioffset0+vdwjidx0C,
 487                                          vdwparam+vdwioffset0+vdwjidx0D,
 488                                          &c6_00,&c12_00);
 489
 490             /* LENNARD-JONES DISPERSION/REPULSION */
 491
 492             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 493             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 494
 495             fscal            = fvdw;
 496
 497              /* Update vectorial force */
 498             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 499             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 500             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 501
 502             fjptrA             = f+j_coord_offsetA;
 503             fjptrB             = f+j_coord_offsetB;
 504             fjptrC             = f+j_coord_offsetC;
 505             fjptrD             = f+j_coord_offsetD;
 506             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 507                                                    _mm_mul_ps(dx00,fscal),
 508                                                    _mm_mul_ps(dy00,fscal),
 509                                                    _mm_mul_ps(dz00,fscal));
 510
 511             /* Inner loop uses 30 flops */
 512         }
 513
 514         if(jidx<j_index_end)
 515         {
 516
 517             /* Get j neighbor index, and coordinate index */
 518             jnrlistA         = jjnr[jidx];
 519             jnrlistB         = jjnr[jidx+1];
 520             jnrlistC         = jjnr[jidx+2];
 521             jnrlistD         = jjnr[jidx+3];
 522             /* Sign of each element will be negative for non-real atoms.
 523              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 524              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 525              */
 526             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 527             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 528             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 529             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 530             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 531             j_coord_offsetA  = DIM*jnrA;
 532             j_coord_offsetB  = DIM*jnrB;
 533             j_coord_offsetC  = DIM*jnrC;
 534             j_coord_offsetD  = DIM*jnrD;
 535
 536             /* load j atom coordinates */
 537             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 538                                               x+j_coord_offsetC,x+j_coord_offsetD,
 539                                               &jx0,&jy0,&jz0);
 540
 541             /* Calculate displacement vector */
 542             dx00             = _mm_sub_ps(ix0,jx0);
 543             dy00             = _mm_sub_ps(iy0,jy0);
 544             dz00             = _mm_sub_ps(iz0,jz0);
 545
 546             /* Calculate squared distance and things based on it */
 547             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 548
 549             rinvsq00         = gmx_mm_inv_ps(rsq00);
 550
 551             /* Load parameters for j particles */
 552             vdwjidx0A        = 2*vdwtype[jnrA+0];
 553             vdwjidx0B        = 2*vdwtype[jnrB+0];
 554             vdwjidx0C        = 2*vdwtype[jnrC+0];
 555             vdwjidx0D        = 2*vdwtype[jnrD+0];
 556
 557             /**************************
 558              * CALCULATE INTERACTIONS *
 559              **************************/
 560
 561             /* Compute parameters for interactions between i and j atoms */
 562             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 563                                          vdwparam+vdwioffset0+vdwjidx0B,
 564                                          vdwparam+vdwioffset0+vdwjidx0C,
 565                                          vdwparam+vdwioffset0+vdwjidx0D,
 566                                          &c6_00,&c12_00);
 567
 568             /* LENNARD-JONES DISPERSION/REPULSION */
 569
 570             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 571             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 572
 573             fscal            = fvdw;
 574
 575             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 576
 577              /* Update vectorial force */
 578             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 579             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 580             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 581
 582             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 583             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 584             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 585             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 586             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 587                                                    _mm_mul_ps(dx00,fscal),
 588                                                    _mm_mul_ps(dy00,fscal),
 589                                                    _mm_mul_ps(dz00,fscal));
 590
 591             /* Inner loop uses 30 flops */
 592         }
 593
 594         /* End of innermost loop */
 595
 596         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 597                                               f+i_coord_offset,fshift+i_shift_offset);
 598
 599         /* Increment number of inner iterations */
 600         inneriter                  += j_index_end - j_index_start;
 601
 602         /* Outer loop uses 6 flops */
 603     }
 604
 605     /* Increment number of outer iterations */
 606     outeriter        += nri;
 607
 608     /* Update outer/inner flops */
 609
 610     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*30);
 611 }