src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecNone_VdwLJSh_GeomP1P1_sse2_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_single.h"
  48 #include "kernelutil_x86_sse2_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSh_GeomP1P1_VF_sse2_single
  52  * Electrostatics interaction: None
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecNone_VdwLJSh_GeomP1P1_VF_sse2_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  86     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     int              nvdwtype;
  89     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  90     int              *vdwtype;
  91     real             *vdwparam;
  92     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  93     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  94     __m128           dummy_mask,cutoff_mask;
  95     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  96     __m128           one     = _mm_set1_ps(1.0);
  97     __m128           two     = _mm_set1_ps(2.0);
  98     x                = xx[0];
  99     f                = ff[0];
 100
 101     nri              = nlist->nri;
 102     iinr             = nlist->iinr;
 103     jindex           = nlist->jindex;
 104     jjnr             = nlist->jjnr;
 105     shiftidx         = nlist->shift;
 106     gid              = nlist->gid;
 107     shiftvec         = fr->shift_vec[0];
 108     fshift           = fr->fshift[0];
 109     nvdwtype         = fr->ntype;
 110     vdwparam         = fr->nbfp;
 111     vdwtype          = mdatoms->typeA;
 112
 113     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_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 217                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),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             /* Calculate temporary vectorial force */
 231             tx               = _mm_mul_ps(fscal,dx00);
 232             ty               = _mm_mul_ps(fscal,dy00);
 233             tz               = _mm_mul_ps(fscal,dz00);
 234
 235             /* Update vectorial force */
 236             fix0             = _mm_add_ps(fix0,tx);
 237             fiy0             = _mm_add_ps(fiy0,ty);
 238             fiz0             = _mm_add_ps(fiz0,tz);
 239
 240             fjptrA             = f+j_coord_offsetA;
 241             fjptrB             = f+j_coord_offsetB;
 242             fjptrC             = f+j_coord_offsetC;
 243             fjptrD             = f+j_coord_offsetD;
 244             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 245
 246             }
 247
 248             /* Inner loop uses 41 flops */
 249         }
 250
 251         if(jidx<j_index_end)
 252         {
 253
 254             /* Get j neighbor index, and coordinate index */
 255             jnrlistA         = jjnr[jidx];
 256             jnrlistB         = jjnr[jidx+1];
 257             jnrlistC         = jjnr[jidx+2];
 258             jnrlistD         = jjnr[jidx+3];
 259             /* Sign of each element will be negative for non-real atoms.
 260              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 261              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 262              */
 263             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 264             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 265             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 266             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 267             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 268             j_coord_offsetA  = DIM*jnrA;
 269             j_coord_offsetB  = DIM*jnrB;
 270             j_coord_offsetC  = DIM*jnrC;
 271             j_coord_offsetD  = DIM*jnrD;
 272
 273             /* load j atom coordinates */
 274             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 275                                               x+j_coord_offsetC,x+j_coord_offsetD,
 276                                               &jx0,&jy0,&jz0);
 277
 278             /* Calculate displacement vector */
 279             dx00             = _mm_sub_ps(ix0,jx0);
 280             dy00             = _mm_sub_ps(iy0,jy0);
 281             dz00             = _mm_sub_ps(iz0,jz0);
 282
 283             /* Calculate squared distance and things based on it */
 284             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 285
 286             rinvsq00         = gmx_mm_inv_ps(rsq00);
 287
 288             /* Load parameters for j particles */
 289             vdwjidx0A        = 2*vdwtype[jnrA+0];
 290             vdwjidx0B        = 2*vdwtype[jnrB+0];
 291             vdwjidx0C        = 2*vdwtype[jnrC+0];
 292             vdwjidx0D        = 2*vdwtype[jnrD+0];
 293
 294             /**************************
 295              * CALCULATE INTERACTIONS *
 296              **************************/
 297
 298             if (gmx_mm_any_lt(rsq00,rcutoff2))
 299             {
 300
 301             /* Compute parameters for interactions between i and j atoms */
 302             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 303                                          vdwparam+vdwioffset0+vdwjidx0B,
 304                                          vdwparam+vdwioffset0+vdwjidx0C,
 305                                          vdwparam+vdwioffset0+vdwjidx0D,
 306                                          &c6_00,&c12_00);
 307
 308             /* LENNARD-JONES DISPERSION/REPULSION */
 309
 310             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 311             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 312             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 313             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 314                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 315             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 316
 317             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 318
 319             /* Update potential sum for this i atom from the interaction with this j atom. */
 320             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 321             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 322             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 323
 324             fscal            = fvdw;
 325
 326             fscal            = _mm_and_ps(fscal,cutoff_mask);
 327
 328             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 329
 330             /* Calculate temporary vectorial force */
 331             tx               = _mm_mul_ps(fscal,dx00);
 332             ty               = _mm_mul_ps(fscal,dy00);
 333             tz               = _mm_mul_ps(fscal,dz00);
 334
 335             /* Update vectorial force */
 336             fix0             = _mm_add_ps(fix0,tx);
 337             fiy0             = _mm_add_ps(fiy0,ty);
 338             fiz0             = _mm_add_ps(fiz0,tz);
 339
 340             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 341             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 342             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 343             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 344             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 345
 346             }
 347
 348             /* Inner loop uses 41 flops */
 349         }
 350
 351         /* End of innermost loop */
 352
 353         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 354                                               f+i_coord_offset,fshift+i_shift_offset);
 355
 356         ggid                        = gid[iidx];
 357         /* Update potential energies */
 358         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 359
 360         /* Increment number of inner iterations */
 361         inneriter                  += j_index_end - j_index_start;
 362
 363         /* Outer loop uses 7 flops */
 364     }
 365
 366     /* Increment number of outer iterations */
 367     outeriter        += nri;
 368
 369     /* Update outer/inner flops */
 370
 371     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*41);
 372 }
 373 /*
 374  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSh_GeomP1P1_F_sse2_single
 375  * Electrostatics interaction: None
 376  * VdW interaction:            LennardJones
 377  * Geometry:                   Particle-Particle
 378  * Calculate force/pot:        Force
 379  */
 380 void
 381 nb_kernel_ElecNone_VdwLJSh_GeomP1P1_F_sse2_single
 382                     (t_nblist                    * gmx_restrict       nlist,
 383                      rvec                        * gmx_restrict          xx,
 384                      rvec                        * gmx_restrict          ff,
 385                      t_forcerec                  * gmx_restrict          fr,
 386                      t_mdatoms                   * gmx_restrict     mdatoms,
 387                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 388                      t_nrnb                      * gmx_restrict        nrnb)
 389 {
 390     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 391      * just 0 for non-waters.
 392      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 393      * jnr indices corresponding to data put in the four positions in the SIMD register.
 394      */
 395     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 396     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 397     int              jnrA,jnrB,jnrC,jnrD;
 398     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 399     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 400     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 401     real             rcutoff_scalar;
 402     real             *shiftvec,*fshift,*x,*f;
 403     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 404     real             scratch[4*DIM];
 405     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 406     int              vdwioffset0;
 407     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 408     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 409     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 410     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 411     int              nvdwtype;
 412     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 413     int              *vdwtype;
 414     real             *vdwparam;
 415     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 416     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 417     __m128           dummy_mask,cutoff_mask;
 418     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 419     __m128           one     = _mm_set1_ps(1.0);
 420     __m128           two     = _mm_set1_ps(2.0);
 421     x                = xx[0];
 422     f                = ff[0];
 423
 424     nri              = nlist->nri;
 425     iinr             = nlist->iinr;
 426     jindex           = nlist->jindex;
 427     jjnr             = nlist->jjnr;
 428     shiftidx         = nlist->shift;
 429     gid              = nlist->gid;
 430     shiftvec         = fr->shift_vec[0];
 431     fshift           = fr->fshift[0];
 432     nvdwtype         = fr->ntype;
 433     vdwparam         = fr->nbfp;
 434     vdwtype          = mdatoms->typeA;
 435
 436     rcutoff_scalar   = fr->rvdw;
 437     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 438     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 439
 440     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 441     rvdw             = _mm_set1_ps(fr->rvdw);
 442
 443     /* Avoid stupid compiler warnings */
 444     jnrA = jnrB = jnrC = jnrD = 0;
 445     j_coord_offsetA = 0;
 446     j_coord_offsetB = 0;
 447     j_coord_offsetC = 0;
 448     j_coord_offsetD = 0;
 449
 450     outeriter        = 0;
 451     inneriter        = 0;
 452
 453     for(iidx=0;iidx<4*DIM;iidx++)
 454     {
 455         scratch[iidx] = 0.0;
 456     }
 457
 458     /* Start outer loop over neighborlists */
 459     for(iidx=0; iidx<nri; iidx++)
 460     {
 461         /* Load shift vector for this list */
 462         i_shift_offset   = DIM*shiftidx[iidx];
 463
 464         /* Load limits for loop over neighbors */
 465         j_index_start    = jindex[iidx];
 466         j_index_end      = jindex[iidx+1];
 467
 468         /* Get outer coordinate index */
 469         inr              = iinr[iidx];
 470         i_coord_offset   = DIM*inr;
 471
 472         /* Load i particle coords and add shift vector */
 473         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 474
 475         fix0             = _mm_setzero_ps();
 476         fiy0             = _mm_setzero_ps();
 477         fiz0             = _mm_setzero_ps();
 478
 479         /* Load parameters for i particles */
 480         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 481
 482         /* Start inner kernel loop */
 483         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 484         {
 485
 486             /* Get j neighbor index, and coordinate index */
 487             jnrA             = jjnr[jidx];
 488             jnrB             = jjnr[jidx+1];
 489             jnrC             = jjnr[jidx+2];
 490             jnrD             = jjnr[jidx+3];
 491             j_coord_offsetA  = DIM*jnrA;
 492             j_coord_offsetB  = DIM*jnrB;
 493             j_coord_offsetC  = DIM*jnrC;
 494             j_coord_offsetD  = DIM*jnrD;
 495
 496             /* load j atom coordinates */
 497             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 498                                               x+j_coord_offsetC,x+j_coord_offsetD,
 499                                               &jx0,&jy0,&jz0);
 500
 501             /* Calculate displacement vector */
 502             dx00             = _mm_sub_ps(ix0,jx0);
 503             dy00             = _mm_sub_ps(iy0,jy0);
 504             dz00             = _mm_sub_ps(iz0,jz0);
 505
 506             /* Calculate squared distance and things based on it */
 507             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 508
 509             rinvsq00         = gmx_mm_inv_ps(rsq00);
 510
 511             /* Load parameters for j particles */
 512             vdwjidx0A        = 2*vdwtype[jnrA+0];
 513             vdwjidx0B        = 2*vdwtype[jnrB+0];
 514             vdwjidx0C        = 2*vdwtype[jnrC+0];
 515             vdwjidx0D        = 2*vdwtype[jnrD+0];
 516
 517             /**************************
 518              * CALCULATE INTERACTIONS *
 519              **************************/
 520
 521             if (gmx_mm_any_lt(rsq00,rcutoff2))
 522             {
 523
 524             /* Compute parameters for interactions between i and j atoms */
 525             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 526                                          vdwparam+vdwioffset0+vdwjidx0B,
 527                                          vdwparam+vdwioffset0+vdwjidx0C,
 528                                          vdwparam+vdwioffset0+vdwjidx0D,
 529                                          &c6_00,&c12_00);
 530
 531             /* LENNARD-JONES DISPERSION/REPULSION */
 532
 533             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 534             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 535
 536             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 537
 538             fscal            = fvdw;
 539
 540             fscal            = _mm_and_ps(fscal,cutoff_mask);
 541
 542             /* Calculate temporary vectorial force */
 543             tx               = _mm_mul_ps(fscal,dx00);
 544             ty               = _mm_mul_ps(fscal,dy00);
 545             tz               = _mm_mul_ps(fscal,dz00);
 546
 547             /* Update vectorial force */
 548             fix0             = _mm_add_ps(fix0,tx);
 549             fiy0             = _mm_add_ps(fiy0,ty);
 550             fiz0             = _mm_add_ps(fiz0,tz);
 551
 552             fjptrA             = f+j_coord_offsetA;
 553             fjptrB             = f+j_coord_offsetB;
 554             fjptrC             = f+j_coord_offsetC;
 555             fjptrD             = f+j_coord_offsetD;
 556             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 557
 558             }
 559
 560             /* Inner loop uses 30 flops */
 561         }
 562
 563         if(jidx<j_index_end)
 564         {
 565
 566             /* Get j neighbor index, and coordinate index */
 567             jnrlistA         = jjnr[jidx];
 568             jnrlistB         = jjnr[jidx+1];
 569             jnrlistC         = jjnr[jidx+2];
 570             jnrlistD         = jjnr[jidx+3];
 571             /* Sign of each element will be negative for non-real atoms.
 572              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 573              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 574              */
 575             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 576             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 577             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 578             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 579             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 580             j_coord_offsetA  = DIM*jnrA;
 581             j_coord_offsetB  = DIM*jnrB;
 582             j_coord_offsetC  = DIM*jnrC;
 583             j_coord_offsetD  = DIM*jnrD;
 584
 585             /* load j atom coordinates */
 586             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 587                                               x+j_coord_offsetC,x+j_coord_offsetD,
 588                                               &jx0,&jy0,&jz0);
 589
 590             /* Calculate displacement vector */
 591             dx00             = _mm_sub_ps(ix0,jx0);
 592             dy00             = _mm_sub_ps(iy0,jy0);
 593             dz00             = _mm_sub_ps(iz0,jz0);
 594
 595             /* Calculate squared distance and things based on it */
 596             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 597
 598             rinvsq00         = gmx_mm_inv_ps(rsq00);
 599
 600             /* Load parameters for j particles */
 601             vdwjidx0A        = 2*vdwtype[jnrA+0];
 602             vdwjidx0B        = 2*vdwtype[jnrB+0];
 603             vdwjidx0C        = 2*vdwtype[jnrC+0];
 604             vdwjidx0D        = 2*vdwtype[jnrD+0];
 605
 606             /**************************
 607              * CALCULATE INTERACTIONS *
 608              **************************/
 609
 610             if (gmx_mm_any_lt(rsq00,rcutoff2))
 611             {
 612
 613             /* Compute parameters for interactions between i and j atoms */
 614             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 615                                          vdwparam+vdwioffset0+vdwjidx0B,
 616                                          vdwparam+vdwioffset0+vdwjidx0C,
 617                                          vdwparam+vdwioffset0+vdwjidx0D,
 618                                          &c6_00,&c12_00);
 619
 620             /* LENNARD-JONES DISPERSION/REPULSION */
 621
 622             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 623             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 624
 625             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 626
 627             fscal            = fvdw;
 628
 629             fscal            = _mm_and_ps(fscal,cutoff_mask);
 630
 631             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 632
 633             /* Calculate temporary vectorial force */
 634             tx               = _mm_mul_ps(fscal,dx00);
 635             ty               = _mm_mul_ps(fscal,dy00);
 636             tz               = _mm_mul_ps(fscal,dz00);
 637
 638             /* Update vectorial force */
 639             fix0             = _mm_add_ps(fix0,tx);
 640             fiy0             = _mm_add_ps(fiy0,ty);
 641             fiz0             = _mm_add_ps(fiz0,tz);
 642
 643             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 644             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 645             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 646             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 647             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 648
 649             }
 650
 651             /* Inner loop uses 30 flops */
 652         }
 653
 654         /* End of innermost loop */
 655
 656         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 657                                               f+i_coord_offset,fshift+i_shift_offset);
 658
 659         /* Increment number of inner iterations */
 660         inneriter                  += j_index_end - j_index_start;
 661
 662         /* Outer loop uses 6 flops */
 663     }
 664
 665     /* Increment number of outer iterations */
 666     outeriter        += nri;
 667
 668     /* Update outer/inner flops */
 669
 670     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*30);
 671 }