src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecCoul_VdwLJ_GeomP1P1_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  50 #include "kernelutil_x86_avx_128_fma_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomP1P1_VF_avx_128_fma_single
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_VdwLJ_GeomP1P1_VF_avx_128_fma_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  88     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     int              nvdwtype;
  93     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  97     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  98     __m128           dummy_mask,cutoff_mask;
  99     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 100     __m128           one     = _mm_set1_ps(1.0);
 101     __m128           two     = _mm_set1_ps(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm_set1_ps(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115     nvdwtype         = fr->ntype;
 116     vdwparam         = fr->nbfp;
 117     vdwtype          = mdatoms->typeA;
 118
 119     /* Avoid stupid compiler warnings */
 120     jnrA = jnrB = jnrC = jnrD = 0;
 121     j_coord_offsetA = 0;
 122     j_coord_offsetB = 0;
 123     j_coord_offsetC = 0;
 124     j_coord_offsetD = 0;
 125
 126     outeriter        = 0;
 127     inneriter        = 0;
 128
 129     for(iidx=0;iidx<4*DIM;iidx++)
 130     {
 131         scratch[iidx] = 0.0;
 132     }
 133
 134     /* Start outer loop over neighborlists */
 135     for(iidx=0; iidx<nri; iidx++)
 136     {
 137         /* Load shift vector for this list */
 138         i_shift_offset   = DIM*shiftidx[iidx];
 139
 140         /* Load limits for loop over neighbors */
 141         j_index_start    = jindex[iidx];
 142         j_index_end      = jindex[iidx+1];
 143
 144         /* Get outer coordinate index */
 145         inr              = iinr[iidx];
 146         i_coord_offset   = DIM*inr;
 147
 148         /* Load i particle coords and add shift vector */
 149         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 150
 151         fix0             = _mm_setzero_ps();
 152         fiy0             = _mm_setzero_ps();
 153         fiz0             = _mm_setzero_ps();
 154
 155         /* Load parameters for i particles */
 156         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 157         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 158
 159         /* Reset potential sums */
 160         velecsum         = _mm_setzero_ps();
 161         vvdwsum          = _mm_setzero_ps();
 162
 163         /* Start inner kernel loop */
 164         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 165         {
 166
 167             /* Get j neighbor index, and coordinate index */
 168             jnrA             = jjnr[jidx];
 169             jnrB             = jjnr[jidx+1];
 170             jnrC             = jjnr[jidx+2];
 171             jnrD             = jjnr[jidx+3];
 172             j_coord_offsetA  = DIM*jnrA;
 173             j_coord_offsetB  = DIM*jnrB;
 174             j_coord_offsetC  = DIM*jnrC;
 175             j_coord_offsetD  = DIM*jnrD;
 176
 177             /* load j atom coordinates */
 178             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 179                                               x+j_coord_offsetC,x+j_coord_offsetD,
 180                                               &jx0,&jy0,&jz0);
 181
 182             /* Calculate displacement vector */
 183             dx00             = _mm_sub_ps(ix0,jx0);
 184             dy00             = _mm_sub_ps(iy0,jy0);
 185             dz00             = _mm_sub_ps(iz0,jz0);
 186
 187             /* Calculate squared distance and things based on it */
 188             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 189
 190             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 191
 192             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 193
 194             /* Load parameters for j particles */
 195             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 196                                                               charge+jnrC+0,charge+jnrD+0);
 197             vdwjidx0A        = 2*vdwtype[jnrA+0];
 198             vdwjidx0B        = 2*vdwtype[jnrB+0];
 199             vdwjidx0C        = 2*vdwtype[jnrC+0];
 200             vdwjidx0D        = 2*vdwtype[jnrD+0];
 201
 202             /**************************
 203              * CALCULATE INTERACTIONS *
 204              **************************/
 205
 206             /* Compute parameters for interactions between i and j atoms */
 207             qq00             = _mm_mul_ps(iq0,jq0);
 208             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 209                                          vdwparam+vdwioffset0+vdwjidx0B,
 210                                          vdwparam+vdwioffset0+vdwjidx0C,
 211                                          vdwparam+vdwioffset0+vdwjidx0D,
 212                                          &c6_00,&c12_00);
 213
 214             /* COULOMB ELECTROSTATICS */
 215             velec            = _mm_mul_ps(qq00,rinv00);
 216             felec            = _mm_mul_ps(velec,rinvsq00);
 217
 218             /* LENNARD-JONES DISPERSION/REPULSION */
 219
 220             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 221             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 222             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 223             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 224             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 225
 226             /* Update potential sum for this i atom from the interaction with this j atom. */
 227             velecsum         = _mm_add_ps(velecsum,velec);
 228             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 229
 230             fscal            = _mm_add_ps(felec,fvdw);
 231
 232              /* Update vectorial force */
 233             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 234             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 235             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 236
 237             fjptrA             = f+j_coord_offsetA;
 238             fjptrB             = f+j_coord_offsetB;
 239             fjptrC             = f+j_coord_offsetC;
 240             fjptrD             = f+j_coord_offsetD;
 241             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 242                                                    _mm_mul_ps(dx00,fscal),
 243                                                    _mm_mul_ps(dy00,fscal),
 244                                                    _mm_mul_ps(dz00,fscal));
 245
 246             /* Inner loop uses 43 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_msub_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              /* Update vectorial force */
 331             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 332             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 333             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 334
 335             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 336             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 337             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 338             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 339             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 340                                                    _mm_mul_ps(dx00,fscal),
 341                                                    _mm_mul_ps(dy00,fscal),
 342                                                    _mm_mul_ps(dz00,fscal));
 343
 344             /* Inner loop uses 43 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(velecsum,kernel_data->energygrp_elec+ggid);
 355         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 356
 357         /* Increment number of inner iterations */
 358         inneriter                  += j_index_end - j_index_start;
 359
 360         /* Outer loop uses 9 flops */
 361     }
 362
 363     /* Increment number of outer iterations */
 364     outeriter        += nri;
 365
 366     /* Update outer/inner flops */
 367
 368     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*43);
 369 }
 370 /*
 371  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomP1P1_F_avx_128_fma_single
 372  * Electrostatics interaction: Coulomb
 373  * VdW interaction:            LennardJones
 374  * Geometry:                   Particle-Particle
 375  * Calculate force/pot:        Force
 376  */
 377 void
 378 nb_kernel_ElecCoul_VdwLJ_GeomP1P1_F_avx_128_fma_single
 379                     (t_nblist                    * gmx_restrict       nlist,
 380                      rvec                        * gmx_restrict          xx,
 381                      rvec                        * gmx_restrict          ff,
 382                      t_forcerec                  * gmx_restrict          fr,
 383                      t_mdatoms                   * gmx_restrict     mdatoms,
 384                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 385                      t_nrnb                      * gmx_restrict        nrnb)
 386 {
 387     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 388      * just 0 for non-waters.
 389      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 390      * jnr indices corresponding to data put in the four positions in the SIMD register.
 391      */
 392     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 393     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 394     int              jnrA,jnrB,jnrC,jnrD;
 395     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 396     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 397     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 398     real             rcutoff_scalar;
 399     real             *shiftvec,*fshift,*x,*f;
 400     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 401     real             scratch[4*DIM];
 402     __m128           fscal,rcutoff,rcutoff2,jidxall;
 403     int              vdwioffset0;
 404     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 405     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 406     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 407     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 408     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 409     real             *charge;
 410     int              nvdwtype;
 411     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 412     int              *vdwtype;
 413     real             *vdwparam;
 414     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 415     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 416     __m128           dummy_mask,cutoff_mask;
 417     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 418     __m128           one     = _mm_set1_ps(1.0);
 419     __m128           two     = _mm_set1_ps(2.0);
 420     x                = xx[0];
 421     f                = ff[0];
 422
 423     nri              = nlist->nri;
 424     iinr             = nlist->iinr;
 425     jindex           = nlist->jindex;
 426     jjnr             = nlist->jjnr;
 427     shiftidx         = nlist->shift;
 428     gid              = nlist->gid;
 429     shiftvec         = fr->shift_vec[0];
 430     fshift           = fr->fshift[0];
 431     facel            = _mm_set1_ps(fr->epsfac);
 432     charge           = mdatoms->chargeA;
 433     nvdwtype         = fr->ntype;
 434     vdwparam         = fr->nbfp;
 435     vdwtype          = mdatoms->typeA;
 436
 437     /* Avoid stupid compiler warnings */
 438     jnrA = jnrB = jnrC = jnrD = 0;
 439     j_coord_offsetA = 0;
 440     j_coord_offsetB = 0;
 441     j_coord_offsetC = 0;
 442     j_coord_offsetD = 0;
 443
 444     outeriter        = 0;
 445     inneriter        = 0;
 446
 447     for(iidx=0;iidx<4*DIM;iidx++)
 448     {
 449         scratch[iidx] = 0.0;
 450     }
 451
 452     /* Start outer loop over neighborlists */
 453     for(iidx=0; iidx<nri; iidx++)
 454     {
 455         /* Load shift vector for this list */
 456         i_shift_offset   = DIM*shiftidx[iidx];
 457
 458         /* Load limits for loop over neighbors */
 459         j_index_start    = jindex[iidx];
 460         j_index_end      = jindex[iidx+1];
 461
 462         /* Get outer coordinate index */
 463         inr              = iinr[iidx];
 464         i_coord_offset   = DIM*inr;
 465
 466         /* Load i particle coords and add shift vector */
 467         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 468
 469         fix0             = _mm_setzero_ps();
 470         fiy0             = _mm_setzero_ps();
 471         fiz0             = _mm_setzero_ps();
 472
 473         /* Load parameters for i particles */
 474         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 475         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 476
 477         /* Start inner kernel loop */
 478         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 479         {
 480
 481             /* Get j neighbor index, and coordinate index */
 482             jnrA             = jjnr[jidx];
 483             jnrB             = jjnr[jidx+1];
 484             jnrC             = jjnr[jidx+2];
 485             jnrD             = jjnr[jidx+3];
 486             j_coord_offsetA  = DIM*jnrA;
 487             j_coord_offsetB  = DIM*jnrB;
 488             j_coord_offsetC  = DIM*jnrC;
 489             j_coord_offsetD  = DIM*jnrD;
 490
 491             /* load j atom coordinates */
 492             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 493                                               x+j_coord_offsetC,x+j_coord_offsetD,
 494                                               &jx0,&jy0,&jz0);
 495
 496             /* Calculate displacement vector */
 497             dx00             = _mm_sub_ps(ix0,jx0);
 498             dy00             = _mm_sub_ps(iy0,jy0);
 499             dz00             = _mm_sub_ps(iz0,jz0);
 500
 501             /* Calculate squared distance and things based on it */
 502             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 503
 504             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 505
 506             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 507
 508             /* Load parameters for j particles */
 509             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 510                                                               charge+jnrC+0,charge+jnrD+0);
 511             vdwjidx0A        = 2*vdwtype[jnrA+0];
 512             vdwjidx0B        = 2*vdwtype[jnrB+0];
 513             vdwjidx0C        = 2*vdwtype[jnrC+0];
 514             vdwjidx0D        = 2*vdwtype[jnrD+0];
 515
 516             /**************************
 517              * CALCULATE INTERACTIONS *
 518              **************************/
 519
 520             /* Compute parameters for interactions between i and j atoms */
 521             qq00             = _mm_mul_ps(iq0,jq0);
 522             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 523                                          vdwparam+vdwioffset0+vdwjidx0B,
 524                                          vdwparam+vdwioffset0+vdwjidx0C,
 525                                          vdwparam+vdwioffset0+vdwjidx0D,
 526                                          &c6_00,&c12_00);
 527
 528             /* COULOMB ELECTROSTATICS */
 529             velec            = _mm_mul_ps(qq00,rinv00);
 530             felec            = _mm_mul_ps(velec,rinvsq00);
 531
 532             /* LENNARD-JONES DISPERSION/REPULSION */
 533
 534             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 535             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 536
 537             fscal            = _mm_add_ps(felec,fvdw);
 538
 539              /* Update vectorial force */
 540             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 541             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 542             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 543
 544             fjptrA             = f+j_coord_offsetA;
 545             fjptrB             = f+j_coord_offsetB;
 546             fjptrC             = f+j_coord_offsetC;
 547             fjptrD             = f+j_coord_offsetD;
 548             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 549                                                    _mm_mul_ps(dx00,fscal),
 550                                                    _mm_mul_ps(dy00,fscal),
 551                                                    _mm_mul_ps(dz00,fscal));
 552
 553             /* Inner loop uses 37 flops */
 554         }
 555
 556         if(jidx<j_index_end)
 557         {
 558
 559             /* Get j neighbor index, and coordinate index */
 560             jnrlistA         = jjnr[jidx];
 561             jnrlistB         = jjnr[jidx+1];
 562             jnrlistC         = jjnr[jidx+2];
 563             jnrlistD         = jjnr[jidx+3];
 564             /* Sign of each element will be negative for non-real atoms.
 565              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 566              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 567              */
 568             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 569             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 570             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 571             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 572             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 573             j_coord_offsetA  = DIM*jnrA;
 574             j_coord_offsetB  = DIM*jnrB;
 575             j_coord_offsetC  = DIM*jnrC;
 576             j_coord_offsetD  = DIM*jnrD;
 577
 578             /* load j atom coordinates */
 579             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 580                                               x+j_coord_offsetC,x+j_coord_offsetD,
 581                                               &jx0,&jy0,&jz0);
 582
 583             /* Calculate displacement vector */
 584             dx00             = _mm_sub_ps(ix0,jx0);
 585             dy00             = _mm_sub_ps(iy0,jy0);
 586             dz00             = _mm_sub_ps(iz0,jz0);
 587
 588             /* Calculate squared distance and things based on it */
 589             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 590
 591             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 592
 593             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 594
 595             /* Load parameters for j particles */
 596             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 597                                                               charge+jnrC+0,charge+jnrD+0);
 598             vdwjidx0A        = 2*vdwtype[jnrA+0];
 599             vdwjidx0B        = 2*vdwtype[jnrB+0];
 600             vdwjidx0C        = 2*vdwtype[jnrC+0];
 601             vdwjidx0D        = 2*vdwtype[jnrD+0];
 602
 603             /**************************
 604              * CALCULATE INTERACTIONS *
 605              **************************/
 606
 607             /* Compute parameters for interactions between i and j atoms */
 608             qq00             = _mm_mul_ps(iq0,jq0);
 609             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 610                                          vdwparam+vdwioffset0+vdwjidx0B,
 611                                          vdwparam+vdwioffset0+vdwjidx0C,
 612                                          vdwparam+vdwioffset0+vdwjidx0D,
 613                                          &c6_00,&c12_00);
 614
 615             /* COULOMB ELECTROSTATICS */
 616             velec            = _mm_mul_ps(qq00,rinv00);
 617             felec            = _mm_mul_ps(velec,rinvsq00);
 618
 619             /* LENNARD-JONES DISPERSION/REPULSION */
 620
 621             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 622             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 623
 624             fscal            = _mm_add_ps(felec,fvdw);
 625
 626             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 627
 628              /* Update vectorial force */
 629             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 630             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 631             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 632
 633             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 634             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 635             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 636             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 637             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 638                                                    _mm_mul_ps(dx00,fscal),
 639                                                    _mm_mul_ps(dy00,fscal),
 640                                                    _mm_mul_ps(dz00,fscal));
 641
 642             /* Inner loop uses 37 flops */
 643         }
 644
 645         /* End of innermost loop */
 646
 647         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 648                                               f+i_coord_offset,fshift+i_shift_offset);
 649
 650         /* Increment number of inner iterations */
 651         inneriter                  += j_index_end - j_index_start;
 652
 653         /* Outer loop uses 7 flops */
 654     }
 655
 656     /* Increment number of outer iterations */
 657     outeriter        += nri;
 658
 659     /* Update outer/inner flops */
 660
 661     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*37);
 662 }