src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  50 #include "kernelutil_x86_avx_128_fma_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_VF_avx_128_fma_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSw_VdwLJSw_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     __m128i          ewitab;
 105     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 106     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 107     real             *ewtab;
 108     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 109     real             rswitch_scalar,d_scalar;
 110     __m128           dummy_mask,cutoff_mask;
 111     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 112     __m128           one     = _mm_set1_ps(1.0);
 113     __m128           two     = _mm_set1_ps(2.0);
 114     x                = xx[0];
 115     f                = ff[0];
 116
 117     nri              = nlist->nri;
 118     iinr             = nlist->iinr;
 119     jindex           = nlist->jindex;
 120     jjnr             = nlist->jjnr;
 121     shiftidx         = nlist->shift;
 122     gid              = nlist->gid;
 123     shiftvec         = fr->shift_vec[0];
 124     fshift           = fr->fshift[0];
 125     facel            = _mm_set1_ps(fr->epsfac);
 126     charge           = mdatoms->chargeA;
 127     nvdwtype         = fr->ntype;
 128     vdwparam         = fr->nbfp;
 129     vdwtype          = mdatoms->typeA;
 130
 131     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 132     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
 133     beta2            = _mm_mul_ps(beta,beta);
 134     beta3            = _mm_mul_ps(beta,beta2);
 135     ewtab            = fr->ic->tabq_coul_FDV0;
 136     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 137     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 138
 139     /* Setup water-specific parameters */
 140     inr              = nlist->iinr[0];
 141     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 142     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 143     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 144     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 145
 146     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 147     rcutoff_scalar   = fr->rcoulomb;
 148     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 149     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 150
 151     rswitch_scalar   = fr->rcoulomb_switch;
 152     rswitch          = _mm_set1_ps(rswitch_scalar);
 153     /* Setup switch parameters */
 154     d_scalar         = rcutoff_scalar-rswitch_scalar;
 155     d                = _mm_set1_ps(d_scalar);
 156     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 157     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 158     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 159     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 160     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 161     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 162
 163     /* Avoid stupid compiler warnings */
 164     jnrA = jnrB = jnrC = jnrD = 0;
 165     j_coord_offsetA = 0;
 166     j_coord_offsetB = 0;
 167     j_coord_offsetC = 0;
 168     j_coord_offsetD = 0;
 169
 170     outeriter        = 0;
 171     inneriter        = 0;
 172
 173     for(iidx=0;iidx<4*DIM;iidx++)
 174     {
 175         scratch[iidx] = 0.0;
 176     }
 177
 178     /* Start outer loop over neighborlists */
 179     for(iidx=0; iidx<nri; iidx++)
 180     {
 181         /* Load shift vector for this list */
 182         i_shift_offset   = DIM*shiftidx[iidx];
 183
 184         /* Load limits for loop over neighbors */
 185         j_index_start    = jindex[iidx];
 186         j_index_end      = jindex[iidx+1];
 187
 188         /* Get outer coordinate index */
 189         inr              = iinr[iidx];
 190         i_coord_offset   = DIM*inr;
 191
 192         /* Load i particle coords and add shift vector */
 193         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 194                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 195
 196         fix0             = _mm_setzero_ps();
 197         fiy0             = _mm_setzero_ps();
 198         fiz0             = _mm_setzero_ps();
 199         fix1             = _mm_setzero_ps();
 200         fiy1             = _mm_setzero_ps();
 201         fiz1             = _mm_setzero_ps();
 202         fix2             = _mm_setzero_ps();
 203         fiy2             = _mm_setzero_ps();
 204         fiz2             = _mm_setzero_ps();
 205
 206         /* Reset potential sums */
 207         velecsum         = _mm_setzero_ps();
 208         vvdwsum          = _mm_setzero_ps();
 209
 210         /* Start inner kernel loop */
 211         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 212         {
 213
 214             /* Get j neighbor index, and coordinate index */
 215             jnrA             = jjnr[jidx];
 216             jnrB             = jjnr[jidx+1];
 217             jnrC             = jjnr[jidx+2];
 218             jnrD             = jjnr[jidx+3];
 219             j_coord_offsetA  = DIM*jnrA;
 220             j_coord_offsetB  = DIM*jnrB;
 221             j_coord_offsetC  = DIM*jnrC;
 222             j_coord_offsetD  = DIM*jnrD;
 223
 224             /* load j atom coordinates */
 225             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 226                                               x+j_coord_offsetC,x+j_coord_offsetD,
 227                                               &jx0,&jy0,&jz0);
 228
 229             /* Calculate displacement vector */
 230             dx00             = _mm_sub_ps(ix0,jx0);
 231             dy00             = _mm_sub_ps(iy0,jy0);
 232             dz00             = _mm_sub_ps(iz0,jz0);
 233             dx10             = _mm_sub_ps(ix1,jx0);
 234             dy10             = _mm_sub_ps(iy1,jy0);
 235             dz10             = _mm_sub_ps(iz1,jz0);
 236             dx20             = _mm_sub_ps(ix2,jx0);
 237             dy20             = _mm_sub_ps(iy2,jy0);
 238             dz20             = _mm_sub_ps(iz2,jz0);
 239
 240             /* Calculate squared distance and things based on it */
 241             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 242             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 243             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 244
 245             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 246             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 247             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 248
 249             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 250             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 251             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 252
 253             /* Load parameters for j particles */
 254             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 255                                                               charge+jnrC+0,charge+jnrD+0);
 256             vdwjidx0A        = 2*vdwtype[jnrA+0];
 257             vdwjidx0B        = 2*vdwtype[jnrB+0];
 258             vdwjidx0C        = 2*vdwtype[jnrC+0];
 259             vdwjidx0D        = 2*vdwtype[jnrD+0];
 260
 261             fjx0             = _mm_setzero_ps();
 262             fjy0             = _mm_setzero_ps();
 263             fjz0             = _mm_setzero_ps();
 264
 265             /**************************
 266              * CALCULATE INTERACTIONS *
 267              **************************/
 268
 269             if (gmx_mm_any_lt(rsq00,rcutoff2))
 270             {
 271
 272             r00              = _mm_mul_ps(rsq00,rinv00);
 273
 274             /* Compute parameters for interactions between i and j atoms */
 275             qq00             = _mm_mul_ps(iq0,jq0);
 276             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 277                                          vdwparam+vdwioffset0+vdwjidx0B,
 278                                          vdwparam+vdwioffset0+vdwjidx0C,
 279                                          vdwparam+vdwioffset0+vdwjidx0D,
 280                                          &c6_00,&c12_00);
 281
 282             /* EWALD ELECTROSTATICS */
 283
 284             /* Analytical PME correction */
 285             zeta2            = _mm_mul_ps(beta2,rsq00);
 286             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 287             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 288             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 289             felec            = _mm_mul_ps(qq00,felec);
 290             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 291             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 292             velec            = _mm_mul_ps(qq00,velec);
 293
 294             /* LENNARD-JONES DISPERSION/REPULSION */
 295
 296             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 297             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 298             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 299             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 300             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 301
 302             d                = _mm_sub_ps(r00,rswitch);
 303             d                = _mm_max_ps(d,_mm_setzero_ps());
 304             d2               = _mm_mul_ps(d,d);
 305             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 306
 307             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 308
 309             /* Evaluate switch function */
 310             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 311             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 312             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 313             velec            = _mm_mul_ps(velec,sw);
 314             vvdw             = _mm_mul_ps(vvdw,sw);
 315             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 316
 317             /* Update potential sum for this i atom from the interaction with this j atom. */
 318             velec            = _mm_and_ps(velec,cutoff_mask);
 319             velecsum         = _mm_add_ps(velecsum,velec);
 320             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 321             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 322
 323             fscal            = _mm_add_ps(felec,fvdw);
 324
 325             fscal            = _mm_and_ps(fscal,cutoff_mask);
 326
 327              /* Update vectorial force */
 328             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 329             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 330             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 331
 332             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 333             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 334             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 335
 336             }
 337
 338             /**************************
 339              * CALCULATE INTERACTIONS *
 340              **************************/
 341
 342             if (gmx_mm_any_lt(rsq10,rcutoff2))
 343             {
 344
 345             r10              = _mm_mul_ps(rsq10,rinv10);
 346
 347             /* Compute parameters for interactions between i and j atoms */
 348             qq10             = _mm_mul_ps(iq1,jq0);
 349
 350             /* EWALD ELECTROSTATICS */
 351
 352             /* Analytical PME correction */
 353             zeta2            = _mm_mul_ps(beta2,rsq10);
 354             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 355             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 356             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 357             felec            = _mm_mul_ps(qq10,felec);
 358             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 359             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 360             velec            = _mm_mul_ps(qq10,velec);
 361
 362             d                = _mm_sub_ps(r10,rswitch);
 363             d                = _mm_max_ps(d,_mm_setzero_ps());
 364             d2               = _mm_mul_ps(d,d);
 365             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 366
 367             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 368
 369             /* Evaluate switch function */
 370             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 371             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 372             velec            = _mm_mul_ps(velec,sw);
 373             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 374
 375             /* Update potential sum for this i atom from the interaction with this j atom. */
 376             velec            = _mm_and_ps(velec,cutoff_mask);
 377             velecsum         = _mm_add_ps(velecsum,velec);
 378
 379             fscal            = felec;
 380
 381             fscal            = _mm_and_ps(fscal,cutoff_mask);
 382
 383              /* Update vectorial force */
 384             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 385             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 386             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 387
 388             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 389             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 390             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 391
 392             }
 393
 394             /**************************
 395              * CALCULATE INTERACTIONS *
 396              **************************/
 397
 398             if (gmx_mm_any_lt(rsq20,rcutoff2))
 399             {
 400
 401             r20              = _mm_mul_ps(rsq20,rinv20);
 402
 403             /* Compute parameters for interactions between i and j atoms */
 404             qq20             = _mm_mul_ps(iq2,jq0);
 405
 406             /* EWALD ELECTROSTATICS */
 407
 408             /* Analytical PME correction */
 409             zeta2            = _mm_mul_ps(beta2,rsq20);
 410             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 411             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 412             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 413             felec            = _mm_mul_ps(qq20,felec);
 414             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 415             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 416             velec            = _mm_mul_ps(qq20,velec);
 417
 418             d                = _mm_sub_ps(r20,rswitch);
 419             d                = _mm_max_ps(d,_mm_setzero_ps());
 420             d2               = _mm_mul_ps(d,d);
 421             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 422
 423             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 424
 425             /* Evaluate switch function */
 426             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 427             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
 428             velec            = _mm_mul_ps(velec,sw);
 429             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 430
 431             /* Update potential sum for this i atom from the interaction with this j atom. */
 432             velec            = _mm_and_ps(velec,cutoff_mask);
 433             velecsum         = _mm_add_ps(velecsum,velec);
 434
 435             fscal            = felec;
 436
 437             fscal            = _mm_and_ps(fscal,cutoff_mask);
 438
 439              /* Update vectorial force */
 440             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 441             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 442             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 443
 444             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 445             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 446             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 447
 448             }
 449
 450             fjptrA             = f+j_coord_offsetA;
 451             fjptrB             = f+j_coord_offsetB;
 452             fjptrC             = f+j_coord_offsetC;
 453             fjptrD             = f+j_coord_offsetD;
 454
 455             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 456
 457             /* Inner loop uses 177 flops */
 458         }
 459
 460         if(jidx<j_index_end)
 461         {
 462
 463             /* Get j neighbor index, and coordinate index */
 464             jnrlistA         = jjnr[jidx];
 465             jnrlistB         = jjnr[jidx+1];
 466             jnrlistC         = jjnr[jidx+2];
 467             jnrlistD         = jjnr[jidx+3];
 468             /* Sign of each element will be negative for non-real atoms.
 469              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 470              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 471              */
 472             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 473             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 474             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 475             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 476             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 477             j_coord_offsetA  = DIM*jnrA;
 478             j_coord_offsetB  = DIM*jnrB;
 479             j_coord_offsetC  = DIM*jnrC;
 480             j_coord_offsetD  = DIM*jnrD;
 481
 482             /* load j atom coordinates */
 483             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 484                                               x+j_coord_offsetC,x+j_coord_offsetD,
 485                                               &jx0,&jy0,&jz0);
 486
 487             /* Calculate displacement vector */
 488             dx00             = _mm_sub_ps(ix0,jx0);
 489             dy00             = _mm_sub_ps(iy0,jy0);
 490             dz00             = _mm_sub_ps(iz0,jz0);
 491             dx10             = _mm_sub_ps(ix1,jx0);
 492             dy10             = _mm_sub_ps(iy1,jy0);
 493             dz10             = _mm_sub_ps(iz1,jz0);
 494             dx20             = _mm_sub_ps(ix2,jx0);
 495             dy20             = _mm_sub_ps(iy2,jy0);
 496             dz20             = _mm_sub_ps(iz2,jz0);
 497
 498             /* Calculate squared distance and things based on it */
 499             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 500             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 501             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 502
 503             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 504             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 505             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 506
 507             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 508             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 509             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 510
 511             /* Load parameters for j particles */
 512             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 513                                                               charge+jnrC+0,charge+jnrD+0);
 514             vdwjidx0A        = 2*vdwtype[jnrA+0];
 515             vdwjidx0B        = 2*vdwtype[jnrB+0];
 516             vdwjidx0C        = 2*vdwtype[jnrC+0];
 517             vdwjidx0D        = 2*vdwtype[jnrD+0];
 518
 519             fjx0             = _mm_setzero_ps();
 520             fjy0             = _mm_setzero_ps();
 521             fjz0             = _mm_setzero_ps();
 522
 523             /**************************
 524              * CALCULATE INTERACTIONS *
 525              **************************/
 526
 527             if (gmx_mm_any_lt(rsq00,rcutoff2))
 528             {
 529
 530             r00              = _mm_mul_ps(rsq00,rinv00);
 531             r00              = _mm_andnot_ps(dummy_mask,r00);
 532
 533             /* Compute parameters for interactions between i and j atoms */
 534             qq00             = _mm_mul_ps(iq0,jq0);
 535             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 536                                          vdwparam+vdwioffset0+vdwjidx0B,
 537                                          vdwparam+vdwioffset0+vdwjidx0C,
 538                                          vdwparam+vdwioffset0+vdwjidx0D,
 539                                          &c6_00,&c12_00);
 540
 541             /* EWALD ELECTROSTATICS */
 542
 543             /* Analytical PME correction */
 544             zeta2            = _mm_mul_ps(beta2,rsq00);
 545             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 546             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 547             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 548             felec            = _mm_mul_ps(qq00,felec);
 549             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 550             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 551             velec            = _mm_mul_ps(qq00,velec);
 552
 553             /* LENNARD-JONES DISPERSION/REPULSION */
 554
 555             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 556             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 557             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 558             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 559             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 560
 561             d                = _mm_sub_ps(r00,rswitch);
 562             d                = _mm_max_ps(d,_mm_setzero_ps());
 563             d2               = _mm_mul_ps(d,d);
 564             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 565
 566             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 567
 568             /* Evaluate switch function */
 569             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 570             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 571             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 572             velec            = _mm_mul_ps(velec,sw);
 573             vvdw             = _mm_mul_ps(vvdw,sw);
 574             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 575
 576             /* Update potential sum for this i atom from the interaction with this j atom. */
 577             velec            = _mm_and_ps(velec,cutoff_mask);
 578             velec            = _mm_andnot_ps(dummy_mask,velec);
 579             velecsum         = _mm_add_ps(velecsum,velec);
 580             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 581             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 582             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 583
 584             fscal            = _mm_add_ps(felec,fvdw);
 585
 586             fscal            = _mm_and_ps(fscal,cutoff_mask);
 587
 588             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 589
 590              /* Update vectorial force */
 591             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 592             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 593             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 594
 595             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 596             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 597             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 598
 599             }
 600
 601             /**************************
 602              * CALCULATE INTERACTIONS *
 603              **************************/
 604
 605             if (gmx_mm_any_lt(rsq10,rcutoff2))
 606             {
 607
 608             r10              = _mm_mul_ps(rsq10,rinv10);
 609             r10              = _mm_andnot_ps(dummy_mask,r10);
 610
 611             /* Compute parameters for interactions between i and j atoms */
 612             qq10             = _mm_mul_ps(iq1,jq0);
 613
 614             /* EWALD ELECTROSTATICS */
 615
 616             /* Analytical PME correction */
 617             zeta2            = _mm_mul_ps(beta2,rsq10);
 618             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 619             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 620             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 621             felec            = _mm_mul_ps(qq10,felec);
 622             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 623             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 624             velec            = _mm_mul_ps(qq10,velec);
 625
 626             d                = _mm_sub_ps(r10,rswitch);
 627             d                = _mm_max_ps(d,_mm_setzero_ps());
 628             d2               = _mm_mul_ps(d,d);
 629             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 630
 631             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 632
 633             /* Evaluate switch function */
 634             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 635             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 636             velec            = _mm_mul_ps(velec,sw);
 637             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 638
 639             /* Update potential sum for this i atom from the interaction with this j atom. */
 640             velec            = _mm_and_ps(velec,cutoff_mask);
 641             velec            = _mm_andnot_ps(dummy_mask,velec);
 642             velecsum         = _mm_add_ps(velecsum,velec);
 643
 644             fscal            = felec;
 645
 646             fscal            = _mm_and_ps(fscal,cutoff_mask);
 647
 648             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 649
 650              /* Update vectorial force */
 651             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 652             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 653             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 654
 655             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 656             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 657             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 658
 659             }
 660
 661             /**************************
 662              * CALCULATE INTERACTIONS *
 663              **************************/
 664
 665             if (gmx_mm_any_lt(rsq20,rcutoff2))
 666             {
 667
 668             r20              = _mm_mul_ps(rsq20,rinv20);
 669             r20              = _mm_andnot_ps(dummy_mask,r20);
 670
 671             /* Compute parameters for interactions between i and j atoms */
 672             qq20             = _mm_mul_ps(iq2,jq0);
 673
 674             /* EWALD ELECTROSTATICS */
 675
 676             /* Analytical PME correction */
 677             zeta2            = _mm_mul_ps(beta2,rsq20);
 678             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 679             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 680             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 681             felec            = _mm_mul_ps(qq20,felec);
 682             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 683             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 684             velec            = _mm_mul_ps(qq20,velec);
 685
 686             d                = _mm_sub_ps(r20,rswitch);
 687             d                = _mm_max_ps(d,_mm_setzero_ps());
 688             d2               = _mm_mul_ps(d,d);
 689             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 690
 691             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 692
 693             /* Evaluate switch function */
 694             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 695             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
 696             velec            = _mm_mul_ps(velec,sw);
 697             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 698
 699             /* Update potential sum for this i atom from the interaction with this j atom. */
 700             velec            = _mm_and_ps(velec,cutoff_mask);
 701             velec            = _mm_andnot_ps(dummy_mask,velec);
 702             velecsum         = _mm_add_ps(velecsum,velec);
 703
 704             fscal            = felec;
 705
 706             fscal            = _mm_and_ps(fscal,cutoff_mask);
 707
 708             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 709
 710              /* Update vectorial force */
 711             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 712             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 713             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 714
 715             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 716             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 717             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 718
 719             }
 720
 721             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 722             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 723             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 724             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 725
 726             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 727
 728             /* Inner loop uses 180 flops */
 729         }
 730
 731         /* End of innermost loop */
 732
 733         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 734                                               f+i_coord_offset,fshift+i_shift_offset);
 735
 736         ggid                        = gid[iidx];
 737         /* Update potential energies */
 738         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 739         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 740
 741         /* Increment number of inner iterations */
 742         inneriter                  += j_index_end - j_index_start;
 743
 744         /* Outer loop uses 20 flops */
 745     }
 746
 747     /* Increment number of outer iterations */
 748     outeriter        += nri;
 749
 750     /* Update outer/inner flops */
 751
 752     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*180);
 753 }
 754 /*
 755  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_avx_128_fma_single
 756  * Electrostatics interaction: Ewald
 757  * VdW interaction:            LennardJones
 758  * Geometry:                   Water3-Particle
 759  * Calculate force/pot:        Force
 760  */
 761 void
 762 nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_avx_128_fma_single
 763                     (t_nblist                    * gmx_restrict       nlist,
 764                      rvec                        * gmx_restrict          xx,
 765                      rvec                        * gmx_restrict          ff,
 766                      t_forcerec                  * gmx_restrict          fr,
 767                      t_mdatoms                   * gmx_restrict     mdatoms,
 768                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 769                      t_nrnb                      * gmx_restrict        nrnb)
 770 {
 771     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 772      * just 0 for non-waters.
 773      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 774      * jnr indices corresponding to data put in the four positions in the SIMD register.
 775      */
 776     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 777     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 778     int              jnrA,jnrB,jnrC,jnrD;
 779     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 780     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 781     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 782     real             rcutoff_scalar;
 783     real             *shiftvec,*fshift,*x,*f;
 784     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 785     real             scratch[4*DIM];
 786     __m128           fscal,rcutoff,rcutoff2,jidxall;
 787     int              vdwioffset0;
 788     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 789     int              vdwioffset1;
 790     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 791     int              vdwioffset2;
 792     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 793     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 794     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 795     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 796     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 797     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 798     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 799     real             *charge;
 800     int              nvdwtype;
 801     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 802     int              *vdwtype;
 803     real             *vdwparam;
 804     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 805     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 806     __m128i          ewitab;
 807     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 808     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 809     real             *ewtab;
 810     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 811     real             rswitch_scalar,d_scalar;
 812     __m128           dummy_mask,cutoff_mask;
 813     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 814     __m128           one     = _mm_set1_ps(1.0);
 815     __m128           two     = _mm_set1_ps(2.0);
 816     x                = xx[0];
 817     f                = ff[0];
 818
 819     nri              = nlist->nri;
 820     iinr             = nlist->iinr;
 821     jindex           = nlist->jindex;
 822     jjnr             = nlist->jjnr;
 823     shiftidx         = nlist->shift;
 824     gid              = nlist->gid;
 825     shiftvec         = fr->shift_vec[0];
 826     fshift           = fr->fshift[0];
 827     facel            = _mm_set1_ps(fr->epsfac);
 828     charge           = mdatoms->chargeA;
 829     nvdwtype         = fr->ntype;
 830     vdwparam         = fr->nbfp;
 831     vdwtype          = mdatoms->typeA;
 832
 833     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 834     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
 835     beta2            = _mm_mul_ps(beta,beta);
 836     beta3            = _mm_mul_ps(beta,beta2);
 837     ewtab            = fr->ic->tabq_coul_FDV0;
 838     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 839     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 840
 841     /* Setup water-specific parameters */
 842     inr              = nlist->iinr[0];
 843     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 844     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 845     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 846     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 847
 848     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 849     rcutoff_scalar   = fr->rcoulomb;
 850     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 851     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 852
 853     rswitch_scalar   = fr->rcoulomb_switch;
 854     rswitch          = _mm_set1_ps(rswitch_scalar);
 855     /* Setup switch parameters */
 856     d_scalar         = rcutoff_scalar-rswitch_scalar;
 857     d                = _mm_set1_ps(d_scalar);
 858     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 859     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 860     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 861     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 862     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 863     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 864
 865     /* Avoid stupid compiler warnings */
 866     jnrA = jnrB = jnrC = jnrD = 0;
 867     j_coord_offsetA = 0;
 868     j_coord_offsetB = 0;
 869     j_coord_offsetC = 0;
 870     j_coord_offsetD = 0;
 871
 872     outeriter        = 0;
 873     inneriter        = 0;
 874
 875     for(iidx=0;iidx<4*DIM;iidx++)
 876     {
 877         scratch[iidx] = 0.0;
 878     }
 879
 880     /* Start outer loop over neighborlists */
 881     for(iidx=0; iidx<nri; iidx++)
 882     {
 883         /* Load shift vector for this list */
 884         i_shift_offset   = DIM*shiftidx[iidx];
 885
 886         /* Load limits for loop over neighbors */
 887         j_index_start    = jindex[iidx];
 888         j_index_end      = jindex[iidx+1];
 889
 890         /* Get outer coordinate index */
 891         inr              = iinr[iidx];
 892         i_coord_offset   = DIM*inr;
 893
 894         /* Load i particle coords and add shift vector */
 895         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 896                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 897
 898         fix0             = _mm_setzero_ps();
 899         fiy0             = _mm_setzero_ps();
 900         fiz0             = _mm_setzero_ps();
 901         fix1             = _mm_setzero_ps();
 902         fiy1             = _mm_setzero_ps();
 903         fiz1             = _mm_setzero_ps();
 904         fix2             = _mm_setzero_ps();
 905         fiy2             = _mm_setzero_ps();
 906         fiz2             = _mm_setzero_ps();
 907
 908         /* Start inner kernel loop */
 909         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 910         {
 911
 912             /* Get j neighbor index, and coordinate index */
 913             jnrA             = jjnr[jidx];
 914             jnrB             = jjnr[jidx+1];
 915             jnrC             = jjnr[jidx+2];
 916             jnrD             = jjnr[jidx+3];
 917             j_coord_offsetA  = DIM*jnrA;
 918             j_coord_offsetB  = DIM*jnrB;
 919             j_coord_offsetC  = DIM*jnrC;
 920             j_coord_offsetD  = DIM*jnrD;
 921
 922             /* load j atom coordinates */
 923             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 924                                               x+j_coord_offsetC,x+j_coord_offsetD,
 925                                               &jx0,&jy0,&jz0);
 926
 927             /* Calculate displacement vector */
 928             dx00             = _mm_sub_ps(ix0,jx0);
 929             dy00             = _mm_sub_ps(iy0,jy0);
 930             dz00             = _mm_sub_ps(iz0,jz0);
 931             dx10             = _mm_sub_ps(ix1,jx0);
 932             dy10             = _mm_sub_ps(iy1,jy0);
 933             dz10             = _mm_sub_ps(iz1,jz0);
 934             dx20             = _mm_sub_ps(ix2,jx0);
 935             dy20             = _mm_sub_ps(iy2,jy0);
 936             dz20             = _mm_sub_ps(iz2,jz0);
 937
 938             /* Calculate squared distance and things based on it */
 939             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 940             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 941             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 942
 943             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 944             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 945             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 946
 947             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 948             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 949             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 950
 951             /* Load parameters for j particles */
 952             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 953                                                               charge+jnrC+0,charge+jnrD+0);
 954             vdwjidx0A        = 2*vdwtype[jnrA+0];
 955             vdwjidx0B        = 2*vdwtype[jnrB+0];
 956             vdwjidx0C        = 2*vdwtype[jnrC+0];
 957             vdwjidx0D        = 2*vdwtype[jnrD+0];
 958
 959             fjx0             = _mm_setzero_ps();
 960             fjy0             = _mm_setzero_ps();
 961             fjz0             = _mm_setzero_ps();
 962
 963             /**************************
 964              * CALCULATE INTERACTIONS *
 965              **************************/
 966
 967             if (gmx_mm_any_lt(rsq00,rcutoff2))
 968             {
 969
 970             r00              = _mm_mul_ps(rsq00,rinv00);
 971
 972             /* Compute parameters for interactions between i and j atoms */
 973             qq00             = _mm_mul_ps(iq0,jq0);
 974             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 975                                          vdwparam+vdwioffset0+vdwjidx0B,
 976                                          vdwparam+vdwioffset0+vdwjidx0C,
 977                                          vdwparam+vdwioffset0+vdwjidx0D,
 978                                          &c6_00,&c12_00);
 979
 980             /* EWALD ELECTROSTATICS */
 981
 982             /* Analytical PME correction */
 983             zeta2            = _mm_mul_ps(beta2,rsq00);
 984             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 985             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 986             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 987             felec            = _mm_mul_ps(qq00,felec);
 988             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 989             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 990             velec            = _mm_mul_ps(qq00,velec);
 991
 992             /* LENNARD-JONES DISPERSION/REPULSION */
 993
 994             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 995             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 996             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 997             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 998             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 999
1000             d                = _mm_sub_ps(r00,rswitch);
1001             d                = _mm_max_ps(d,_mm_setzero_ps());
1002             d2               = _mm_mul_ps(d,d);
1003             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1004
1005             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1006
1007             /* Evaluate switch function */
1008             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1009             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
1010             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1011             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1012
1013             fscal            = _mm_add_ps(felec,fvdw);
1014
1015             fscal            = _mm_and_ps(fscal,cutoff_mask);
1016
1017              /* Update vectorial force */
1018             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1019             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1020             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1021
1022             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1023             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1024             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1025
1026             }
1027
1028             /**************************
1029              * CALCULATE INTERACTIONS *
1030              **************************/
1031
1032             if (gmx_mm_any_lt(rsq10,rcutoff2))
1033             {
1034
1035             r10              = _mm_mul_ps(rsq10,rinv10);
1036
1037             /* Compute parameters for interactions between i and j atoms */
1038             qq10             = _mm_mul_ps(iq1,jq0);
1039
1040             /* EWALD ELECTROSTATICS */
1041
1042             /* Analytical PME correction */
1043             zeta2            = _mm_mul_ps(beta2,rsq10);
1044             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1045             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1046             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1047             felec            = _mm_mul_ps(qq10,felec);
1048             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1049             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
1050             velec            = _mm_mul_ps(qq10,velec);
1051
1052             d                = _mm_sub_ps(r10,rswitch);
1053             d                = _mm_max_ps(d,_mm_setzero_ps());
1054             d2               = _mm_mul_ps(d,d);
1055             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1056
1057             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1058
1059             /* Evaluate switch function */
1060             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1061             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
1062             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1063
1064             fscal            = felec;
1065
1066             fscal            = _mm_and_ps(fscal,cutoff_mask);
1067
1068              /* Update vectorial force */
1069             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1070             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1071             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1072
1073             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1074             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1075             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1076
1077             }
1078
1079             /**************************
1080              * CALCULATE INTERACTIONS *
1081              **************************/
1082
1083             if (gmx_mm_any_lt(rsq20,rcutoff2))
1084             {
1085
1086             r20              = _mm_mul_ps(rsq20,rinv20);
1087
1088             /* Compute parameters for interactions between i and j atoms */
1089             qq20             = _mm_mul_ps(iq2,jq0);
1090
1091             /* EWALD ELECTROSTATICS */
1092
1093             /* Analytical PME correction */
1094             zeta2            = _mm_mul_ps(beta2,rsq20);
1095             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1096             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1097             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1098             felec            = _mm_mul_ps(qq20,felec);
1099             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1100             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
1101             velec            = _mm_mul_ps(qq20,velec);
1102
1103             d                = _mm_sub_ps(r20,rswitch);
1104             d                = _mm_max_ps(d,_mm_setzero_ps());
1105             d2               = _mm_mul_ps(d,d);
1106             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1107
1108             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1109
1110             /* Evaluate switch function */
1111             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1112             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1113             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1114
1115             fscal            = felec;
1116
1117             fscal            = _mm_and_ps(fscal,cutoff_mask);
1118
1119              /* Update vectorial force */
1120             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1121             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1122             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1123
1124             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1125             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1126             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1127
1128             }
1129
1130             fjptrA             = f+j_coord_offsetA;
1131             fjptrB             = f+j_coord_offsetB;
1132             fjptrC             = f+j_coord_offsetC;
1133             fjptrD             = f+j_coord_offsetD;
1134
1135             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1136
1137             /* Inner loop uses 165 flops */
1138         }
1139
1140         if(jidx<j_index_end)
1141         {
1142
1143             /* Get j neighbor index, and coordinate index */
1144             jnrlistA         = jjnr[jidx];
1145             jnrlistB         = jjnr[jidx+1];
1146             jnrlistC         = jjnr[jidx+2];
1147             jnrlistD         = jjnr[jidx+3];
1148             /* Sign of each element will be negative for non-real atoms.
1149              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1150              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1151              */
1152             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1153             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1154             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1155             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1156             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1157             j_coord_offsetA  = DIM*jnrA;
1158             j_coord_offsetB  = DIM*jnrB;
1159             j_coord_offsetC  = DIM*jnrC;
1160             j_coord_offsetD  = DIM*jnrD;
1161
1162             /* load j atom coordinates */
1163             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1164                                               x+j_coord_offsetC,x+j_coord_offsetD,
1165                                               &jx0,&jy0,&jz0);
1166
1167             /* Calculate displacement vector */
1168             dx00             = _mm_sub_ps(ix0,jx0);
1169             dy00             = _mm_sub_ps(iy0,jy0);
1170             dz00             = _mm_sub_ps(iz0,jz0);
1171             dx10             = _mm_sub_ps(ix1,jx0);
1172             dy10             = _mm_sub_ps(iy1,jy0);
1173             dz10             = _mm_sub_ps(iz1,jz0);
1174             dx20             = _mm_sub_ps(ix2,jx0);
1175             dy20             = _mm_sub_ps(iy2,jy0);
1176             dz20             = _mm_sub_ps(iz2,jz0);
1177
1178             /* Calculate squared distance and things based on it */
1179             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1180             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1181             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1182
1183             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1184             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1185             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1186
1187             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1188             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1189             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1190
1191             /* Load parameters for j particles */
1192             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1193                                                               charge+jnrC+0,charge+jnrD+0);
1194             vdwjidx0A        = 2*vdwtype[jnrA+0];
1195             vdwjidx0B        = 2*vdwtype[jnrB+0];
1196             vdwjidx0C        = 2*vdwtype[jnrC+0];
1197             vdwjidx0D        = 2*vdwtype[jnrD+0];
1198
1199             fjx0             = _mm_setzero_ps();
1200             fjy0             = _mm_setzero_ps();
1201             fjz0             = _mm_setzero_ps();
1202
1203             /**************************
1204              * CALCULATE INTERACTIONS *
1205              **************************/
1206
1207             if (gmx_mm_any_lt(rsq00,rcutoff2))
1208             {
1209
1210             r00              = _mm_mul_ps(rsq00,rinv00);
1211             r00              = _mm_andnot_ps(dummy_mask,r00);
1212
1213             /* Compute parameters for interactions between i and j atoms */
1214             qq00             = _mm_mul_ps(iq0,jq0);
1215             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1216                                          vdwparam+vdwioffset0+vdwjidx0B,
1217                                          vdwparam+vdwioffset0+vdwjidx0C,
1218                                          vdwparam+vdwioffset0+vdwjidx0D,
1219                                          &c6_00,&c12_00);
1220
1221             /* EWALD ELECTROSTATICS */
1222
1223             /* Analytical PME correction */
1224             zeta2            = _mm_mul_ps(beta2,rsq00);
1225             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
1226             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1227             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1228             felec            = _mm_mul_ps(qq00,felec);
1229             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1230             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
1231             velec            = _mm_mul_ps(qq00,velec);
1232
1233             /* LENNARD-JONES DISPERSION/REPULSION */
1234
1235             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1236             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1237             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1238             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
1239             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1240
1241             d                = _mm_sub_ps(r00,rswitch);
1242             d                = _mm_max_ps(d,_mm_setzero_ps());
1243             d2               = _mm_mul_ps(d,d);
1244             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1245
1246             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1247
1248             /* Evaluate switch function */
1249             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1250             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
1251             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1252             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1253
1254             fscal            = _mm_add_ps(felec,fvdw);
1255
1256             fscal            = _mm_and_ps(fscal,cutoff_mask);
1257
1258             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1259
1260              /* Update vectorial force */
1261             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1262             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1263             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1264
1265             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1266             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1267             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1268
1269             }
1270
1271             /**************************
1272              * CALCULATE INTERACTIONS *
1273              **************************/
1274
1275             if (gmx_mm_any_lt(rsq10,rcutoff2))
1276             {
1277
1278             r10              = _mm_mul_ps(rsq10,rinv10);
1279             r10              = _mm_andnot_ps(dummy_mask,r10);
1280
1281             /* Compute parameters for interactions between i and j atoms */
1282             qq10             = _mm_mul_ps(iq1,jq0);
1283
1284             /* EWALD ELECTROSTATICS */
1285
1286             /* Analytical PME correction */
1287             zeta2            = _mm_mul_ps(beta2,rsq10);
1288             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1289             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1290             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1291             felec            = _mm_mul_ps(qq10,felec);
1292             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1293             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
1294             velec            = _mm_mul_ps(qq10,velec);
1295
1296             d                = _mm_sub_ps(r10,rswitch);
1297             d                = _mm_max_ps(d,_mm_setzero_ps());
1298             d2               = _mm_mul_ps(d,d);
1299             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1300
1301             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1302
1303             /* Evaluate switch function */
1304             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1305             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
1306             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1307
1308             fscal            = felec;
1309
1310             fscal            = _mm_and_ps(fscal,cutoff_mask);
1311
1312             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1313
1314              /* Update vectorial force */
1315             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1316             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1317             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1318
1319             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1320             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1321             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1322
1323             }
1324
1325             /**************************
1326              * CALCULATE INTERACTIONS *
1327              **************************/
1328
1329             if (gmx_mm_any_lt(rsq20,rcutoff2))
1330             {
1331
1332             r20              = _mm_mul_ps(rsq20,rinv20);
1333             r20              = _mm_andnot_ps(dummy_mask,r20);
1334
1335             /* Compute parameters for interactions between i and j atoms */
1336             qq20             = _mm_mul_ps(iq2,jq0);
1337
1338             /* EWALD ELECTROSTATICS */
1339
1340             /* Analytical PME correction */
1341             zeta2            = _mm_mul_ps(beta2,rsq20);
1342             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1343             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1344             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1345             felec            = _mm_mul_ps(qq20,felec);
1346             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1347             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
1348             velec            = _mm_mul_ps(qq20,velec);
1349
1350             d                = _mm_sub_ps(r20,rswitch);
1351             d                = _mm_max_ps(d,_mm_setzero_ps());
1352             d2               = _mm_mul_ps(d,d);
1353             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1354
1355             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1356
1357             /* Evaluate switch function */
1358             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1359             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1360             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1361
1362             fscal            = felec;
1363
1364             fscal            = _mm_and_ps(fscal,cutoff_mask);
1365
1366             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1367
1368              /* Update vectorial force */
1369             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1370             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1371             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1372
1373             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1374             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1375             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1376
1377             }
1378
1379             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1380             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1381             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1382             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1383
1384             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1385
1386             /* Inner loop uses 168 flops */
1387         }
1388
1389         /* End of innermost loop */
1390
1391         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1392                                               f+i_coord_offset,fshift+i_shift_offset);
1393
1394         /* Increment number of inner iterations */
1395         inneriter                  += j_index_end - j_index_start;
1396
1397         /* Outer loop uses 18 flops */
1398     }
1399
1400     /* Increment number of outer iterations */
1401     outeriter        += nri;
1402
1403     /* Update outer/inner flops */
1404
1405     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*168);
1406 }