src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecCoul_VdwLJ_GeomW3P1_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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/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_GeomW3P1_VF_avx_128_fma_single
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_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              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128           dummy_mask,cutoff_mask;
 105     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 106     __m128           one     = _mm_set1_ps(1.0);
 107     __m128           two     = _mm_set1_ps(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm_set1_ps(fr->epsfac);
 120     charge           = mdatoms->chargeA;
 121     nvdwtype         = fr->ntype;
 122     vdwparam         = fr->nbfp;
 123     vdwtype          = mdatoms->typeA;
 124
 125     /* Setup water-specific parameters */
 126     inr              = nlist->iinr[0];
 127     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 128     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 129     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 130     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 131
 132     /* Avoid stupid compiler warnings */
 133     jnrA = jnrB = jnrC = jnrD = 0;
 134     j_coord_offsetA = 0;
 135     j_coord_offsetB = 0;
 136     j_coord_offsetC = 0;
 137     j_coord_offsetD = 0;
 138
 139     outeriter        = 0;
 140     inneriter        = 0;
 141
 142     for(iidx=0;iidx<4*DIM;iidx++)
 143     {
 144         scratch[iidx] = 0.0;
 145     }
 146
 147     /* Start outer loop over neighborlists */
 148     for(iidx=0; iidx<nri; iidx++)
 149     {
 150         /* Load shift vector for this list */
 151         i_shift_offset   = DIM*shiftidx[iidx];
 152
 153         /* Load limits for loop over neighbors */
 154         j_index_start    = jindex[iidx];
 155         j_index_end      = jindex[iidx+1];
 156
 157         /* Get outer coordinate index */
 158         inr              = iinr[iidx];
 159         i_coord_offset   = DIM*inr;
 160
 161         /* Load i particle coords and add shift vector */
 162         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 163                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 164
 165         fix0             = _mm_setzero_ps();
 166         fiy0             = _mm_setzero_ps();
 167         fiz0             = _mm_setzero_ps();
 168         fix1             = _mm_setzero_ps();
 169         fiy1             = _mm_setzero_ps();
 170         fiz1             = _mm_setzero_ps();
 171         fix2             = _mm_setzero_ps();
 172         fiy2             = _mm_setzero_ps();
 173         fiz2             = _mm_setzero_ps();
 174
 175         /* Reset potential sums */
 176         velecsum         = _mm_setzero_ps();
 177         vvdwsum          = _mm_setzero_ps();
 178
 179         /* Start inner kernel loop */
 180         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 181         {
 182
 183             /* Get j neighbor index, and coordinate index */
 184             jnrA             = jjnr[jidx];
 185             jnrB             = jjnr[jidx+1];
 186             jnrC             = jjnr[jidx+2];
 187             jnrD             = jjnr[jidx+3];
 188             j_coord_offsetA  = DIM*jnrA;
 189             j_coord_offsetB  = DIM*jnrB;
 190             j_coord_offsetC  = DIM*jnrC;
 191             j_coord_offsetD  = DIM*jnrD;
 192
 193             /* load j atom coordinates */
 194             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 195                                               x+j_coord_offsetC,x+j_coord_offsetD,
 196                                               &jx0,&jy0,&jz0);
 197
 198             /* Calculate displacement vector */
 199             dx00             = _mm_sub_ps(ix0,jx0);
 200             dy00             = _mm_sub_ps(iy0,jy0);
 201             dz00             = _mm_sub_ps(iz0,jz0);
 202             dx10             = _mm_sub_ps(ix1,jx0);
 203             dy10             = _mm_sub_ps(iy1,jy0);
 204             dz10             = _mm_sub_ps(iz1,jz0);
 205             dx20             = _mm_sub_ps(ix2,jx0);
 206             dy20             = _mm_sub_ps(iy2,jy0);
 207             dz20             = _mm_sub_ps(iz2,jz0);
 208
 209             /* Calculate squared distance and things based on it */
 210             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 211             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 212             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 213
 214             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 215             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 216             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 217
 218             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 219             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 220             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 221
 222             /* Load parameters for j particles */
 223             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 224                                                               charge+jnrC+0,charge+jnrD+0);
 225             vdwjidx0A        = 2*vdwtype[jnrA+0];
 226             vdwjidx0B        = 2*vdwtype[jnrB+0];
 227             vdwjidx0C        = 2*vdwtype[jnrC+0];
 228             vdwjidx0D        = 2*vdwtype[jnrD+0];
 229
 230             fjx0             = _mm_setzero_ps();
 231             fjy0             = _mm_setzero_ps();
 232             fjz0             = _mm_setzero_ps();
 233
 234             /**************************
 235              * CALCULATE INTERACTIONS *
 236              **************************/
 237
 238             /* Compute parameters for interactions between i and j atoms */
 239             qq00             = _mm_mul_ps(iq0,jq0);
 240             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 241                                          vdwparam+vdwioffset0+vdwjidx0B,
 242                                          vdwparam+vdwioffset0+vdwjidx0C,
 243                                          vdwparam+vdwioffset0+vdwjidx0D,
 244                                          &c6_00,&c12_00);
 245
 246             /* COULOMB ELECTROSTATICS */
 247             velec            = _mm_mul_ps(qq00,rinv00);
 248             felec            = _mm_mul_ps(velec,rinvsq00);
 249
 250             /* LENNARD-JONES DISPERSION/REPULSION */
 251
 252             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 253             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 254             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 255             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 256             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 257
 258             /* Update potential sum for this i atom from the interaction with this j atom. */
 259             velecsum         = _mm_add_ps(velecsum,velec);
 260             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 261
 262             fscal            = _mm_add_ps(felec,fvdw);
 263
 264              /* Update vectorial force */
 265             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 266             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 267             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 268
 269             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 270             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 271             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 272
 273             /**************************
 274              * CALCULATE INTERACTIONS *
 275              **************************/
 276
 277             /* Compute parameters for interactions between i and j atoms */
 278             qq10             = _mm_mul_ps(iq1,jq0);
 279
 280             /* COULOMB ELECTROSTATICS */
 281             velec            = _mm_mul_ps(qq10,rinv10);
 282             felec            = _mm_mul_ps(velec,rinvsq10);
 283
 284             /* Update potential sum for this i atom from the interaction with this j atom. */
 285             velecsum         = _mm_add_ps(velecsum,velec);
 286
 287             fscal            = felec;
 288
 289              /* Update vectorial force */
 290             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 291             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 292             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 293
 294             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 295             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 296             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 297
 298             /**************************
 299              * CALCULATE INTERACTIONS *
 300              **************************/
 301
 302             /* Compute parameters for interactions between i and j atoms */
 303             qq20             = _mm_mul_ps(iq2,jq0);
 304
 305             /* COULOMB ELECTROSTATICS */
 306             velec            = _mm_mul_ps(qq20,rinv20);
 307             felec            = _mm_mul_ps(velec,rinvsq20);
 308
 309             /* Update potential sum for this i atom from the interaction with this j atom. */
 310             velecsum         = _mm_add_ps(velecsum,velec);
 311
 312             fscal            = felec;
 313
 314              /* Update vectorial force */
 315             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 316             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 317             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 318
 319             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 320             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 321             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 322
 323             fjptrA             = f+j_coord_offsetA;
 324             fjptrB             = f+j_coord_offsetB;
 325             fjptrC             = f+j_coord_offsetC;
 326             fjptrD             = f+j_coord_offsetD;
 327
 328             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 329
 330             /* Inner loop uses 105 flops */
 331         }
 332
 333         if(jidx<j_index_end)
 334         {
 335
 336             /* Get j neighbor index, and coordinate index */
 337             jnrlistA         = jjnr[jidx];
 338             jnrlistB         = jjnr[jidx+1];
 339             jnrlistC         = jjnr[jidx+2];
 340             jnrlistD         = jjnr[jidx+3];
 341             /* Sign of each element will be negative for non-real atoms.
 342              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 343              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 344              */
 345             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 346             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 347             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 348             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 349             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 350             j_coord_offsetA  = DIM*jnrA;
 351             j_coord_offsetB  = DIM*jnrB;
 352             j_coord_offsetC  = DIM*jnrC;
 353             j_coord_offsetD  = DIM*jnrD;
 354
 355             /* load j atom coordinates */
 356             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 357                                               x+j_coord_offsetC,x+j_coord_offsetD,
 358                                               &jx0,&jy0,&jz0);
 359
 360             /* Calculate displacement vector */
 361             dx00             = _mm_sub_ps(ix0,jx0);
 362             dy00             = _mm_sub_ps(iy0,jy0);
 363             dz00             = _mm_sub_ps(iz0,jz0);
 364             dx10             = _mm_sub_ps(ix1,jx0);
 365             dy10             = _mm_sub_ps(iy1,jy0);
 366             dz10             = _mm_sub_ps(iz1,jz0);
 367             dx20             = _mm_sub_ps(ix2,jx0);
 368             dy20             = _mm_sub_ps(iy2,jy0);
 369             dz20             = _mm_sub_ps(iz2,jz0);
 370
 371             /* Calculate squared distance and things based on it */
 372             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 373             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 374             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 375
 376             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 377             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 378             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 379
 380             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 381             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 382             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 383
 384             /* Load parameters for j particles */
 385             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 386                                                               charge+jnrC+0,charge+jnrD+0);
 387             vdwjidx0A        = 2*vdwtype[jnrA+0];
 388             vdwjidx0B        = 2*vdwtype[jnrB+0];
 389             vdwjidx0C        = 2*vdwtype[jnrC+0];
 390             vdwjidx0D        = 2*vdwtype[jnrD+0];
 391
 392             fjx0             = _mm_setzero_ps();
 393             fjy0             = _mm_setzero_ps();
 394             fjz0             = _mm_setzero_ps();
 395
 396             /**************************
 397              * CALCULATE INTERACTIONS *
 398              **************************/
 399
 400             /* Compute parameters for interactions between i and j atoms */
 401             qq00             = _mm_mul_ps(iq0,jq0);
 402             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 403                                          vdwparam+vdwioffset0+vdwjidx0B,
 404                                          vdwparam+vdwioffset0+vdwjidx0C,
 405                                          vdwparam+vdwioffset0+vdwjidx0D,
 406                                          &c6_00,&c12_00);
 407
 408             /* COULOMB ELECTROSTATICS */
 409             velec            = _mm_mul_ps(qq00,rinv00);
 410             felec            = _mm_mul_ps(velec,rinvsq00);
 411
 412             /* LENNARD-JONES DISPERSION/REPULSION */
 413
 414             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 415             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 416             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 417             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 418             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 419
 420             /* Update potential sum for this i atom from the interaction with this j atom. */
 421             velec            = _mm_andnot_ps(dummy_mask,velec);
 422             velecsum         = _mm_add_ps(velecsum,velec);
 423             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 424             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 425
 426             fscal            = _mm_add_ps(felec,fvdw);
 427
 428             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 429
 430              /* Update vectorial force */
 431             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 432             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 433             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 434
 435             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 436             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 437             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 438
 439             /**************************
 440              * CALCULATE INTERACTIONS *
 441              **************************/
 442
 443             /* Compute parameters for interactions between i and j atoms */
 444             qq10             = _mm_mul_ps(iq1,jq0);
 445
 446             /* COULOMB ELECTROSTATICS */
 447             velec            = _mm_mul_ps(qq10,rinv10);
 448             felec            = _mm_mul_ps(velec,rinvsq10);
 449
 450             /* Update potential sum for this i atom from the interaction with this j atom. */
 451             velec            = _mm_andnot_ps(dummy_mask,velec);
 452             velecsum         = _mm_add_ps(velecsum,velec);
 453
 454             fscal            = felec;
 455
 456             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 457
 458              /* Update vectorial force */
 459             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 460             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 461             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 462
 463             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 464             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 465             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 466
 467             /**************************
 468              * CALCULATE INTERACTIONS *
 469              **************************/
 470
 471             /* Compute parameters for interactions between i and j atoms */
 472             qq20             = _mm_mul_ps(iq2,jq0);
 473
 474             /* COULOMB ELECTROSTATICS */
 475             velec            = _mm_mul_ps(qq20,rinv20);
 476             felec            = _mm_mul_ps(velec,rinvsq20);
 477
 478             /* Update potential sum for this i atom from the interaction with this j atom. */
 479             velec            = _mm_andnot_ps(dummy_mask,velec);
 480             velecsum         = _mm_add_ps(velecsum,velec);
 481
 482             fscal            = felec;
 483
 484             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 485
 486              /* Update vectorial force */
 487             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 488             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 489             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 490
 491             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 492             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 493             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 494
 495             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 496             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 497             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 498             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 499
 500             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 501
 502             /* Inner loop uses 105 flops */
 503         }
 504
 505         /* End of innermost loop */
 506
 507         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 508                                               f+i_coord_offset,fshift+i_shift_offset);
 509
 510         ggid                        = gid[iidx];
 511         /* Update potential energies */
 512         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 513         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 514
 515         /* Increment number of inner iterations */
 516         inneriter                  += j_index_end - j_index_start;
 517
 518         /* Outer loop uses 20 flops */
 519     }
 520
 521     /* Increment number of outer iterations */
 522     outeriter        += nri;
 523
 524     /* Update outer/inner flops */
 525
 526     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*105);
 527 }
 528 /*
 529  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_avx_128_fma_single
 530  * Electrostatics interaction: Coulomb
 531  * VdW interaction:            LennardJones
 532  * Geometry:                   Water3-Particle
 533  * Calculate force/pot:        Force
 534  */
 535 void
 536 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_avx_128_fma_single
 537                     (t_nblist                    * gmx_restrict       nlist,
 538                      rvec                        * gmx_restrict          xx,
 539                      rvec                        * gmx_restrict          ff,
 540                      t_forcerec                  * gmx_restrict          fr,
 541                      t_mdatoms                   * gmx_restrict     mdatoms,
 542                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 543                      t_nrnb                      * gmx_restrict        nrnb)
 544 {
 545     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 546      * just 0 for non-waters.
 547      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 548      * jnr indices corresponding to data put in the four positions in the SIMD register.
 549      */
 550     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 551     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 552     int              jnrA,jnrB,jnrC,jnrD;
 553     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 554     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 555     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 556     real             rcutoff_scalar;
 557     real             *shiftvec,*fshift,*x,*f;
 558     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 559     real             scratch[4*DIM];
 560     __m128           fscal,rcutoff,rcutoff2,jidxall;
 561     int              vdwioffset0;
 562     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 563     int              vdwioffset1;
 564     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 565     int              vdwioffset2;
 566     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 567     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 568     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 569     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 570     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 571     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 572     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 573     real             *charge;
 574     int              nvdwtype;
 575     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 576     int              *vdwtype;
 577     real             *vdwparam;
 578     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 579     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 580     __m128           dummy_mask,cutoff_mask;
 581     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 582     __m128           one     = _mm_set1_ps(1.0);
 583     __m128           two     = _mm_set1_ps(2.0);
 584     x                = xx[0];
 585     f                = ff[0];
 586
 587     nri              = nlist->nri;
 588     iinr             = nlist->iinr;
 589     jindex           = nlist->jindex;
 590     jjnr             = nlist->jjnr;
 591     shiftidx         = nlist->shift;
 592     gid              = nlist->gid;
 593     shiftvec         = fr->shift_vec[0];
 594     fshift           = fr->fshift[0];
 595     facel            = _mm_set1_ps(fr->epsfac);
 596     charge           = mdatoms->chargeA;
 597     nvdwtype         = fr->ntype;
 598     vdwparam         = fr->nbfp;
 599     vdwtype          = mdatoms->typeA;
 600
 601     /* Setup water-specific parameters */
 602     inr              = nlist->iinr[0];
 603     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 604     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 605     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 606     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 607
 608     /* Avoid stupid compiler warnings */
 609     jnrA = jnrB = jnrC = jnrD = 0;
 610     j_coord_offsetA = 0;
 611     j_coord_offsetB = 0;
 612     j_coord_offsetC = 0;
 613     j_coord_offsetD = 0;
 614
 615     outeriter        = 0;
 616     inneriter        = 0;
 617
 618     for(iidx=0;iidx<4*DIM;iidx++)
 619     {
 620         scratch[iidx] = 0.0;
 621     }
 622
 623     /* Start outer loop over neighborlists */
 624     for(iidx=0; iidx<nri; iidx++)
 625     {
 626         /* Load shift vector for this list */
 627         i_shift_offset   = DIM*shiftidx[iidx];
 628
 629         /* Load limits for loop over neighbors */
 630         j_index_start    = jindex[iidx];
 631         j_index_end      = jindex[iidx+1];
 632
 633         /* Get outer coordinate index */
 634         inr              = iinr[iidx];
 635         i_coord_offset   = DIM*inr;
 636
 637         /* Load i particle coords and add shift vector */
 638         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 639                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 640
 641         fix0             = _mm_setzero_ps();
 642         fiy0             = _mm_setzero_ps();
 643         fiz0             = _mm_setzero_ps();
 644         fix1             = _mm_setzero_ps();
 645         fiy1             = _mm_setzero_ps();
 646         fiz1             = _mm_setzero_ps();
 647         fix2             = _mm_setzero_ps();
 648         fiy2             = _mm_setzero_ps();
 649         fiz2             = _mm_setzero_ps();
 650
 651         /* Start inner kernel loop */
 652         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 653         {
 654
 655             /* Get j neighbor index, and coordinate index */
 656             jnrA             = jjnr[jidx];
 657             jnrB             = jjnr[jidx+1];
 658             jnrC             = jjnr[jidx+2];
 659             jnrD             = jjnr[jidx+3];
 660             j_coord_offsetA  = DIM*jnrA;
 661             j_coord_offsetB  = DIM*jnrB;
 662             j_coord_offsetC  = DIM*jnrC;
 663             j_coord_offsetD  = DIM*jnrD;
 664
 665             /* load j atom coordinates */
 666             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 667                                               x+j_coord_offsetC,x+j_coord_offsetD,
 668                                               &jx0,&jy0,&jz0);
 669
 670             /* Calculate displacement vector */
 671             dx00             = _mm_sub_ps(ix0,jx0);
 672             dy00             = _mm_sub_ps(iy0,jy0);
 673             dz00             = _mm_sub_ps(iz0,jz0);
 674             dx10             = _mm_sub_ps(ix1,jx0);
 675             dy10             = _mm_sub_ps(iy1,jy0);
 676             dz10             = _mm_sub_ps(iz1,jz0);
 677             dx20             = _mm_sub_ps(ix2,jx0);
 678             dy20             = _mm_sub_ps(iy2,jy0);
 679             dz20             = _mm_sub_ps(iz2,jz0);
 680
 681             /* Calculate squared distance and things based on it */
 682             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 683             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 684             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 685
 686             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 687             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 688             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 689
 690             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 691             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 692             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 693
 694             /* Load parameters for j particles */
 695             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 696                                                               charge+jnrC+0,charge+jnrD+0);
 697             vdwjidx0A        = 2*vdwtype[jnrA+0];
 698             vdwjidx0B        = 2*vdwtype[jnrB+0];
 699             vdwjidx0C        = 2*vdwtype[jnrC+0];
 700             vdwjidx0D        = 2*vdwtype[jnrD+0];
 701
 702             fjx0             = _mm_setzero_ps();
 703             fjy0             = _mm_setzero_ps();
 704             fjz0             = _mm_setzero_ps();
 705
 706             /**************************
 707              * CALCULATE INTERACTIONS *
 708              **************************/
 709
 710             /* Compute parameters for interactions between i and j atoms */
 711             qq00             = _mm_mul_ps(iq0,jq0);
 712             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 713                                          vdwparam+vdwioffset0+vdwjidx0B,
 714                                          vdwparam+vdwioffset0+vdwjidx0C,
 715                                          vdwparam+vdwioffset0+vdwjidx0D,
 716                                          &c6_00,&c12_00);
 717
 718             /* COULOMB ELECTROSTATICS */
 719             velec            = _mm_mul_ps(qq00,rinv00);
 720             felec            = _mm_mul_ps(velec,rinvsq00);
 721
 722             /* LENNARD-JONES DISPERSION/REPULSION */
 723
 724             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 725             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 726
 727             fscal            = _mm_add_ps(felec,fvdw);
 728
 729              /* Update vectorial force */
 730             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 731             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 732             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 733
 734             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 735             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 736             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 737
 738             /**************************
 739              * CALCULATE INTERACTIONS *
 740              **************************/
 741
 742             /* Compute parameters for interactions between i and j atoms */
 743             qq10             = _mm_mul_ps(iq1,jq0);
 744
 745             /* COULOMB ELECTROSTATICS */
 746             velec            = _mm_mul_ps(qq10,rinv10);
 747             felec            = _mm_mul_ps(velec,rinvsq10);
 748
 749             fscal            = felec;
 750
 751              /* Update vectorial force */
 752             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 753             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 754             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 755
 756             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 757             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 758             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 759
 760             /**************************
 761              * CALCULATE INTERACTIONS *
 762              **************************/
 763
 764             /* Compute parameters for interactions between i and j atoms */
 765             qq20             = _mm_mul_ps(iq2,jq0);
 766
 767             /* COULOMB ELECTROSTATICS */
 768             velec            = _mm_mul_ps(qq20,rinv20);
 769             felec            = _mm_mul_ps(velec,rinvsq20);
 770
 771             fscal            = felec;
 772
 773              /* Update vectorial force */
 774             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 775             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 776             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 777
 778             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 779             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 780             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 781
 782             fjptrA             = f+j_coord_offsetA;
 783             fjptrB             = f+j_coord_offsetB;
 784             fjptrC             = f+j_coord_offsetC;
 785             fjptrD             = f+j_coord_offsetD;
 786
 787             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 788
 789             /* Inner loop uses 97 flops */
 790         }
 791
 792         if(jidx<j_index_end)
 793         {
 794
 795             /* Get j neighbor index, and coordinate index */
 796             jnrlistA         = jjnr[jidx];
 797             jnrlistB         = jjnr[jidx+1];
 798             jnrlistC         = jjnr[jidx+2];
 799             jnrlistD         = jjnr[jidx+3];
 800             /* Sign of each element will be negative for non-real atoms.
 801              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 802              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 803              */
 804             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 805             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 806             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 807             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 808             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 809             j_coord_offsetA  = DIM*jnrA;
 810             j_coord_offsetB  = DIM*jnrB;
 811             j_coord_offsetC  = DIM*jnrC;
 812             j_coord_offsetD  = DIM*jnrD;
 813
 814             /* load j atom coordinates */
 815             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 816                                               x+j_coord_offsetC,x+j_coord_offsetD,
 817                                               &jx0,&jy0,&jz0);
 818
 819             /* Calculate displacement vector */
 820             dx00             = _mm_sub_ps(ix0,jx0);
 821             dy00             = _mm_sub_ps(iy0,jy0);
 822             dz00             = _mm_sub_ps(iz0,jz0);
 823             dx10             = _mm_sub_ps(ix1,jx0);
 824             dy10             = _mm_sub_ps(iy1,jy0);
 825             dz10             = _mm_sub_ps(iz1,jz0);
 826             dx20             = _mm_sub_ps(ix2,jx0);
 827             dy20             = _mm_sub_ps(iy2,jy0);
 828             dz20             = _mm_sub_ps(iz2,jz0);
 829
 830             /* Calculate squared distance and things based on it */
 831             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 832             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 833             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 834
 835             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 836             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 837             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 838
 839             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 840             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 841             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 842
 843             /* Load parameters for j particles */
 844             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 845                                                               charge+jnrC+0,charge+jnrD+0);
 846             vdwjidx0A        = 2*vdwtype[jnrA+0];
 847             vdwjidx0B        = 2*vdwtype[jnrB+0];
 848             vdwjidx0C        = 2*vdwtype[jnrC+0];
 849             vdwjidx0D        = 2*vdwtype[jnrD+0];
 850
 851             fjx0             = _mm_setzero_ps();
 852             fjy0             = _mm_setzero_ps();
 853             fjz0             = _mm_setzero_ps();
 854
 855             /**************************
 856              * CALCULATE INTERACTIONS *
 857              **************************/
 858
 859             /* Compute parameters for interactions between i and j atoms */
 860             qq00             = _mm_mul_ps(iq0,jq0);
 861             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 862                                          vdwparam+vdwioffset0+vdwjidx0B,
 863                                          vdwparam+vdwioffset0+vdwjidx0C,
 864                                          vdwparam+vdwioffset0+vdwjidx0D,
 865                                          &c6_00,&c12_00);
 866
 867             /* COULOMB ELECTROSTATICS */
 868             velec            = _mm_mul_ps(qq00,rinv00);
 869             felec            = _mm_mul_ps(velec,rinvsq00);
 870
 871             /* LENNARD-JONES DISPERSION/REPULSION */
 872
 873             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 874             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 875
 876             fscal            = _mm_add_ps(felec,fvdw);
 877
 878             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 879
 880              /* Update vectorial force */
 881             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 882             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 883             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 884
 885             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 886             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 887             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 888
 889             /**************************
 890              * CALCULATE INTERACTIONS *
 891              **************************/
 892
 893             /* Compute parameters for interactions between i and j atoms */
 894             qq10             = _mm_mul_ps(iq1,jq0);
 895
 896             /* COULOMB ELECTROSTATICS */
 897             velec            = _mm_mul_ps(qq10,rinv10);
 898             felec            = _mm_mul_ps(velec,rinvsq10);
 899
 900             fscal            = felec;
 901
 902             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 903
 904              /* Update vectorial force */
 905             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 906             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 907             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 908
 909             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 910             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 911             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 912
 913             /**************************
 914              * CALCULATE INTERACTIONS *
 915              **************************/
 916
 917             /* Compute parameters for interactions between i and j atoms */
 918             qq20             = _mm_mul_ps(iq2,jq0);
 919
 920             /* COULOMB ELECTROSTATICS */
 921             velec            = _mm_mul_ps(qq20,rinv20);
 922             felec            = _mm_mul_ps(velec,rinvsq20);
 923
 924             fscal            = felec;
 925
 926             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 927
 928              /* Update vectorial force */
 929             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 930             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 931             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 932
 933             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 934             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 935             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 936
 937             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 938             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 939             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 940             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 941
 942             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 943
 944             /* Inner loop uses 97 flops */
 945         }
 946
 947         /* End of innermost loop */
 948
 949         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 950                                               f+i_coord_offset,fshift+i_shift_offset);
 951
 952         /* Increment number of inner iterations */
 953         inneriter                  += j_index_end - j_index_start;
 954
 955         /* Outer loop uses 18 flops */
 956     }
 957
 958     /* Increment number of outer iterations */
 959     outeriter        += nri;
 960
 961     /* Update outer/inner flops */
 962
 963     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*97);
 964 }