src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_avx_128_fma_double.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_double 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_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_VF_avx_128_fma_double
  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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 100     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 101     __m128d          dummy_mask,cutoff_mask;
 102     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 103     __m128d          one     = _mm_set1_pd(1.0);
 104     __m128d          two     = _mm_set1_pd(2.0);
 105     x                = xx[0];
 106     f                = ff[0];
 107
 108     nri              = nlist->nri;
 109     iinr             = nlist->iinr;
 110     jindex           = nlist->jindex;
 111     jjnr             = nlist->jjnr;
 112     shiftidx         = nlist->shift;
 113     gid              = nlist->gid;
 114     shiftvec         = fr->shift_vec[0];
 115     fshift           = fr->fshift[0];
 116     facel            = _mm_set1_pd(fr->epsfac);
 117     charge           = mdatoms->chargeA;
 118     krf              = _mm_set1_pd(fr->ic->k_rf);
 119     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 120     crf              = _mm_set1_pd(fr->ic->c_rf);
 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_pd(facel,_mm_set1_pd(charge[inr+0]));
 128     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 129     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 130     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 131
 132     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 133     rcutoff_scalar   = fr->rcoulomb;
 134     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 135     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 136
 137     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 138     rvdw             = _mm_set1_pd(fr->rvdw);
 139
 140     /* Avoid stupid compiler warnings */
 141     jnrA = jnrB = 0;
 142     j_coord_offsetA = 0;
 143     j_coord_offsetB = 0;
 144
 145     outeriter        = 0;
 146     inneriter        = 0;
 147
 148     /* Start outer loop over neighborlists */
 149     for(iidx=0; iidx<nri; iidx++)
 150     {
 151         /* Load shift vector for this list */
 152         i_shift_offset   = DIM*shiftidx[iidx];
 153
 154         /* Load limits for loop over neighbors */
 155         j_index_start    = jindex[iidx];
 156         j_index_end      = jindex[iidx+1];
 157
 158         /* Get outer coordinate index */
 159         inr              = iinr[iidx];
 160         i_coord_offset   = DIM*inr;
 161
 162         /* Load i particle coords and add shift vector */
 163         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 164                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 165
 166         fix0             = _mm_setzero_pd();
 167         fiy0             = _mm_setzero_pd();
 168         fiz0             = _mm_setzero_pd();
 169         fix1             = _mm_setzero_pd();
 170         fiy1             = _mm_setzero_pd();
 171         fiz1             = _mm_setzero_pd();
 172         fix2             = _mm_setzero_pd();
 173         fiy2             = _mm_setzero_pd();
 174         fiz2             = _mm_setzero_pd();
 175
 176         /* Reset potential sums */
 177         velecsum         = _mm_setzero_pd();
 178         vvdwsum          = _mm_setzero_pd();
 179
 180         /* Start inner kernel loop */
 181         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 182         {
 183
 184             /* Get j neighbor index, and coordinate index */
 185             jnrA             = jjnr[jidx];
 186             jnrB             = jjnr[jidx+1];
 187             j_coord_offsetA  = DIM*jnrA;
 188             j_coord_offsetB  = DIM*jnrB;
 189
 190             /* load j atom coordinates */
 191             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 192                                               &jx0,&jy0,&jz0);
 193
 194             /* Calculate displacement vector */
 195             dx00             = _mm_sub_pd(ix0,jx0);
 196             dy00             = _mm_sub_pd(iy0,jy0);
 197             dz00             = _mm_sub_pd(iz0,jz0);
 198             dx10             = _mm_sub_pd(ix1,jx0);
 199             dy10             = _mm_sub_pd(iy1,jy0);
 200             dz10             = _mm_sub_pd(iz1,jz0);
 201             dx20             = _mm_sub_pd(ix2,jx0);
 202             dy20             = _mm_sub_pd(iy2,jy0);
 203             dz20             = _mm_sub_pd(iz2,jz0);
 204
 205             /* Calculate squared distance and things based on it */
 206             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 207             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 208             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 209
 210             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 211             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 212             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 213
 214             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 215             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 216             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 217
 218             /* Load parameters for j particles */
 219             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 220             vdwjidx0A        = 2*vdwtype[jnrA+0];
 221             vdwjidx0B        = 2*vdwtype[jnrB+0];
 222
 223             fjx0             = _mm_setzero_pd();
 224             fjy0             = _mm_setzero_pd();
 225             fjz0             = _mm_setzero_pd();
 226
 227             /**************************
 228              * CALCULATE INTERACTIONS *
 229              **************************/
 230
 231             if (gmx_mm_any_lt(rsq00,rcutoff2))
 232             {
 233
 234             /* Compute parameters for interactions between i and j atoms */
 235             qq00             = _mm_mul_pd(iq0,jq0);
 236             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 237                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 238
 239             /* REACTION-FIELD ELECTROSTATICS */
 240             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 241             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 242
 243             /* LENNARD-JONES DISPERSION/REPULSION */
 244
 245             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 246             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 247             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 248             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 249                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 250             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 251
 252             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 253
 254             /* Update potential sum for this i atom from the interaction with this j atom. */
 255             velec            = _mm_and_pd(velec,cutoff_mask);
 256             velecsum         = _mm_add_pd(velecsum,velec);
 257             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 258             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 259
 260             fscal            = _mm_add_pd(felec,fvdw);
 261
 262             fscal            = _mm_and_pd(fscal,cutoff_mask);
 263
 264             /* Update vectorial force */
 265             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 266             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 267             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 268
 269             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 270             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 271             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 272
 273             }
 274
 275             /**************************
 276              * CALCULATE INTERACTIONS *
 277              **************************/
 278
 279             if (gmx_mm_any_lt(rsq10,rcutoff2))
 280             {
 281
 282             /* Compute parameters for interactions between i and j atoms */
 283             qq10             = _mm_mul_pd(iq1,jq0);
 284
 285             /* REACTION-FIELD ELECTROSTATICS */
 286             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 287             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 288
 289             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 290
 291             /* Update potential sum for this i atom from the interaction with this j atom. */
 292             velec            = _mm_and_pd(velec,cutoff_mask);
 293             velecsum         = _mm_add_pd(velecsum,velec);
 294
 295             fscal            = felec;
 296
 297             fscal            = _mm_and_pd(fscal,cutoff_mask);
 298
 299             /* Update vectorial force */
 300             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 301             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 302             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 303
 304             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 305             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 306             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 307
 308             }
 309
 310             /**************************
 311              * CALCULATE INTERACTIONS *
 312              **************************/
 313
 314             if (gmx_mm_any_lt(rsq20,rcutoff2))
 315             {
 316
 317             /* Compute parameters for interactions between i and j atoms */
 318             qq20             = _mm_mul_pd(iq2,jq0);
 319
 320             /* REACTION-FIELD ELECTROSTATICS */
 321             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 322             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 323
 324             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 325
 326             /* Update potential sum for this i atom from the interaction with this j atom. */
 327             velec            = _mm_and_pd(velec,cutoff_mask);
 328             velecsum         = _mm_add_pd(velecsum,velec);
 329
 330             fscal            = felec;
 331
 332             fscal            = _mm_and_pd(fscal,cutoff_mask);
 333
 334             /* Update vectorial force */
 335             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 336             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 337             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 338
 339             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 340             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 341             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 342
 343             }
 344
 345             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 346
 347             /* Inner loop uses 138 flops */
 348         }
 349
 350         if(jidx<j_index_end)
 351         {
 352
 353             jnrA             = jjnr[jidx];
 354             j_coord_offsetA  = DIM*jnrA;
 355
 356             /* load j atom coordinates */
 357             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 358                                               &jx0,&jy0,&jz0);
 359
 360             /* Calculate displacement vector */
 361             dx00             = _mm_sub_pd(ix0,jx0);
 362             dy00             = _mm_sub_pd(iy0,jy0);
 363             dz00             = _mm_sub_pd(iz0,jz0);
 364             dx10             = _mm_sub_pd(ix1,jx0);
 365             dy10             = _mm_sub_pd(iy1,jy0);
 366             dz10             = _mm_sub_pd(iz1,jz0);
 367             dx20             = _mm_sub_pd(ix2,jx0);
 368             dy20             = _mm_sub_pd(iy2,jy0);
 369             dz20             = _mm_sub_pd(iz2,jz0);
 370
 371             /* Calculate squared distance and things based on it */
 372             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 373             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 374             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 375
 376             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 377             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 378             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 379
 380             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 381             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 382             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 383
 384             /* Load parameters for j particles */
 385             jq0              = _mm_load_sd(charge+jnrA+0);
 386             vdwjidx0A        = 2*vdwtype[jnrA+0];
 387
 388             fjx0             = _mm_setzero_pd();
 389             fjy0             = _mm_setzero_pd();
 390             fjz0             = _mm_setzero_pd();
 391
 392             /**************************
 393              * CALCULATE INTERACTIONS *
 394              **************************/
 395
 396             if (gmx_mm_any_lt(rsq00,rcutoff2))
 397             {
 398
 399             /* Compute parameters for interactions between i and j atoms */
 400             qq00             = _mm_mul_pd(iq0,jq0);
 401             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 402
 403             /* REACTION-FIELD ELECTROSTATICS */
 404             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 405             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 406
 407             /* LENNARD-JONES DISPERSION/REPULSION */
 408
 409             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 410             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 411             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 412             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 413                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 414             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 415
 416             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 417
 418             /* Update potential sum for this i atom from the interaction with this j atom. */
 419             velec            = _mm_and_pd(velec,cutoff_mask);
 420             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 421             velecsum         = _mm_add_pd(velecsum,velec);
 422             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 423             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 424             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 425
 426             fscal            = _mm_add_pd(felec,fvdw);
 427
 428             fscal            = _mm_and_pd(fscal,cutoff_mask);
 429
 430             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 431
 432             /* Update vectorial force */
 433             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 434             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 435             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 436
 437             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 438             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 439             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 440
 441             }
 442
 443             /**************************
 444              * CALCULATE INTERACTIONS *
 445              **************************/
 446
 447             if (gmx_mm_any_lt(rsq10,rcutoff2))
 448             {
 449
 450             /* Compute parameters for interactions between i and j atoms */
 451             qq10             = _mm_mul_pd(iq1,jq0);
 452
 453             /* REACTION-FIELD ELECTROSTATICS */
 454             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 455             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 456
 457             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 458
 459             /* Update potential sum for this i atom from the interaction with this j atom. */
 460             velec            = _mm_and_pd(velec,cutoff_mask);
 461             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 462             velecsum         = _mm_add_pd(velecsum,velec);
 463
 464             fscal            = felec;
 465
 466             fscal            = _mm_and_pd(fscal,cutoff_mask);
 467
 468             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 469
 470             /* Update vectorial force */
 471             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 472             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 473             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 474
 475             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 476             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 477             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 478
 479             }
 480
 481             /**************************
 482              * CALCULATE INTERACTIONS *
 483              **************************/
 484
 485             if (gmx_mm_any_lt(rsq20,rcutoff2))
 486             {
 487
 488             /* Compute parameters for interactions between i and j atoms */
 489             qq20             = _mm_mul_pd(iq2,jq0);
 490
 491             /* REACTION-FIELD ELECTROSTATICS */
 492             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 493             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 494
 495             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 496
 497             /* Update potential sum for this i atom from the interaction with this j atom. */
 498             velec            = _mm_and_pd(velec,cutoff_mask);
 499             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 500             velecsum         = _mm_add_pd(velecsum,velec);
 501
 502             fscal            = felec;
 503
 504             fscal            = _mm_and_pd(fscal,cutoff_mask);
 505
 506             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 507
 508             /* Update vectorial force */
 509             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 510             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 511             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 512
 513             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 514             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 515             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 516
 517             }
 518
 519             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 520
 521             /* Inner loop uses 138 flops */
 522         }
 523
 524         /* End of innermost loop */
 525
 526         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 527                                               f+i_coord_offset,fshift+i_shift_offset);
 528
 529         ggid                        = gid[iidx];
 530         /* Update potential energies */
 531         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 532         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 533
 534         /* Increment number of inner iterations */
 535         inneriter                  += j_index_end - j_index_start;
 536
 537         /* Outer loop uses 20 flops */
 538     }
 539
 540     /* Increment number of outer iterations */
 541     outeriter        += nri;
 542
 543     /* Update outer/inner flops */
 544
 545     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*138);
 546 }
 547 /*
 548  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_avx_128_fma_double
 549  * Electrostatics interaction: ReactionField
 550  * VdW interaction:            LennardJones
 551  * Geometry:                   Water3-Particle
 552  * Calculate force/pot:        Force
 553  */
 554 void
 555 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_avx_128_fma_double
 556                     (t_nblist                    * gmx_restrict       nlist,
 557                      rvec                        * gmx_restrict          xx,
 558                      rvec                        * gmx_restrict          ff,
 559                      t_forcerec                  * gmx_restrict          fr,
 560                      t_mdatoms                   * gmx_restrict     mdatoms,
 561                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 562                      t_nrnb                      * gmx_restrict        nrnb)
 563 {
 564     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 565      * just 0 for non-waters.
 566      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 567      * jnr indices corresponding to data put in the four positions in the SIMD register.
 568      */
 569     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 570     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 571     int              jnrA,jnrB;
 572     int              j_coord_offsetA,j_coord_offsetB;
 573     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 574     real             rcutoff_scalar;
 575     real             *shiftvec,*fshift,*x,*f;
 576     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 577     int              vdwioffset0;
 578     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 579     int              vdwioffset1;
 580     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 581     int              vdwioffset2;
 582     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 583     int              vdwjidx0A,vdwjidx0B;
 584     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 585     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 586     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 587     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 588     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 589     real             *charge;
 590     int              nvdwtype;
 591     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 592     int              *vdwtype;
 593     real             *vdwparam;
 594     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 595     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 596     __m128d          dummy_mask,cutoff_mask;
 597     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 598     __m128d          one     = _mm_set1_pd(1.0);
 599     __m128d          two     = _mm_set1_pd(2.0);
 600     x                = xx[0];
 601     f                = ff[0];
 602
 603     nri              = nlist->nri;
 604     iinr             = nlist->iinr;
 605     jindex           = nlist->jindex;
 606     jjnr             = nlist->jjnr;
 607     shiftidx         = nlist->shift;
 608     gid              = nlist->gid;
 609     shiftvec         = fr->shift_vec[0];
 610     fshift           = fr->fshift[0];
 611     facel            = _mm_set1_pd(fr->epsfac);
 612     charge           = mdatoms->chargeA;
 613     krf              = _mm_set1_pd(fr->ic->k_rf);
 614     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 615     crf              = _mm_set1_pd(fr->ic->c_rf);
 616     nvdwtype         = fr->ntype;
 617     vdwparam         = fr->nbfp;
 618     vdwtype          = mdatoms->typeA;
 619
 620     /* Setup water-specific parameters */
 621     inr              = nlist->iinr[0];
 622     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 623     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 624     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 625     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 626
 627     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 628     rcutoff_scalar   = fr->rcoulomb;
 629     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 630     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 631
 632     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 633     rvdw             = _mm_set1_pd(fr->rvdw);
 634
 635     /* Avoid stupid compiler warnings */
 636     jnrA = jnrB = 0;
 637     j_coord_offsetA = 0;
 638     j_coord_offsetB = 0;
 639
 640     outeriter        = 0;
 641     inneriter        = 0;
 642
 643     /* Start outer loop over neighborlists */
 644     for(iidx=0; iidx<nri; iidx++)
 645     {
 646         /* Load shift vector for this list */
 647         i_shift_offset   = DIM*shiftidx[iidx];
 648
 649         /* Load limits for loop over neighbors */
 650         j_index_start    = jindex[iidx];
 651         j_index_end      = jindex[iidx+1];
 652
 653         /* Get outer coordinate index */
 654         inr              = iinr[iidx];
 655         i_coord_offset   = DIM*inr;
 656
 657         /* Load i particle coords and add shift vector */
 658         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 659                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 660
 661         fix0             = _mm_setzero_pd();
 662         fiy0             = _mm_setzero_pd();
 663         fiz0             = _mm_setzero_pd();
 664         fix1             = _mm_setzero_pd();
 665         fiy1             = _mm_setzero_pd();
 666         fiz1             = _mm_setzero_pd();
 667         fix2             = _mm_setzero_pd();
 668         fiy2             = _mm_setzero_pd();
 669         fiz2             = _mm_setzero_pd();
 670
 671         /* Start inner kernel loop */
 672         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 673         {
 674
 675             /* Get j neighbor index, and coordinate index */
 676             jnrA             = jjnr[jidx];
 677             jnrB             = jjnr[jidx+1];
 678             j_coord_offsetA  = DIM*jnrA;
 679             j_coord_offsetB  = DIM*jnrB;
 680
 681             /* load j atom coordinates */
 682             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 683                                               &jx0,&jy0,&jz0);
 684
 685             /* Calculate displacement vector */
 686             dx00             = _mm_sub_pd(ix0,jx0);
 687             dy00             = _mm_sub_pd(iy0,jy0);
 688             dz00             = _mm_sub_pd(iz0,jz0);
 689             dx10             = _mm_sub_pd(ix1,jx0);
 690             dy10             = _mm_sub_pd(iy1,jy0);
 691             dz10             = _mm_sub_pd(iz1,jz0);
 692             dx20             = _mm_sub_pd(ix2,jx0);
 693             dy20             = _mm_sub_pd(iy2,jy0);
 694             dz20             = _mm_sub_pd(iz2,jz0);
 695
 696             /* Calculate squared distance and things based on it */
 697             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 698             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 699             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 700
 701             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 702             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 703             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 704
 705             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 706             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 707             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 708
 709             /* Load parameters for j particles */
 710             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 711             vdwjidx0A        = 2*vdwtype[jnrA+0];
 712             vdwjidx0B        = 2*vdwtype[jnrB+0];
 713
 714             fjx0             = _mm_setzero_pd();
 715             fjy0             = _mm_setzero_pd();
 716             fjz0             = _mm_setzero_pd();
 717
 718             /**************************
 719              * CALCULATE INTERACTIONS *
 720              **************************/
 721
 722             if (gmx_mm_any_lt(rsq00,rcutoff2))
 723             {
 724
 725             /* Compute parameters for interactions between i and j atoms */
 726             qq00             = _mm_mul_pd(iq0,jq0);
 727             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 728                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 729
 730             /* REACTION-FIELD ELECTROSTATICS */
 731             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 732
 733             /* LENNARD-JONES DISPERSION/REPULSION */
 734
 735             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 736             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 737
 738             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 739
 740             fscal            = _mm_add_pd(felec,fvdw);
 741
 742             fscal            = _mm_and_pd(fscal,cutoff_mask);
 743
 744             /* Update vectorial force */
 745             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 746             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 747             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 748
 749             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 750             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 751             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 752
 753             }
 754
 755             /**************************
 756              * CALCULATE INTERACTIONS *
 757              **************************/
 758
 759             if (gmx_mm_any_lt(rsq10,rcutoff2))
 760             {
 761
 762             /* Compute parameters for interactions between i and j atoms */
 763             qq10             = _mm_mul_pd(iq1,jq0);
 764
 765             /* REACTION-FIELD ELECTROSTATICS */
 766             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 767
 768             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 769
 770             fscal            = felec;
 771
 772             fscal            = _mm_and_pd(fscal,cutoff_mask);
 773
 774             /* Update vectorial force */
 775             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 776             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 777             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 778
 779             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 780             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 781             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 782
 783             }
 784
 785             /**************************
 786              * CALCULATE INTERACTIONS *
 787              **************************/
 788
 789             if (gmx_mm_any_lt(rsq20,rcutoff2))
 790             {
 791
 792             /* Compute parameters for interactions between i and j atoms */
 793             qq20             = _mm_mul_pd(iq2,jq0);
 794
 795             /* REACTION-FIELD ELECTROSTATICS */
 796             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 797
 798             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 799
 800             fscal            = felec;
 801
 802             fscal            = _mm_and_pd(fscal,cutoff_mask);
 803
 804             /* Update vectorial force */
 805             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 806             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 807             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 808
 809             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 810             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 811             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 812
 813             }
 814
 815             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 816
 817             /* Inner loop uses 109 flops */
 818         }
 819
 820         if(jidx<j_index_end)
 821         {
 822
 823             jnrA             = jjnr[jidx];
 824             j_coord_offsetA  = DIM*jnrA;
 825
 826             /* load j atom coordinates */
 827             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 828                                               &jx0,&jy0,&jz0);
 829
 830             /* Calculate displacement vector */
 831             dx00             = _mm_sub_pd(ix0,jx0);
 832             dy00             = _mm_sub_pd(iy0,jy0);
 833             dz00             = _mm_sub_pd(iz0,jz0);
 834             dx10             = _mm_sub_pd(ix1,jx0);
 835             dy10             = _mm_sub_pd(iy1,jy0);
 836             dz10             = _mm_sub_pd(iz1,jz0);
 837             dx20             = _mm_sub_pd(ix2,jx0);
 838             dy20             = _mm_sub_pd(iy2,jy0);
 839             dz20             = _mm_sub_pd(iz2,jz0);
 840
 841             /* Calculate squared distance and things based on it */
 842             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 843             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 844             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 845
 846             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 847             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 848             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 849
 850             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 851             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 852             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 853
 854             /* Load parameters for j particles */
 855             jq0              = _mm_load_sd(charge+jnrA+0);
 856             vdwjidx0A        = 2*vdwtype[jnrA+0];
 857
 858             fjx0             = _mm_setzero_pd();
 859             fjy0             = _mm_setzero_pd();
 860             fjz0             = _mm_setzero_pd();
 861
 862             /**************************
 863              * CALCULATE INTERACTIONS *
 864              **************************/
 865
 866             if (gmx_mm_any_lt(rsq00,rcutoff2))
 867             {
 868
 869             /* Compute parameters for interactions between i and j atoms */
 870             qq00             = _mm_mul_pd(iq0,jq0);
 871             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 872
 873             /* REACTION-FIELD ELECTROSTATICS */
 874             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 875
 876             /* LENNARD-JONES DISPERSION/REPULSION */
 877
 878             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 879             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 880
 881             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 882
 883             fscal            = _mm_add_pd(felec,fvdw);
 884
 885             fscal            = _mm_and_pd(fscal,cutoff_mask);
 886
 887             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 888
 889             /* Update vectorial force */
 890             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 891             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 892             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 893
 894             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 895             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 896             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 897
 898             }
 899
 900             /**************************
 901              * CALCULATE INTERACTIONS *
 902              **************************/
 903
 904             if (gmx_mm_any_lt(rsq10,rcutoff2))
 905             {
 906
 907             /* Compute parameters for interactions between i and j atoms */
 908             qq10             = _mm_mul_pd(iq1,jq0);
 909
 910             /* REACTION-FIELD ELECTROSTATICS */
 911             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 912
 913             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 914
 915             fscal            = felec;
 916
 917             fscal            = _mm_and_pd(fscal,cutoff_mask);
 918
 919             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 920
 921             /* Update vectorial force */
 922             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 923             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 924             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 925
 926             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 927             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 928             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 929
 930             }
 931
 932             /**************************
 933              * CALCULATE INTERACTIONS *
 934              **************************/
 935
 936             if (gmx_mm_any_lt(rsq20,rcutoff2))
 937             {
 938
 939             /* Compute parameters for interactions between i and j atoms */
 940             qq20             = _mm_mul_pd(iq2,jq0);
 941
 942             /* REACTION-FIELD ELECTROSTATICS */
 943             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 944
 945             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 946
 947             fscal            = felec;
 948
 949             fscal            = _mm_and_pd(fscal,cutoff_mask);
 950
 951             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 952
 953             /* Update vectorial force */
 954             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 955             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 956             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 957
 958             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 959             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 960             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 961
 962             }
 963
 964             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 965
 966             /* Inner loop uses 109 flops */
 967         }
 968
 969         /* End of innermost loop */
 970
 971         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 972                                               f+i_coord_offset,fshift+i_shift_offset);
 973
 974         /* Increment number of inner iterations */
 975         inneriter                  += j_index_end - j_index_start;
 976
 977         /* Outer loop uses 18 flops */
 978     }
 979
 980     /* Increment number of outer iterations */
 981     outeriter        += nri;
 982
 983     /* Update outer/inner flops */
 984
 985     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*109);
 986 }