src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecNone_VdwLJSh_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_VdwLJSh_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_VdwLJSh_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     rcutoff_scalar   = fr->rvdw;
 114     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 115     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 116
 117     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 118     rvdw             = _mm_set1_ps(fr->rvdw);
 119
 120     /* Avoid stupid compiler warnings */
 121     jnrA = jnrB = jnrC = jnrD = 0;
 122     j_coord_offsetA = 0;
 123     j_coord_offsetB = 0;
 124     j_coord_offsetC = 0;
 125     j_coord_offsetD = 0;
 126
 127     outeriter        = 0;
 128     inneriter        = 0;
 129
 130     for(iidx=0;iidx<4*DIM;iidx++)
 131     {
 132         scratch[iidx] = 0.0;
 133     }
 134
 135     /* Start outer loop over neighborlists */
 136     for(iidx=0; iidx<nri; iidx++)
 137     {
 138         /* Load shift vector for this list */
 139         i_shift_offset   = DIM*shiftidx[iidx];
 140
 141         /* Load limits for loop over neighbors */
 142         j_index_start    = jindex[iidx];
 143         j_index_end      = jindex[iidx+1];
 144
 145         /* Get outer coordinate index */
 146         inr              = iinr[iidx];
 147         i_coord_offset   = DIM*inr;
 148
 149         /* Load i particle coords and add shift vector */
 150         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 151
 152         fix0             = _mm_setzero_ps();
 153         fiy0             = _mm_setzero_ps();
 154         fiz0             = _mm_setzero_ps();
 155
 156         /* Load parameters for i particles */
 157         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 158
 159         /* Reset potential sums */
 160         vvdwsum          = _mm_setzero_ps();
 161
 162         /* Start inner kernel loop */
 163         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 164         {
 165
 166             /* Get j neighbor index, and coordinate index */
 167             jnrA             = jjnr[jidx];
 168             jnrB             = jjnr[jidx+1];
 169             jnrC             = jjnr[jidx+2];
 170             jnrD             = jjnr[jidx+3];
 171             j_coord_offsetA  = DIM*jnrA;
 172             j_coord_offsetB  = DIM*jnrB;
 173             j_coord_offsetC  = DIM*jnrC;
 174             j_coord_offsetD  = DIM*jnrD;
 175
 176             /* load j atom coordinates */
 177             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 178                                               x+j_coord_offsetC,x+j_coord_offsetD,
 179                                               &jx0,&jy0,&jz0);
 180
 181             /* Calculate displacement vector */
 182             dx00             = _mm_sub_ps(ix0,jx0);
 183             dy00             = _mm_sub_ps(iy0,jy0);
 184             dz00             = _mm_sub_ps(iz0,jz0);
 185
 186             /* Calculate squared distance and things based on it */
 187             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 188
 189             rinvsq00         = gmx_mm_inv_ps(rsq00);
 190
 191             /* Load parameters for j particles */
 192             vdwjidx0A        = 2*vdwtype[jnrA+0];
 193             vdwjidx0B        = 2*vdwtype[jnrB+0];
 194             vdwjidx0C        = 2*vdwtype[jnrC+0];
 195             vdwjidx0D        = 2*vdwtype[jnrD+0];
 196
 197             /**************************
 198              * CALCULATE INTERACTIONS *
 199              **************************/
 200
 201             if (gmx_mm_any_lt(rsq00,rcutoff2))
 202             {
 203
 204             /* Compute parameters for interactions between i and j atoms */
 205             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 206                                          vdwparam+vdwioffset0+vdwjidx0B,
 207                                          vdwparam+vdwioffset0+vdwjidx0C,
 208                                          vdwparam+vdwioffset0+vdwjidx0D,
 209                                          &c6_00,&c12_00);
 210
 211             /* LENNARD-JONES DISPERSION/REPULSION */
 212
 213             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 214             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 215             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 216             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 217                                           _mm_mul_ps( _mm_nmacc_ps(c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 218             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 219
 220             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 221
 222             /* Update potential sum for this i atom from the interaction with this j atom. */
 223             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 224             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 225
 226             fscal            = fvdw;
 227
 228             fscal            = _mm_and_ps(fscal,cutoff_mask);
 229
 230              /* Update vectorial force */
 231             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 232             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 233             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 234
 235             fjptrA             = f+j_coord_offsetA;
 236             fjptrB             = f+j_coord_offsetB;
 237             fjptrC             = f+j_coord_offsetC;
 238             fjptrD             = f+j_coord_offsetD;
 239             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 240                                                    _mm_mul_ps(dx00,fscal),
 241                                                    _mm_mul_ps(dy00,fscal),
 242                                                    _mm_mul_ps(dz00,fscal));
 243
 244             }
 245
 246             /* Inner loop uses 44 flops */
 247         }
 248
 249         if(jidx<j_index_end)
 250         {
 251
 252             /* Get j neighbor index, and coordinate index */
 253             jnrlistA         = jjnr[jidx];
 254             jnrlistB         = jjnr[jidx+1];
 255             jnrlistC         = jjnr[jidx+2];
 256             jnrlistD         = jjnr[jidx+3];
 257             /* Sign of each element will be negative for non-real atoms.
 258              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 259              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 260              */
 261             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 262             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 263             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 264             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 265             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 266             j_coord_offsetA  = DIM*jnrA;
 267             j_coord_offsetB  = DIM*jnrB;
 268             j_coord_offsetC  = DIM*jnrC;
 269             j_coord_offsetD  = DIM*jnrD;
 270
 271             /* load j atom coordinates */
 272             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 273                                               x+j_coord_offsetC,x+j_coord_offsetD,
 274                                               &jx0,&jy0,&jz0);
 275
 276             /* Calculate displacement vector */
 277             dx00             = _mm_sub_ps(ix0,jx0);
 278             dy00             = _mm_sub_ps(iy0,jy0);
 279             dz00             = _mm_sub_ps(iz0,jz0);
 280
 281             /* Calculate squared distance and things based on it */
 282             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 283
 284             rinvsq00         = gmx_mm_inv_ps(rsq00);
 285
 286             /* Load parameters for j particles */
 287             vdwjidx0A        = 2*vdwtype[jnrA+0];
 288             vdwjidx0B        = 2*vdwtype[jnrB+0];
 289             vdwjidx0C        = 2*vdwtype[jnrC+0];
 290             vdwjidx0D        = 2*vdwtype[jnrD+0];
 291
 292             /**************************
 293              * CALCULATE INTERACTIONS *
 294              **************************/
 295
 296             if (gmx_mm_any_lt(rsq00,rcutoff2))
 297             {
 298
 299             /* Compute parameters for interactions between i and j atoms */
 300             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 301                                          vdwparam+vdwioffset0+vdwjidx0B,
 302                                          vdwparam+vdwioffset0+vdwjidx0C,
 303                                          vdwparam+vdwioffset0+vdwjidx0D,
 304                                          &c6_00,&c12_00);
 305
 306             /* LENNARD-JONES DISPERSION/REPULSION */
 307
 308             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 309             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 310             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 311             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 312                                           _mm_mul_ps( _mm_nmacc_ps(c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 313             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 314
 315             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 316
 317             /* Update potential sum for this i atom from the interaction with this j atom. */
 318             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 319             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 320             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 321
 322             fscal            = fvdw;
 323
 324             fscal            = _mm_and_ps(fscal,cutoff_mask);
 325
 326             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 327
 328              /* Update vectorial force */
 329             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 330             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 331             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 332
 333             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 334             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 335             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 336             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 337             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 338                                                    _mm_mul_ps(dx00,fscal),
 339                                                    _mm_mul_ps(dy00,fscal),
 340                                                    _mm_mul_ps(dz00,fscal));
 341
 342             }
 343
 344             /* Inner loop uses 44 flops */
 345         }
 346
 347         /* End of innermost loop */
 348
 349         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 350                                               f+i_coord_offset,fshift+i_shift_offset);
 351
 352         ggid                        = gid[iidx];
 353         /* Update potential energies */
 354         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 355
 356         /* Increment number of inner iterations */
 357         inneriter                  += j_index_end - j_index_start;
 358
 359         /* Outer loop uses 7 flops */
 360     }
 361
 362     /* Increment number of outer iterations */
 363     outeriter        += nri;
 364
 365     /* Update outer/inner flops */
 366
 367     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*44);
 368 }
 369 /*
 370  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSh_GeomP1P1_F_avx_128_fma_single
 371  * Electrostatics interaction: None
 372  * VdW interaction:            LennardJones
 373  * Geometry:                   Particle-Particle
 374  * Calculate force/pot:        Force
 375  */
 376 void
 377 nb_kernel_ElecNone_VdwLJSh_GeomP1P1_F_avx_128_fma_single
 378                     (t_nblist                    * gmx_restrict       nlist,
 379                      rvec                        * gmx_restrict          xx,
 380                      rvec                        * gmx_restrict          ff,
 381                      t_forcerec                  * gmx_restrict          fr,
 382                      t_mdatoms                   * gmx_restrict     mdatoms,
 383                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 384                      t_nrnb                      * gmx_restrict        nrnb)
 385 {
 386     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 387      * just 0 for non-waters.
 388      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 389      * jnr indices corresponding to data put in the four positions in the SIMD register.
 390      */
 391     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 392     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 393     int              jnrA,jnrB,jnrC,jnrD;
 394     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 395     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 396     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 397     real             rcutoff_scalar;
 398     real             *shiftvec,*fshift,*x,*f;
 399     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 400     real             scratch[4*DIM];
 401     __m128           fscal,rcutoff,rcutoff2,jidxall;
 402     int              vdwioffset0;
 403     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 404     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 405     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 406     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 407     int              nvdwtype;
 408     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 409     int              *vdwtype;
 410     real             *vdwparam;
 411     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 412     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 413     __m128           dummy_mask,cutoff_mask;
 414     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 415     __m128           one     = _mm_set1_ps(1.0);
 416     __m128           two     = _mm_set1_ps(2.0);
 417     x                = xx[0];
 418     f                = ff[0];
 419
 420     nri              = nlist->nri;
 421     iinr             = nlist->iinr;
 422     jindex           = nlist->jindex;
 423     jjnr             = nlist->jjnr;
 424     shiftidx         = nlist->shift;
 425     gid              = nlist->gid;
 426     shiftvec         = fr->shift_vec[0];
 427     fshift           = fr->fshift[0];
 428     nvdwtype         = fr->ntype;
 429     vdwparam         = fr->nbfp;
 430     vdwtype          = mdatoms->typeA;
 431
 432     rcutoff_scalar   = fr->rvdw;
 433     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 434     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 435
 436     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 437     rvdw             = _mm_set1_ps(fr->rvdw);
 438
 439     /* Avoid stupid compiler warnings */
 440     jnrA = jnrB = jnrC = jnrD = 0;
 441     j_coord_offsetA = 0;
 442     j_coord_offsetB = 0;
 443     j_coord_offsetC = 0;
 444     j_coord_offsetD = 0;
 445
 446     outeriter        = 0;
 447     inneriter        = 0;
 448
 449     for(iidx=0;iidx<4*DIM;iidx++)
 450     {
 451         scratch[iidx] = 0.0;
 452     }
 453
 454     /* Start outer loop over neighborlists */
 455     for(iidx=0; iidx<nri; iidx++)
 456     {
 457         /* Load shift vector for this list */
 458         i_shift_offset   = DIM*shiftidx[iidx];
 459
 460         /* Load limits for loop over neighbors */
 461         j_index_start    = jindex[iidx];
 462         j_index_end      = jindex[iidx+1];
 463
 464         /* Get outer coordinate index */
 465         inr              = iinr[iidx];
 466         i_coord_offset   = DIM*inr;
 467
 468         /* Load i particle coords and add shift vector */
 469         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 470
 471         fix0             = _mm_setzero_ps();
 472         fiy0             = _mm_setzero_ps();
 473         fiz0             = _mm_setzero_ps();
 474
 475         /* Load parameters for i particles */
 476         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 477
 478         /* Start inner kernel loop */
 479         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 480         {
 481
 482             /* Get j neighbor index, and coordinate index */
 483             jnrA             = jjnr[jidx];
 484             jnrB             = jjnr[jidx+1];
 485             jnrC             = jjnr[jidx+2];
 486             jnrD             = jjnr[jidx+3];
 487             j_coord_offsetA  = DIM*jnrA;
 488             j_coord_offsetB  = DIM*jnrB;
 489             j_coord_offsetC  = DIM*jnrC;
 490             j_coord_offsetD  = DIM*jnrD;
 491
 492             /* load j atom coordinates */
 493             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 494                                               x+j_coord_offsetC,x+j_coord_offsetD,
 495                                               &jx0,&jy0,&jz0);
 496
 497             /* Calculate displacement vector */
 498             dx00             = _mm_sub_ps(ix0,jx0);
 499             dy00             = _mm_sub_ps(iy0,jy0);
 500             dz00             = _mm_sub_ps(iz0,jz0);
 501
 502             /* Calculate squared distance and things based on it */
 503             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 504
 505             rinvsq00         = gmx_mm_inv_ps(rsq00);
 506
 507             /* Load parameters for j particles */
 508             vdwjidx0A        = 2*vdwtype[jnrA+0];
 509             vdwjidx0B        = 2*vdwtype[jnrB+0];
 510             vdwjidx0C        = 2*vdwtype[jnrC+0];
 511             vdwjidx0D        = 2*vdwtype[jnrD+0];
 512
 513             /**************************
 514              * CALCULATE INTERACTIONS *
 515              **************************/
 516
 517             if (gmx_mm_any_lt(rsq00,rcutoff2))
 518             {
 519
 520             /* Compute parameters for interactions between i and j atoms */
 521             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 522                                          vdwparam+vdwioffset0+vdwjidx0B,
 523                                          vdwparam+vdwioffset0+vdwjidx0C,
 524                                          vdwparam+vdwioffset0+vdwjidx0D,
 525                                          &c6_00,&c12_00);
 526
 527             /* LENNARD-JONES DISPERSION/REPULSION */
 528
 529             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 530             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 531
 532             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 533
 534             fscal            = fvdw;
 535
 536             fscal            = _mm_and_ps(fscal,cutoff_mask);
 537
 538              /* Update vectorial force */
 539             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 540             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 541             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 542
 543             fjptrA             = f+j_coord_offsetA;
 544             fjptrB             = f+j_coord_offsetB;
 545             fjptrC             = f+j_coord_offsetC;
 546             fjptrD             = f+j_coord_offsetD;
 547             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 548                                                    _mm_mul_ps(dx00,fscal),
 549                                                    _mm_mul_ps(dy00,fscal),
 550                                                    _mm_mul_ps(dz00,fscal));
 551
 552             }
 553
 554             /* Inner loop uses 33 flops */
 555         }
 556
 557         if(jidx<j_index_end)
 558         {
 559
 560             /* Get j neighbor index, and coordinate index */
 561             jnrlistA         = jjnr[jidx];
 562             jnrlistB         = jjnr[jidx+1];
 563             jnrlistC         = jjnr[jidx+2];
 564             jnrlistD         = jjnr[jidx+3];
 565             /* Sign of each element will be negative for non-real atoms.
 566              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 567              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 568              */
 569             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 570             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 571             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 572             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 573             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 574             j_coord_offsetA  = DIM*jnrA;
 575             j_coord_offsetB  = DIM*jnrB;
 576             j_coord_offsetC  = DIM*jnrC;
 577             j_coord_offsetD  = DIM*jnrD;
 578
 579             /* load j atom coordinates */
 580             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 581                                               x+j_coord_offsetC,x+j_coord_offsetD,
 582                                               &jx0,&jy0,&jz0);
 583
 584             /* Calculate displacement vector */
 585             dx00             = _mm_sub_ps(ix0,jx0);
 586             dy00             = _mm_sub_ps(iy0,jy0);
 587             dz00             = _mm_sub_ps(iz0,jz0);
 588
 589             /* Calculate squared distance and things based on it */
 590             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 591
 592             rinvsq00         = gmx_mm_inv_ps(rsq00);
 593
 594             /* Load parameters for j particles */
 595             vdwjidx0A        = 2*vdwtype[jnrA+0];
 596             vdwjidx0B        = 2*vdwtype[jnrB+0];
 597             vdwjidx0C        = 2*vdwtype[jnrC+0];
 598             vdwjidx0D        = 2*vdwtype[jnrD+0];
 599
 600             /**************************
 601              * CALCULATE INTERACTIONS *
 602              **************************/
 603
 604             if (gmx_mm_any_lt(rsq00,rcutoff2))
 605             {
 606
 607             /* Compute parameters for interactions between i and j atoms */
 608             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 609                                          vdwparam+vdwioffset0+vdwjidx0B,
 610                                          vdwparam+vdwioffset0+vdwjidx0C,
 611                                          vdwparam+vdwioffset0+vdwjidx0D,
 612                                          &c6_00,&c12_00);
 613
 614             /* LENNARD-JONES DISPERSION/REPULSION */
 615
 616             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 617             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 618
 619             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 620
 621             fscal            = fvdw;
 622
 623             fscal            = _mm_and_ps(fscal,cutoff_mask);
 624
 625             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 626
 627              /* Update vectorial force */
 628             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 629             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 630             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 631
 632             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 633             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 634             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 635             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 636             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 637                                                    _mm_mul_ps(dx00,fscal),
 638                                                    _mm_mul_ps(dy00,fscal),
 639                                                    _mm_mul_ps(dz00,fscal));
 640
 641             }
 642
 643             /* Inner loop uses 33 flops */
 644         }
 645
 646         /* End of innermost loop */
 647
 648         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 649                                               f+i_coord_offset,fshift+i_shift_offset);
 650
 651         /* Increment number of inner iterations */
 652         inneriter                  += j_index_end - j_index_start;
 653
 654         /* Outer loop uses 6 flops */
 655     }
 656
 657     /* Increment number of outer iterations */
 658     outeriter        += nri;
 659
 660     /* Update outer/inner flops */
 661
 662     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*33);
 663 }