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) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomP1P1_VF_avx_128_fma_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_avx_128_fma_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  86     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __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_msub_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              /* Update vectorial force */
 231             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 232             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 233             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 234
 235             fjptrA             = f+j_coord_offsetA;
 236             fjptrB             = f+j_coord_offsetB;
 237             fjptrC             = f+j_coord_offsetC;
 238             fjptrD             = f+j_coord_offsetD;
 239             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 240                                                    _mm_mul_ps(dx00,fscal),
 241                                                    _mm_mul_ps(dy00,fscal),
 242                                                    _mm_mul_ps(dz00,fscal));
 243
 244             /* Inner loop uses 43 flops */
 245         }
 246
 247         if(jidx<j_index_end)
 248         {
 249
 250             /* Get j neighbor index, and coordinate index */
 251             jnrlistA         = jjnr[jidx];
 252             jnrlistB         = jjnr[jidx+1];
 253             jnrlistC         = jjnr[jidx+2];
 254             jnrlistD         = jjnr[jidx+3];
 255             /* Sign of each element will be negative for non-real atoms.
 256              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 257              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 258              */
 259             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 260             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 261             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 262             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 263             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 264             j_coord_offsetA  = DIM*jnrA;
 265             j_coord_offsetB  = DIM*jnrB;
 266             j_coord_offsetC  = DIM*jnrC;
 267             j_coord_offsetD  = DIM*jnrD;
 268
 269             /* load j atom coordinates */
 270             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 271                                               x+j_coord_offsetC,x+j_coord_offsetD,
 272                                               &jx0,&jy0,&jz0);
 273
 274             /* Calculate displacement vector */
 275             dx00             = _mm_sub_ps(ix0,jx0);
 276             dy00             = _mm_sub_ps(iy0,jy0);
 277             dz00             = _mm_sub_ps(iz0,jz0);
 278
 279             /* Calculate squared distance and things based on it */
 280             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 281
 282             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 283
 284             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 285
 286             /* Load parameters for j particles */
 287             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 288                                                               charge+jnrC+0,charge+jnrD+0);
 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             /* Compute parameters for interactions between i and j atoms */
 299             qq00             = _mm_mul_ps(iq0,jq0);
 300             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 301                                          vdwparam+vdwioffset0+vdwjidx0B,
 302                                          vdwparam+vdwioffset0+vdwjidx0C,
 303                                          vdwparam+vdwioffset0+vdwjidx0D,
 304                                          &c6_00,&c12_00);
 305
 306             /* COULOMB ELECTROSTATICS */
 307             velec            = _mm_mul_ps(qq00,rinv00);
 308             felec            = _mm_mul_ps(velec,rinvsq00);
 309
 310             /* LENNARD-JONES DISPERSION/REPULSION */
 311
 312             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 313             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 314             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 315             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 316             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 317
 318             /* Update potential sum for this i atom from the interaction with this j atom. */
 319             velec            = _mm_andnot_ps(dummy_mask,velec);
 320             velecsum         = _mm_add_ps(velecsum,velec);
 321             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 322             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 323
 324             fscal            = _mm_add_ps(felec,fvdw);
 325
 326             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 327
 328              /* Update vectorial force */
 329             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 330             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 331             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 332
 333             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 334             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 335             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 336             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 337             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 338                                                    _mm_mul_ps(dx00,fscal),
 339                                                    _mm_mul_ps(dy00,fscal),
 340                                                    _mm_mul_ps(dz00,fscal));
 341
 342             /* Inner loop uses 43 flops */
 343         }
 344
 345         /* End of innermost loop */
 346
 347         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 348                                               f+i_coord_offset,fshift+i_shift_offset);
 349
 350         ggid                        = gid[iidx];
 351         /* Update potential energies */
 352         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 353         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 354
 355         /* Increment number of inner iterations */
 356         inneriter                  += j_index_end - j_index_start;
 357
 358         /* Outer loop uses 9 flops */
 359     }
 360
 361     /* Increment number of outer iterations */
 362     outeriter        += nri;
 363
 364     /* Update outer/inner flops */
 365
 366     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*43);
 367 }
 368 /*
 369  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomP1P1_F_avx_128_fma_single
 370  * Electrostatics interaction: Coulomb
 371  * VdW interaction:            LennardJones
 372  * Geometry:                   Particle-Particle
 373  * Calculate force/pot:        Force
 374  */
 375 void
 376 nb_kernel_ElecCoul_VdwLJ_GeomP1P1_F_avx_128_fma_single
 377                     (t_nblist                    * gmx_restrict       nlist,
 378                      rvec                        * gmx_restrict          xx,
 379                      rvec                        * gmx_restrict          ff,
 380                      t_forcerec                  * gmx_restrict          fr,
 381                      t_mdatoms                   * gmx_restrict     mdatoms,
 382                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 383                      t_nrnb                      * gmx_restrict        nrnb)
 384 {
 385     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 386      * just 0 for non-waters.
 387      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 388      * jnr indices corresponding to data put in the four positions in the SIMD register.
 389      */
 390     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 391     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 392     int              jnrA,jnrB,jnrC,jnrD;
 393     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 394     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 395     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 396     real             rcutoff_scalar;
 397     real             *shiftvec,*fshift,*x,*f;
 398     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 399     real             scratch[4*DIM];
 400     __m128           fscal,rcutoff,rcutoff2,jidxall;
 401     int              vdwioffset0;
 402     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 403     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 404     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 405     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 406     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 407     real             *charge;
 408     int              nvdwtype;
 409     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 410     int              *vdwtype;
 411     real             *vdwparam;
 412     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 413     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 414     __m128           dummy_mask,cutoff_mask;
 415     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 416     __m128           one     = _mm_set1_ps(1.0);
 417     __m128           two     = _mm_set1_ps(2.0);
 418     x                = xx[0];
 419     f                = ff[0];
 420
 421     nri              = nlist->nri;
 422     iinr             = nlist->iinr;
 423     jindex           = nlist->jindex;
 424     jjnr             = nlist->jjnr;
 425     shiftidx         = nlist->shift;
 426     gid              = nlist->gid;
 427     shiftvec         = fr->shift_vec[0];
 428     fshift           = fr->fshift[0];
 429     facel            = _mm_set1_ps(fr->epsfac);
 430     charge           = mdatoms->chargeA;
 431     nvdwtype         = fr->ntype;
 432     vdwparam         = fr->nbfp;
 433     vdwtype          = mdatoms->typeA;
 434
 435     /* Avoid stupid compiler warnings */
 436     jnrA = jnrB = jnrC = jnrD = 0;
 437     j_coord_offsetA = 0;
 438     j_coord_offsetB = 0;
 439     j_coord_offsetC = 0;
 440     j_coord_offsetD = 0;
 441
 442     outeriter        = 0;
 443     inneriter        = 0;
 444
 445     for(iidx=0;iidx<4*DIM;iidx++)
 446     {
 447         scratch[iidx] = 0.0;
 448     }
 449
 450     /* Start outer loop over neighborlists */
 451     for(iidx=0; iidx<nri; iidx++)
 452     {
 453         /* Load shift vector for this list */
 454         i_shift_offset   = DIM*shiftidx[iidx];
 455
 456         /* Load limits for loop over neighbors */
 457         j_index_start    = jindex[iidx];
 458         j_index_end      = jindex[iidx+1];
 459
 460         /* Get outer coordinate index */
 461         inr              = iinr[iidx];
 462         i_coord_offset   = DIM*inr;
 463
 464         /* Load i particle coords and add shift vector */
 465         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 466
 467         fix0             = _mm_setzero_ps();
 468         fiy0             = _mm_setzero_ps();
 469         fiz0             = _mm_setzero_ps();
 470
 471         /* Load parameters for i particles */
 472         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 473         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 474
 475         /* Start inner kernel loop */
 476         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 477         {
 478
 479             /* Get j neighbor index, and coordinate index */
 480             jnrA             = jjnr[jidx];
 481             jnrB             = jjnr[jidx+1];
 482             jnrC             = jjnr[jidx+2];
 483             jnrD             = jjnr[jidx+3];
 484             j_coord_offsetA  = DIM*jnrA;
 485             j_coord_offsetB  = DIM*jnrB;
 486             j_coord_offsetC  = DIM*jnrC;
 487             j_coord_offsetD  = DIM*jnrD;
 488
 489             /* load j atom coordinates */
 490             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 491                                               x+j_coord_offsetC,x+j_coord_offsetD,
 492                                               &jx0,&jy0,&jz0);
 493
 494             /* Calculate displacement vector */
 495             dx00             = _mm_sub_ps(ix0,jx0);
 496             dy00             = _mm_sub_ps(iy0,jy0);
 497             dz00             = _mm_sub_ps(iz0,jz0);
 498
 499             /* Calculate squared distance and things based on it */
 500             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 501
 502             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 503
 504             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 505
 506             /* Load parameters for j particles */
 507             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 508                                                               charge+jnrC+0,charge+jnrD+0);
 509             vdwjidx0A        = 2*vdwtype[jnrA+0];
 510             vdwjidx0B        = 2*vdwtype[jnrB+0];
 511             vdwjidx0C        = 2*vdwtype[jnrC+0];
 512             vdwjidx0D        = 2*vdwtype[jnrD+0];
 513
 514             /**************************
 515              * CALCULATE INTERACTIONS *
 516              **************************/
 517
 518             /* Compute parameters for interactions between i and j atoms */
 519             qq00             = _mm_mul_ps(iq0,jq0);
 520             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 521                                          vdwparam+vdwioffset0+vdwjidx0B,
 522                                          vdwparam+vdwioffset0+vdwjidx0C,
 523                                          vdwparam+vdwioffset0+vdwjidx0D,
 524                                          &c6_00,&c12_00);
 525
 526             /* COULOMB ELECTROSTATICS */
 527             velec            = _mm_mul_ps(qq00,rinv00);
 528             felec            = _mm_mul_ps(velec,rinvsq00);
 529
 530             /* LENNARD-JONES DISPERSION/REPULSION */
 531
 532             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 533             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 534
 535             fscal            = _mm_add_ps(felec,fvdw);
 536
 537              /* Update vectorial force */
 538             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 539             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 540             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 541
 542             fjptrA             = f+j_coord_offsetA;
 543             fjptrB             = f+j_coord_offsetB;
 544             fjptrC             = f+j_coord_offsetC;
 545             fjptrD             = f+j_coord_offsetD;
 546             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 547                                                    _mm_mul_ps(dx00,fscal),
 548                                                    _mm_mul_ps(dy00,fscal),
 549                                                    _mm_mul_ps(dz00,fscal));
 550
 551             /* Inner loop uses 37 flops */
 552         }
 553
 554         if(jidx<j_index_end)
 555         {
 556
 557             /* Get j neighbor index, and coordinate index */
 558             jnrlistA         = jjnr[jidx];
 559             jnrlistB         = jjnr[jidx+1];
 560             jnrlistC         = jjnr[jidx+2];
 561             jnrlistD         = jjnr[jidx+3];
 562             /* Sign of each element will be negative for non-real atoms.
 563              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 564              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 565              */
 566             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 567             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 568             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 569             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 570             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 571             j_coord_offsetA  = DIM*jnrA;
 572             j_coord_offsetB  = DIM*jnrB;
 573             j_coord_offsetC  = DIM*jnrC;
 574             j_coord_offsetD  = DIM*jnrD;
 575
 576             /* load j atom coordinates */
 577             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 578                                               x+j_coord_offsetC,x+j_coord_offsetD,
 579                                               &jx0,&jy0,&jz0);
 580
 581             /* Calculate displacement vector */
 582             dx00             = _mm_sub_ps(ix0,jx0);
 583             dy00             = _mm_sub_ps(iy0,jy0);
 584             dz00             = _mm_sub_ps(iz0,jz0);
 585
 586             /* Calculate squared distance and things based on it */
 587             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 588
 589             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 590
 591             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 592
 593             /* Load parameters for j particles */
 594             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 595                                                               charge+jnrC+0,charge+jnrD+0);
 596             vdwjidx0A        = 2*vdwtype[jnrA+0];
 597             vdwjidx0B        = 2*vdwtype[jnrB+0];
 598             vdwjidx0C        = 2*vdwtype[jnrC+0];
 599             vdwjidx0D        = 2*vdwtype[jnrD+0];
 600
 601             /**************************
 602              * CALCULATE INTERACTIONS *
 603              **************************/
 604
 605             /* Compute parameters for interactions between i and j atoms */
 606             qq00             = _mm_mul_ps(iq0,jq0);
 607             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 608                                          vdwparam+vdwioffset0+vdwjidx0B,
 609                                          vdwparam+vdwioffset0+vdwjidx0C,
 610                                          vdwparam+vdwioffset0+vdwjidx0D,
 611                                          &c6_00,&c12_00);
 612
 613             /* COULOMB ELECTROSTATICS */
 614             velec            = _mm_mul_ps(qq00,rinv00);
 615             felec            = _mm_mul_ps(velec,rinvsq00);
 616
 617             /* LENNARD-JONES DISPERSION/REPULSION */
 618
 619             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 620             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 621
 622             fscal            = _mm_add_ps(felec,fvdw);
 623
 624             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 625
 626              /* Update vectorial force */
 627             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 628             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 629             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 630
 631             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 632             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 633             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 634             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 635             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 636                                                    _mm_mul_ps(dx00,fscal),
 637                                                    _mm_mul_ps(dy00,fscal),
 638                                                    _mm_mul_ps(dz00,fscal));
 639
 640             /* Inner loop uses 37 flops */
 641         }
 642
 643         /* End of innermost loop */
 644
 645         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 646                                               f+i_coord_offset,fshift+i_shift_offset);
 647
 648         /* Increment number of inner iterations */
 649         inneriter                  += j_index_end - j_index_start;
 650
 651         /* Outer loop uses 7 flops */
 652     }
 653
 654     /* Increment number of outer iterations */
 655     outeriter        += nri;
 656
 657     /* Update outer/inner flops */
 658
 659     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*37);
 660 }