src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCoul_VdwLJ_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 "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "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_ElecCoul_VdwLJ_GeomP1P1_VF_sse2_single
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_VdwLJ_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     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  89     real             *charge;
  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     facel            = _mm_set1_ps(fr->epsfac);
 112     charge           = mdatoms->chargeA;
 113     nvdwtype         = fr->ntype;
 114     vdwparam         = fr->nbfp;
 115     vdwtype          = mdatoms->typeA;
 116
 117     /* Avoid stupid compiler warnings */
 118     jnrA = jnrB = jnrC = jnrD = 0;
 119     j_coord_offsetA = 0;
 120     j_coord_offsetB = 0;
 121     j_coord_offsetC = 0;
 122     j_coord_offsetD = 0;
 123
 124     outeriter        = 0;
 125     inneriter        = 0;
 126
 127     for(iidx=0;iidx<4*DIM;iidx++)
 128     {
 129         scratch[iidx] = 0.0;
 130     }
 131
 132     /* Start outer loop over neighborlists */
 133     for(iidx=0; iidx<nri; iidx++)
 134     {
 135         /* Load shift vector for this list */
 136         i_shift_offset   = DIM*shiftidx[iidx];
 137
 138         /* Load limits for loop over neighbors */
 139         j_index_start    = jindex[iidx];
 140         j_index_end      = jindex[iidx+1];
 141
 142         /* Get outer coordinate index */
 143         inr              = iinr[iidx];
 144         i_coord_offset   = DIM*inr;
 145
 146         /* Load i particle coords and add shift vector */
 147         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 148
 149         fix0             = _mm_setzero_ps();
 150         fiy0             = _mm_setzero_ps();
 151         fiz0             = _mm_setzero_ps();
 152
 153         /* Load parameters for i particles */
 154         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 155         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 156
 157         /* Reset potential sums */
 158         velecsum         = _mm_setzero_ps();
 159         vvdwsum          = _mm_setzero_ps();
 160
 161         /* Start inner kernel loop */
 162         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 163         {
 164
 165             /* Get j neighbor index, and coordinate index */
 166             jnrA             = jjnr[jidx];
 167             jnrB             = jjnr[jidx+1];
 168             jnrC             = jjnr[jidx+2];
 169             jnrD             = jjnr[jidx+3];
 170             j_coord_offsetA  = DIM*jnrA;
 171             j_coord_offsetB  = DIM*jnrB;
 172             j_coord_offsetC  = DIM*jnrC;
 173             j_coord_offsetD  = DIM*jnrD;
 174
 175             /* load j atom coordinates */
 176             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 177                                               x+j_coord_offsetC,x+j_coord_offsetD,
 178                                               &jx0,&jy0,&jz0);
 179
 180             /* Calculate displacement vector */
 181             dx00             = _mm_sub_ps(ix0,jx0);
 182             dy00             = _mm_sub_ps(iy0,jy0);
 183             dz00             = _mm_sub_ps(iz0,jz0);
 184
 185             /* Calculate squared distance and things based on it */
 186             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 187
 188             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 189
 190             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 191
 192             /* Load parameters for j particles */
 193             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 194                                                               charge+jnrC+0,charge+jnrD+0);
 195             vdwjidx0A        = 2*vdwtype[jnrA+0];
 196             vdwjidx0B        = 2*vdwtype[jnrB+0];
 197             vdwjidx0C        = 2*vdwtype[jnrC+0];
 198             vdwjidx0D        = 2*vdwtype[jnrD+0];
 199
 200             /**************************
 201              * CALCULATE INTERACTIONS *
 202              **************************/
 203
 204             /* Compute parameters for interactions between i and j atoms */
 205             qq00             = _mm_mul_ps(iq0,jq0);
 206             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 207                                          vdwparam+vdwioffset0+vdwjidx0B,
 208                                          vdwparam+vdwioffset0+vdwjidx0C,
 209                                          vdwparam+vdwioffset0+vdwjidx0D,
 210                                          &c6_00,&c12_00);
 211
 212             /* COULOMB ELECTROSTATICS */
 213             velec            = _mm_mul_ps(qq00,rinv00);
 214             felec            = _mm_mul_ps(velec,rinvsq00);
 215
 216             /* LENNARD-JONES DISPERSION/REPULSION */
 217
 218             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 219             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 220             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 221             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 222             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 223
 224             /* Update potential sum for this i atom from the interaction with this j atom. */
 225             velecsum         = _mm_add_ps(velecsum,velec);
 226             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 227
 228             fscal            = _mm_add_ps(felec,fvdw);
 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             /* Inner loop uses 40 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             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 285
 286             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 287
 288             /* Load parameters for j particles */
 289             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 290                                                               charge+jnrC+0,charge+jnrD+0);
 291             vdwjidx0A        = 2*vdwtype[jnrA+0];
 292             vdwjidx0B        = 2*vdwtype[jnrB+0];
 293             vdwjidx0C        = 2*vdwtype[jnrC+0];
 294             vdwjidx0D        = 2*vdwtype[jnrD+0];
 295
 296             /**************************
 297              * CALCULATE INTERACTIONS *
 298              **************************/
 299
 300             /* Compute parameters for interactions between i and j atoms */
 301             qq00             = _mm_mul_ps(iq0,jq0);
 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             /* COULOMB ELECTROSTATICS */
 309             velec            = _mm_mul_ps(qq00,rinv00);
 310             felec            = _mm_mul_ps(velec,rinvsq00);
 311
 312             /* LENNARD-JONES DISPERSION/REPULSION */
 313
 314             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 315             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 316             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 317             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 318             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 319
 320             /* Update potential sum for this i atom from the interaction with this j atom. */
 321             velec            = _mm_andnot_ps(dummy_mask,velec);
 322             velecsum         = _mm_add_ps(velecsum,velec);
 323             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 324             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 325
 326             fscal            = _mm_add_ps(felec,fvdw);
 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             /* Inner loop uses 40 flops */
 347         }
 348
 349         /* End of innermost loop */
 350
 351         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 352                                               f+i_coord_offset,fshift+i_shift_offset);
 353
 354         ggid                        = gid[iidx];
 355         /* Update potential energies */
 356         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 357         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 358
 359         /* Increment number of inner iterations */
 360         inneriter                  += j_index_end - j_index_start;
 361
 362         /* Outer loop uses 9 flops */
 363     }
 364
 365     /* Increment number of outer iterations */
 366     outeriter        += nri;
 367
 368     /* Update outer/inner flops */
 369
 370     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*40);
 371 }
 372 /*
 373  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomP1P1_F_sse2_single
 374  * Electrostatics interaction: Coulomb
 375  * VdW interaction:            LennardJones
 376  * Geometry:                   Particle-Particle
 377  * Calculate force/pot:        Force
 378  */
 379 void
 380 nb_kernel_ElecCoul_VdwLJ_GeomP1P1_F_sse2_single
 381                     (t_nblist                    * gmx_restrict       nlist,
 382                      rvec                        * gmx_restrict          xx,
 383                      rvec                        * gmx_restrict          ff,
 384                      t_forcerec                  * gmx_restrict          fr,
 385                      t_mdatoms                   * gmx_restrict     mdatoms,
 386                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 387                      t_nrnb                      * gmx_restrict        nrnb)
 388 {
 389     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 390      * just 0 for non-waters.
 391      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 392      * jnr indices corresponding to data put in the four positions in the SIMD register.
 393      */
 394     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 395     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 396     int              jnrA,jnrB,jnrC,jnrD;
 397     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 398     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 399     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 400     real             rcutoff_scalar;
 401     real             *shiftvec,*fshift,*x,*f;
 402     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 403     real             scratch[4*DIM];
 404     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 405     int              vdwioffset0;
 406     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 407     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 408     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 409     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 410     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 411     real             *charge;
 412     int              nvdwtype;
 413     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 414     int              *vdwtype;
 415     real             *vdwparam;
 416     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 417     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 418     __m128           dummy_mask,cutoff_mask;
 419     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 420     __m128           one     = _mm_set1_ps(1.0);
 421     __m128           two     = _mm_set1_ps(2.0);
 422     x                = xx[0];
 423     f                = ff[0];
 424
 425     nri              = nlist->nri;
 426     iinr             = nlist->iinr;
 427     jindex           = nlist->jindex;
 428     jjnr             = nlist->jjnr;
 429     shiftidx         = nlist->shift;
 430     gid              = nlist->gid;
 431     shiftvec         = fr->shift_vec[0];
 432     fshift           = fr->fshift[0];
 433     facel            = _mm_set1_ps(fr->epsfac);
 434     charge           = mdatoms->chargeA;
 435     nvdwtype         = fr->ntype;
 436     vdwparam         = fr->nbfp;
 437     vdwtype          = mdatoms->typeA;
 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         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 477         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 478
 479         /* Start inner kernel loop */
 480         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 481         {
 482
 483             /* Get j neighbor index, and coordinate index */
 484             jnrA             = jjnr[jidx];
 485             jnrB             = jjnr[jidx+1];
 486             jnrC             = jjnr[jidx+2];
 487             jnrD             = jjnr[jidx+3];
 488             j_coord_offsetA  = DIM*jnrA;
 489             j_coord_offsetB  = DIM*jnrB;
 490             j_coord_offsetC  = DIM*jnrC;
 491             j_coord_offsetD  = DIM*jnrD;
 492
 493             /* load j atom coordinates */
 494             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 495                                               x+j_coord_offsetC,x+j_coord_offsetD,
 496                                               &jx0,&jy0,&jz0);
 497
 498             /* Calculate displacement vector */
 499             dx00             = _mm_sub_ps(ix0,jx0);
 500             dy00             = _mm_sub_ps(iy0,jy0);
 501             dz00             = _mm_sub_ps(iz0,jz0);
 502
 503             /* Calculate squared distance and things based on it */
 504             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 505
 506             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 507
 508             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 509
 510             /* Load parameters for j particles */
 511             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 512                                                               charge+jnrC+0,charge+jnrD+0);
 513             vdwjidx0A        = 2*vdwtype[jnrA+0];
 514             vdwjidx0B        = 2*vdwtype[jnrB+0];
 515             vdwjidx0C        = 2*vdwtype[jnrC+0];
 516             vdwjidx0D        = 2*vdwtype[jnrD+0];
 517
 518             /**************************
 519              * CALCULATE INTERACTIONS *
 520              **************************/
 521
 522             /* Compute parameters for interactions between i and j atoms */
 523             qq00             = _mm_mul_ps(iq0,jq0);
 524             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 525                                          vdwparam+vdwioffset0+vdwjidx0B,
 526                                          vdwparam+vdwioffset0+vdwjidx0C,
 527                                          vdwparam+vdwioffset0+vdwjidx0D,
 528                                          &c6_00,&c12_00);
 529
 530             /* COULOMB ELECTROSTATICS */
 531             velec            = _mm_mul_ps(qq00,rinv00);
 532             felec            = _mm_mul_ps(velec,rinvsq00);
 533
 534             /* LENNARD-JONES DISPERSION/REPULSION */
 535
 536             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 537             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 538
 539             fscal            = _mm_add_ps(felec,fvdw);
 540
 541             /* Calculate temporary vectorial force */
 542             tx               = _mm_mul_ps(fscal,dx00);
 543             ty               = _mm_mul_ps(fscal,dy00);
 544             tz               = _mm_mul_ps(fscal,dz00);
 545
 546             /* Update vectorial force */
 547             fix0             = _mm_add_ps(fix0,tx);
 548             fiy0             = _mm_add_ps(fiy0,ty);
 549             fiz0             = _mm_add_ps(fiz0,tz);
 550
 551             fjptrA             = f+j_coord_offsetA;
 552             fjptrB             = f+j_coord_offsetB;
 553             fjptrC             = f+j_coord_offsetC;
 554             fjptrD             = f+j_coord_offsetD;
 555             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 556
 557             /* Inner loop uses 34 flops */
 558         }
 559
 560         if(jidx<j_index_end)
 561         {
 562
 563             /* Get j neighbor index, and coordinate index */
 564             jnrlistA         = jjnr[jidx];
 565             jnrlistB         = jjnr[jidx+1];
 566             jnrlistC         = jjnr[jidx+2];
 567             jnrlistD         = jjnr[jidx+3];
 568             /* Sign of each element will be negative for non-real atoms.
 569              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 570              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 571              */
 572             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 573             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 574             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 575             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 576             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 577             j_coord_offsetA  = DIM*jnrA;
 578             j_coord_offsetB  = DIM*jnrB;
 579             j_coord_offsetC  = DIM*jnrC;
 580             j_coord_offsetD  = DIM*jnrD;
 581
 582             /* load j atom coordinates */
 583             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 584                                               x+j_coord_offsetC,x+j_coord_offsetD,
 585                                               &jx0,&jy0,&jz0);
 586
 587             /* Calculate displacement vector */
 588             dx00             = _mm_sub_ps(ix0,jx0);
 589             dy00             = _mm_sub_ps(iy0,jy0);
 590             dz00             = _mm_sub_ps(iz0,jz0);
 591
 592             /* Calculate squared distance and things based on it */
 593             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 594
 595             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 596
 597             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 598
 599             /* Load parameters for j particles */
 600             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 601                                                               charge+jnrC+0,charge+jnrD+0);
 602             vdwjidx0A        = 2*vdwtype[jnrA+0];
 603             vdwjidx0B        = 2*vdwtype[jnrB+0];
 604             vdwjidx0C        = 2*vdwtype[jnrC+0];
 605             vdwjidx0D        = 2*vdwtype[jnrD+0];
 606
 607             /**************************
 608              * CALCULATE INTERACTIONS *
 609              **************************/
 610
 611             /* Compute parameters for interactions between i and j atoms */
 612             qq00             = _mm_mul_ps(iq0,jq0);
 613             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 614                                          vdwparam+vdwioffset0+vdwjidx0B,
 615                                          vdwparam+vdwioffset0+vdwjidx0C,
 616                                          vdwparam+vdwioffset0+vdwjidx0D,
 617                                          &c6_00,&c12_00);
 618
 619             /* COULOMB ELECTROSTATICS */
 620             velec            = _mm_mul_ps(qq00,rinv00);
 621             felec            = _mm_mul_ps(velec,rinvsq00);
 622
 623             /* LENNARD-JONES DISPERSION/REPULSION */
 624
 625             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 626             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 627
 628             fscal            = _mm_add_ps(felec,fvdw);
 629
 630             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 631
 632             /* Calculate temporary vectorial force */
 633             tx               = _mm_mul_ps(fscal,dx00);
 634             ty               = _mm_mul_ps(fscal,dy00);
 635             tz               = _mm_mul_ps(fscal,dz00);
 636
 637             /* Update vectorial force */
 638             fix0             = _mm_add_ps(fix0,tx);
 639             fiy0             = _mm_add_ps(fiy0,ty);
 640             fiz0             = _mm_add_ps(fiz0,tz);
 641
 642             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 643             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 644             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 645             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 646             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 647
 648             /* Inner loop uses 34 flops */
 649         }
 650
 651         /* End of innermost loop */
 652
 653         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 654                                               f+i_coord_offset,fshift+i_shift_offset);
 655
 656         /* Increment number of inner iterations */
 657         inneriter                  += j_index_end - j_index_start;
 658
 659         /* Outer loop uses 7 flops */
 660     }
 661
 662     /* Increment number of outer iterations */
 663     outeriter        += nri;
 664
 665     /* Update outer/inner flops */
 666
 667     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*34);
 668 }