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) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_VF_avx_128_fma_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 101     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 102     __m128i          ewitab;
 103     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 104     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 105     real             *ewtab;
 106     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 107     real             rswitch_scalar,d_scalar;
 108     __m128           dummy_mask,cutoff_mask;
 109     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 110     __m128           one     = _mm_set1_ps(1.0);
 111     __m128           two     = _mm_set1_ps(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm_set1_ps(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 130     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 131     beta2            = _mm_mul_ps(beta,beta);
 132     beta3            = _mm_mul_ps(beta,beta2);
 133     ewtab            = fr->ic->tabq_coul_FDV0;
 134     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 135     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 136
 137     /* Setup water-specific parameters */
 138     inr              = nlist->iinr[0];
 139     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 140     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 141     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 142     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 143
 144     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 145     rcutoff_scalar   = fr->rcoulomb;
 146     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 147     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 148
 149     rswitch_scalar   = fr->rcoulomb_switch;
 150     rswitch          = _mm_set1_ps(rswitch_scalar);
 151     /* Setup switch parameters */
 152     d_scalar         = rcutoff_scalar-rswitch_scalar;
 153     d                = _mm_set1_ps(d_scalar);
 154     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 155     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 156     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 157     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 158     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 159     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 160
 161     /* Avoid stupid compiler warnings */
 162     jnrA = jnrB = jnrC = jnrD = 0;
 163     j_coord_offsetA = 0;
 164     j_coord_offsetB = 0;
 165     j_coord_offsetC = 0;
 166     j_coord_offsetD = 0;
 167
 168     outeriter        = 0;
 169     inneriter        = 0;
 170
 171     for(iidx=0;iidx<4*DIM;iidx++)
 172     {
 173         scratch[iidx] = 0.0;
 174     }
 175
 176     /* Start outer loop over neighborlists */
 177     for(iidx=0; iidx<nri; iidx++)
 178     {
 179         /* Load shift vector for this list */
 180         i_shift_offset   = DIM*shiftidx[iidx];
 181
 182         /* Load limits for loop over neighbors */
 183         j_index_start    = jindex[iidx];
 184         j_index_end      = jindex[iidx+1];
 185
 186         /* Get outer coordinate index */
 187         inr              = iinr[iidx];
 188         i_coord_offset   = DIM*inr;
 189
 190         /* Load i particle coords and add shift vector */
 191         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 192                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 193
 194         fix0             = _mm_setzero_ps();
 195         fiy0             = _mm_setzero_ps();
 196         fiz0             = _mm_setzero_ps();
 197         fix1             = _mm_setzero_ps();
 198         fiy1             = _mm_setzero_ps();
 199         fiz1             = _mm_setzero_ps();
 200         fix2             = _mm_setzero_ps();
 201         fiy2             = _mm_setzero_ps();
 202         fiz2             = _mm_setzero_ps();
 203
 204         /* Reset potential sums */
 205         velecsum         = _mm_setzero_ps();
 206         vvdwsum          = _mm_setzero_ps();
 207
 208         /* Start inner kernel loop */
 209         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 210         {
 211
 212             /* Get j neighbor index, and coordinate index */
 213             jnrA             = jjnr[jidx];
 214             jnrB             = jjnr[jidx+1];
 215             jnrC             = jjnr[jidx+2];
 216             jnrD             = jjnr[jidx+3];
 217             j_coord_offsetA  = DIM*jnrA;
 218             j_coord_offsetB  = DIM*jnrB;
 219             j_coord_offsetC  = DIM*jnrC;
 220             j_coord_offsetD  = DIM*jnrD;
 221
 222             /* load j atom coordinates */
 223             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 224                                               x+j_coord_offsetC,x+j_coord_offsetD,
 225                                               &jx0,&jy0,&jz0);
 226
 227             /* Calculate displacement vector */
 228             dx00             = _mm_sub_ps(ix0,jx0);
 229             dy00             = _mm_sub_ps(iy0,jy0);
 230             dz00             = _mm_sub_ps(iz0,jz0);
 231             dx10             = _mm_sub_ps(ix1,jx0);
 232             dy10             = _mm_sub_ps(iy1,jy0);
 233             dz10             = _mm_sub_ps(iz1,jz0);
 234             dx20             = _mm_sub_ps(ix2,jx0);
 235             dy20             = _mm_sub_ps(iy2,jy0);
 236             dz20             = _mm_sub_ps(iz2,jz0);
 237
 238             /* Calculate squared distance and things based on it */
 239             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 240             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 241             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 242
 243             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 244             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 245             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 246
 247             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 248             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 249             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 250
 251             /* Load parameters for j particles */
 252             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 253                                                               charge+jnrC+0,charge+jnrD+0);
 254             vdwjidx0A        = 2*vdwtype[jnrA+0];
 255             vdwjidx0B        = 2*vdwtype[jnrB+0];
 256             vdwjidx0C        = 2*vdwtype[jnrC+0];
 257             vdwjidx0D        = 2*vdwtype[jnrD+0];
 258
 259             fjx0             = _mm_setzero_ps();
 260             fjy0             = _mm_setzero_ps();
 261             fjz0             = _mm_setzero_ps();
 262
 263             /**************************
 264              * CALCULATE INTERACTIONS *
 265              **************************/
 266
 267             if (gmx_mm_any_lt(rsq00,rcutoff2))
 268             {
 269
 270             r00              = _mm_mul_ps(rsq00,rinv00);
 271
 272             /* Compute parameters for interactions between i and j atoms */
 273             qq00             = _mm_mul_ps(iq0,jq0);
 274             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 275                                          vdwparam+vdwioffset0+vdwjidx0B,
 276                                          vdwparam+vdwioffset0+vdwjidx0C,
 277                                          vdwparam+vdwioffset0+vdwjidx0D,
 278                                          &c6_00,&c12_00);
 279
 280             /* EWALD ELECTROSTATICS */
 281
 282             /* Analytical PME correction */
 283             zeta2            = _mm_mul_ps(beta2,rsq00);
 284             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 285             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 286             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 287             felec            = _mm_mul_ps(qq00,felec);
 288             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 289             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 290             velec            = _mm_mul_ps(qq00,velec);
 291
 292             /* LENNARD-JONES DISPERSION/REPULSION */
 293
 294             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 295             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 296             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 297             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 298             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 299
 300             d                = _mm_sub_ps(r00,rswitch);
 301             d                = _mm_max_ps(d,_mm_setzero_ps());
 302             d2               = _mm_mul_ps(d,d);
 303             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 304
 305             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 306
 307             /* Evaluate switch function */
 308             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 309             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 310             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 311             velec            = _mm_mul_ps(velec,sw);
 312             vvdw             = _mm_mul_ps(vvdw,sw);
 313             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 314
 315             /* Update potential sum for this i atom from the interaction with this j atom. */
 316             velec            = _mm_and_ps(velec,cutoff_mask);
 317             velecsum         = _mm_add_ps(velecsum,velec);
 318             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 319             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 320
 321             fscal            = _mm_add_ps(felec,fvdw);
 322
 323             fscal            = _mm_and_ps(fscal,cutoff_mask);
 324
 325              /* Update vectorial force */
 326             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 327             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 328             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 329
 330             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 331             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 332             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 333
 334             }
 335
 336             /**************************
 337              * CALCULATE INTERACTIONS *
 338              **************************/
 339
 340             if (gmx_mm_any_lt(rsq10,rcutoff2))
 341             {
 342
 343             r10              = _mm_mul_ps(rsq10,rinv10);
 344
 345             /* Compute parameters for interactions between i and j atoms */
 346             qq10             = _mm_mul_ps(iq1,jq0);
 347
 348             /* EWALD ELECTROSTATICS */
 349
 350             /* Analytical PME correction */
 351             zeta2            = _mm_mul_ps(beta2,rsq10);
 352             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 353             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 354             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 355             felec            = _mm_mul_ps(qq10,felec);
 356             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 357             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 358             velec            = _mm_mul_ps(qq10,velec);
 359
 360             d                = _mm_sub_ps(r10,rswitch);
 361             d                = _mm_max_ps(d,_mm_setzero_ps());
 362             d2               = _mm_mul_ps(d,d);
 363             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 364
 365             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 366
 367             /* Evaluate switch function */
 368             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 369             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 370             velec            = _mm_mul_ps(velec,sw);
 371             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 372
 373             /* Update potential sum for this i atom from the interaction with this j atom. */
 374             velec            = _mm_and_ps(velec,cutoff_mask);
 375             velecsum         = _mm_add_ps(velecsum,velec);
 376
 377             fscal            = felec;
 378
 379             fscal            = _mm_and_ps(fscal,cutoff_mask);
 380
 381              /* Update vectorial force */
 382             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 383             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 384             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 385
 386             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 387             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 388             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 389
 390             }
 391
 392             /**************************
 393              * CALCULATE INTERACTIONS *
 394              **************************/
 395
 396             if (gmx_mm_any_lt(rsq20,rcutoff2))
 397             {
 398
 399             r20              = _mm_mul_ps(rsq20,rinv20);
 400
 401             /* Compute parameters for interactions between i and j atoms */
 402             qq20             = _mm_mul_ps(iq2,jq0);
 403
 404             /* EWALD ELECTROSTATICS */
 405
 406             /* Analytical PME correction */
 407             zeta2            = _mm_mul_ps(beta2,rsq20);
 408             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 409             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 410             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 411             felec            = _mm_mul_ps(qq20,felec);
 412             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 413             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 414             velec            = _mm_mul_ps(qq20,velec);
 415
 416             d                = _mm_sub_ps(r20,rswitch);
 417             d                = _mm_max_ps(d,_mm_setzero_ps());
 418             d2               = _mm_mul_ps(d,d);
 419             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 420
 421             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 422
 423             /* Evaluate switch function */
 424             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 425             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
 426             velec            = _mm_mul_ps(velec,sw);
 427             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 428
 429             /* Update potential sum for this i atom from the interaction with this j atom. */
 430             velec            = _mm_and_ps(velec,cutoff_mask);
 431             velecsum         = _mm_add_ps(velecsum,velec);
 432
 433             fscal            = felec;
 434
 435             fscal            = _mm_and_ps(fscal,cutoff_mask);
 436
 437              /* Update vectorial force */
 438             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 439             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 440             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 441
 442             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 443             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 444             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 445
 446             }
 447
 448             fjptrA             = f+j_coord_offsetA;
 449             fjptrB             = f+j_coord_offsetB;
 450             fjptrC             = f+j_coord_offsetC;
 451             fjptrD             = f+j_coord_offsetD;
 452
 453             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 454
 455             /* Inner loop uses 177 flops */
 456         }
 457
 458         if(jidx<j_index_end)
 459         {
 460
 461             /* Get j neighbor index, and coordinate index */
 462             jnrlistA         = jjnr[jidx];
 463             jnrlistB         = jjnr[jidx+1];
 464             jnrlistC         = jjnr[jidx+2];
 465             jnrlistD         = jjnr[jidx+3];
 466             /* Sign of each element will be negative for non-real atoms.
 467              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 468              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 469              */
 470             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 471             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 472             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 473             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 474             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 475             j_coord_offsetA  = DIM*jnrA;
 476             j_coord_offsetB  = DIM*jnrB;
 477             j_coord_offsetC  = DIM*jnrC;
 478             j_coord_offsetD  = DIM*jnrD;
 479
 480             /* load j atom coordinates */
 481             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 482                                               x+j_coord_offsetC,x+j_coord_offsetD,
 483                                               &jx0,&jy0,&jz0);
 484
 485             /* Calculate displacement vector */
 486             dx00             = _mm_sub_ps(ix0,jx0);
 487             dy00             = _mm_sub_ps(iy0,jy0);
 488             dz00             = _mm_sub_ps(iz0,jz0);
 489             dx10             = _mm_sub_ps(ix1,jx0);
 490             dy10             = _mm_sub_ps(iy1,jy0);
 491             dz10             = _mm_sub_ps(iz1,jz0);
 492             dx20             = _mm_sub_ps(ix2,jx0);
 493             dy20             = _mm_sub_ps(iy2,jy0);
 494             dz20             = _mm_sub_ps(iz2,jz0);
 495
 496             /* Calculate squared distance and things based on it */
 497             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 498             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 499             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 500
 501             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 502             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 503             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 504
 505             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 506             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 507             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 508
 509             /* Load parameters for j particles */
 510             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 511                                                               charge+jnrC+0,charge+jnrD+0);
 512             vdwjidx0A        = 2*vdwtype[jnrA+0];
 513             vdwjidx0B        = 2*vdwtype[jnrB+0];
 514             vdwjidx0C        = 2*vdwtype[jnrC+0];
 515             vdwjidx0D        = 2*vdwtype[jnrD+0];
 516
 517             fjx0             = _mm_setzero_ps();
 518             fjy0             = _mm_setzero_ps();
 519             fjz0             = _mm_setzero_ps();
 520
 521             /**************************
 522              * CALCULATE INTERACTIONS *
 523              **************************/
 524
 525             if (gmx_mm_any_lt(rsq00,rcutoff2))
 526             {
 527
 528             r00              = _mm_mul_ps(rsq00,rinv00);
 529             r00              = _mm_andnot_ps(dummy_mask,r00);
 530
 531             /* Compute parameters for interactions between i and j atoms */
 532             qq00             = _mm_mul_ps(iq0,jq0);
 533             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 534                                          vdwparam+vdwioffset0+vdwjidx0B,
 535                                          vdwparam+vdwioffset0+vdwjidx0C,
 536                                          vdwparam+vdwioffset0+vdwjidx0D,
 537                                          &c6_00,&c12_00);
 538
 539             /* EWALD ELECTROSTATICS */
 540
 541             /* Analytical PME correction */
 542             zeta2            = _mm_mul_ps(beta2,rsq00);
 543             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 544             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 545             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 546             felec            = _mm_mul_ps(qq00,felec);
 547             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 548             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 549             velec            = _mm_mul_ps(qq00,velec);
 550
 551             /* LENNARD-JONES DISPERSION/REPULSION */
 552
 553             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 554             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 555             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 556             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 557             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 558
 559             d                = _mm_sub_ps(r00,rswitch);
 560             d                = _mm_max_ps(d,_mm_setzero_ps());
 561             d2               = _mm_mul_ps(d,d);
 562             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 563
 564             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 565
 566             /* Evaluate switch function */
 567             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 568             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 569             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 570             velec            = _mm_mul_ps(velec,sw);
 571             vvdw             = _mm_mul_ps(vvdw,sw);
 572             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 573
 574             /* Update potential sum for this i atom from the interaction with this j atom. */
 575             velec            = _mm_and_ps(velec,cutoff_mask);
 576             velec            = _mm_andnot_ps(dummy_mask,velec);
 577             velecsum         = _mm_add_ps(velecsum,velec);
 578             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 579             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 580             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 581
 582             fscal            = _mm_add_ps(felec,fvdw);
 583
 584             fscal            = _mm_and_ps(fscal,cutoff_mask);
 585
 586             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 587
 588              /* Update vectorial force */
 589             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 590             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 591             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 592
 593             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 594             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 595             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 596
 597             }
 598
 599             /**************************
 600              * CALCULATE INTERACTIONS *
 601              **************************/
 602
 603             if (gmx_mm_any_lt(rsq10,rcutoff2))
 604             {
 605
 606             r10              = _mm_mul_ps(rsq10,rinv10);
 607             r10              = _mm_andnot_ps(dummy_mask,r10);
 608
 609             /* Compute parameters for interactions between i and j atoms */
 610             qq10             = _mm_mul_ps(iq1,jq0);
 611
 612             /* EWALD ELECTROSTATICS */
 613
 614             /* Analytical PME correction */
 615             zeta2            = _mm_mul_ps(beta2,rsq10);
 616             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 617             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 618             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 619             felec            = _mm_mul_ps(qq10,felec);
 620             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 621             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 622             velec            = _mm_mul_ps(qq10,velec);
 623
 624             d                = _mm_sub_ps(r10,rswitch);
 625             d                = _mm_max_ps(d,_mm_setzero_ps());
 626             d2               = _mm_mul_ps(d,d);
 627             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 628
 629             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 630
 631             /* Evaluate switch function */
 632             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 633             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 634             velec            = _mm_mul_ps(velec,sw);
 635             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 636
 637             /* Update potential sum for this i atom from the interaction with this j atom. */
 638             velec            = _mm_and_ps(velec,cutoff_mask);
 639             velec            = _mm_andnot_ps(dummy_mask,velec);
 640             velecsum         = _mm_add_ps(velecsum,velec);
 641
 642             fscal            = felec;
 643
 644             fscal            = _mm_and_ps(fscal,cutoff_mask);
 645
 646             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 647
 648              /* Update vectorial force */
 649             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 650             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 651             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 652
 653             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 654             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 655             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 656
 657             }
 658
 659             /**************************
 660              * CALCULATE INTERACTIONS *
 661              **************************/
 662
 663             if (gmx_mm_any_lt(rsq20,rcutoff2))
 664             {
 665
 666             r20              = _mm_mul_ps(rsq20,rinv20);
 667             r20              = _mm_andnot_ps(dummy_mask,r20);
 668
 669             /* Compute parameters for interactions between i and j atoms */
 670             qq20             = _mm_mul_ps(iq2,jq0);
 671
 672             /* EWALD ELECTROSTATICS */
 673
 674             /* Analytical PME correction */
 675             zeta2            = _mm_mul_ps(beta2,rsq20);
 676             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 677             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 678             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 679             felec            = _mm_mul_ps(qq20,felec);
 680             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 681             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 682             velec            = _mm_mul_ps(qq20,velec);
 683
 684             d                = _mm_sub_ps(r20,rswitch);
 685             d                = _mm_max_ps(d,_mm_setzero_ps());
 686             d2               = _mm_mul_ps(d,d);
 687             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 688
 689             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 690
 691             /* Evaluate switch function */
 692             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 693             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
 694             velec            = _mm_mul_ps(velec,sw);
 695             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 696
 697             /* Update potential sum for this i atom from the interaction with this j atom. */
 698             velec            = _mm_and_ps(velec,cutoff_mask);
 699             velec            = _mm_andnot_ps(dummy_mask,velec);
 700             velecsum         = _mm_add_ps(velecsum,velec);
 701
 702             fscal            = felec;
 703
 704             fscal            = _mm_and_ps(fscal,cutoff_mask);
 705
 706             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 707
 708              /* Update vectorial force */
 709             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 710             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 711             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 712
 713             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 714             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 715             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 716
 717             }
 718
 719             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 720             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 721             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 722             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 723
 724             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 725
 726             /* Inner loop uses 180 flops */
 727         }
 728
 729         /* End of innermost loop */
 730
 731         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 732                                               f+i_coord_offset,fshift+i_shift_offset);
 733
 734         ggid                        = gid[iidx];
 735         /* Update potential energies */
 736         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 737         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 738
 739         /* Increment number of inner iterations */
 740         inneriter                  += j_index_end - j_index_start;
 741
 742         /* Outer loop uses 20 flops */
 743     }
 744
 745     /* Increment number of outer iterations */
 746     outeriter        += nri;
 747
 748     /* Update outer/inner flops */
 749
 750     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*180);
 751 }
 752 /*
 753  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_avx_128_fma_single
 754  * Electrostatics interaction: Ewald
 755  * VdW interaction:            LennardJones
 756  * Geometry:                   Water3-Particle
 757  * Calculate force/pot:        Force
 758  */
 759 void
 760 nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_avx_128_fma_single
 761                     (t_nblist                    * gmx_restrict       nlist,
 762                      rvec                        * gmx_restrict          xx,
 763                      rvec                        * gmx_restrict          ff,
 764                      t_forcerec                  * gmx_restrict          fr,
 765                      t_mdatoms                   * gmx_restrict     mdatoms,
 766                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 767                      t_nrnb                      * gmx_restrict        nrnb)
 768 {
 769     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 770      * just 0 for non-waters.
 771      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 772      * jnr indices corresponding to data put in the four positions in the SIMD register.
 773      */
 774     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 775     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 776     int              jnrA,jnrB,jnrC,jnrD;
 777     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 778     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 779     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 780     real             rcutoff_scalar;
 781     real             *shiftvec,*fshift,*x,*f;
 782     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 783     real             scratch[4*DIM];
 784     __m128           fscal,rcutoff,rcutoff2,jidxall;
 785     int              vdwioffset0;
 786     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 787     int              vdwioffset1;
 788     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 789     int              vdwioffset2;
 790     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 791     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 792     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 793     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 794     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 795     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 796     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 797     real             *charge;
 798     int              nvdwtype;
 799     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 800     int              *vdwtype;
 801     real             *vdwparam;
 802     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 803     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 804     __m128i          ewitab;
 805     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 806     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 807     real             *ewtab;
 808     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 809     real             rswitch_scalar,d_scalar;
 810     __m128           dummy_mask,cutoff_mask;
 811     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 812     __m128           one     = _mm_set1_ps(1.0);
 813     __m128           two     = _mm_set1_ps(2.0);
 814     x                = xx[0];
 815     f                = ff[0];
 816
 817     nri              = nlist->nri;
 818     iinr             = nlist->iinr;
 819     jindex           = nlist->jindex;
 820     jjnr             = nlist->jjnr;
 821     shiftidx         = nlist->shift;
 822     gid              = nlist->gid;
 823     shiftvec         = fr->shift_vec[0];
 824     fshift           = fr->fshift[0];
 825     facel            = _mm_set1_ps(fr->epsfac);
 826     charge           = mdatoms->chargeA;
 827     nvdwtype         = fr->ntype;
 828     vdwparam         = fr->nbfp;
 829     vdwtype          = mdatoms->typeA;
 830
 831     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 832     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 833     beta2            = _mm_mul_ps(beta,beta);
 834     beta3            = _mm_mul_ps(beta,beta2);
 835     ewtab            = fr->ic->tabq_coul_FDV0;
 836     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 837     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 838
 839     /* Setup water-specific parameters */
 840     inr              = nlist->iinr[0];
 841     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 842     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 843     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 844     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 845
 846     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 847     rcutoff_scalar   = fr->rcoulomb;
 848     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 849     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 850
 851     rswitch_scalar   = fr->rcoulomb_switch;
 852     rswitch          = _mm_set1_ps(rswitch_scalar);
 853     /* Setup switch parameters */
 854     d_scalar         = rcutoff_scalar-rswitch_scalar;
 855     d                = _mm_set1_ps(d_scalar);
 856     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 857     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 858     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 859     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 860     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 861     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 862
 863     /* Avoid stupid compiler warnings */
 864     jnrA = jnrB = jnrC = jnrD = 0;
 865     j_coord_offsetA = 0;
 866     j_coord_offsetB = 0;
 867     j_coord_offsetC = 0;
 868     j_coord_offsetD = 0;
 869
 870     outeriter        = 0;
 871     inneriter        = 0;
 872
 873     for(iidx=0;iidx<4*DIM;iidx++)
 874     {
 875         scratch[iidx] = 0.0;
 876     }
 877
 878     /* Start outer loop over neighborlists */
 879     for(iidx=0; iidx<nri; iidx++)
 880     {
 881         /* Load shift vector for this list */
 882         i_shift_offset   = DIM*shiftidx[iidx];
 883
 884         /* Load limits for loop over neighbors */
 885         j_index_start    = jindex[iidx];
 886         j_index_end      = jindex[iidx+1];
 887
 888         /* Get outer coordinate index */
 889         inr              = iinr[iidx];
 890         i_coord_offset   = DIM*inr;
 891
 892         /* Load i particle coords and add shift vector */
 893         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 894                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 895
 896         fix0             = _mm_setzero_ps();
 897         fiy0             = _mm_setzero_ps();
 898         fiz0             = _mm_setzero_ps();
 899         fix1             = _mm_setzero_ps();
 900         fiy1             = _mm_setzero_ps();
 901         fiz1             = _mm_setzero_ps();
 902         fix2             = _mm_setzero_ps();
 903         fiy2             = _mm_setzero_ps();
 904         fiz2             = _mm_setzero_ps();
 905
 906         /* Start inner kernel loop */
 907         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 908         {
 909
 910             /* Get j neighbor index, and coordinate index */
 911             jnrA             = jjnr[jidx];
 912             jnrB             = jjnr[jidx+1];
 913             jnrC             = jjnr[jidx+2];
 914             jnrD             = jjnr[jidx+3];
 915             j_coord_offsetA  = DIM*jnrA;
 916             j_coord_offsetB  = DIM*jnrB;
 917             j_coord_offsetC  = DIM*jnrC;
 918             j_coord_offsetD  = DIM*jnrD;
 919
 920             /* load j atom coordinates */
 921             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 922                                               x+j_coord_offsetC,x+j_coord_offsetD,
 923                                               &jx0,&jy0,&jz0);
 924
 925             /* Calculate displacement vector */
 926             dx00             = _mm_sub_ps(ix0,jx0);
 927             dy00             = _mm_sub_ps(iy0,jy0);
 928             dz00             = _mm_sub_ps(iz0,jz0);
 929             dx10             = _mm_sub_ps(ix1,jx0);
 930             dy10             = _mm_sub_ps(iy1,jy0);
 931             dz10             = _mm_sub_ps(iz1,jz0);
 932             dx20             = _mm_sub_ps(ix2,jx0);
 933             dy20             = _mm_sub_ps(iy2,jy0);
 934             dz20             = _mm_sub_ps(iz2,jz0);
 935
 936             /* Calculate squared distance and things based on it */
 937             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 938             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 939             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 940
 941             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 942             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 943             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 944
 945             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 946             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 947             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 948
 949             /* Load parameters for j particles */
 950             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 951                                                               charge+jnrC+0,charge+jnrD+0);
 952             vdwjidx0A        = 2*vdwtype[jnrA+0];
 953             vdwjidx0B        = 2*vdwtype[jnrB+0];
 954             vdwjidx0C        = 2*vdwtype[jnrC+0];
 955             vdwjidx0D        = 2*vdwtype[jnrD+0];
 956
 957             fjx0             = _mm_setzero_ps();
 958             fjy0             = _mm_setzero_ps();
 959             fjz0             = _mm_setzero_ps();
 960
 961             /**************************
 962              * CALCULATE INTERACTIONS *
 963              **************************/
 964
 965             if (gmx_mm_any_lt(rsq00,rcutoff2))
 966             {
 967
 968             r00              = _mm_mul_ps(rsq00,rinv00);
 969
 970             /* Compute parameters for interactions between i and j atoms */
 971             qq00             = _mm_mul_ps(iq0,jq0);
 972             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 973                                          vdwparam+vdwioffset0+vdwjidx0B,
 974                                          vdwparam+vdwioffset0+vdwjidx0C,
 975                                          vdwparam+vdwioffset0+vdwjidx0D,
 976                                          &c6_00,&c12_00);
 977
 978             /* EWALD ELECTROSTATICS */
 979
 980             /* Analytical PME correction */
 981             zeta2            = _mm_mul_ps(beta2,rsq00);
 982             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 983             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 984             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 985             felec            = _mm_mul_ps(qq00,felec);
 986             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 987             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 988             velec            = _mm_mul_ps(qq00,velec);
 989
 990             /* LENNARD-JONES DISPERSION/REPULSION */
 991
 992             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 993             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 994             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 995             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 996             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 997
 998             d                = _mm_sub_ps(r00,rswitch);
 999             d                = _mm_max_ps(d,_mm_setzero_ps());
1000             d2               = _mm_mul_ps(d,d);
1001             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1002
1003             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1004
1005             /* Evaluate switch function */
1006             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1007             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
1008             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1009             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1010
1011             fscal            = _mm_add_ps(felec,fvdw);
1012
1013             fscal            = _mm_and_ps(fscal,cutoff_mask);
1014
1015              /* Update vectorial force */
1016             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1017             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1018             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1019
1020             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1021             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1022             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1023
1024             }
1025
1026             /**************************
1027              * CALCULATE INTERACTIONS *
1028              **************************/
1029
1030             if (gmx_mm_any_lt(rsq10,rcutoff2))
1031             {
1032
1033             r10              = _mm_mul_ps(rsq10,rinv10);
1034
1035             /* Compute parameters for interactions between i and j atoms */
1036             qq10             = _mm_mul_ps(iq1,jq0);
1037
1038             /* EWALD ELECTROSTATICS */
1039
1040             /* Analytical PME correction */
1041             zeta2            = _mm_mul_ps(beta2,rsq10);
1042             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1043             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1044             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1045             felec            = _mm_mul_ps(qq10,felec);
1046             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1047             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
1048             velec            = _mm_mul_ps(qq10,velec);
1049
1050             d                = _mm_sub_ps(r10,rswitch);
1051             d                = _mm_max_ps(d,_mm_setzero_ps());
1052             d2               = _mm_mul_ps(d,d);
1053             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1054
1055             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1056
1057             /* Evaluate switch function */
1058             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1059             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
1060             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1061
1062             fscal            = felec;
1063
1064             fscal            = _mm_and_ps(fscal,cutoff_mask);
1065
1066              /* Update vectorial force */
1067             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1068             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1069             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1070
1071             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1072             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1073             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1074
1075             }
1076
1077             /**************************
1078              * CALCULATE INTERACTIONS *
1079              **************************/
1080
1081             if (gmx_mm_any_lt(rsq20,rcutoff2))
1082             {
1083
1084             r20              = _mm_mul_ps(rsq20,rinv20);
1085
1086             /* Compute parameters for interactions between i and j atoms */
1087             qq20             = _mm_mul_ps(iq2,jq0);
1088
1089             /* EWALD ELECTROSTATICS */
1090
1091             /* Analytical PME correction */
1092             zeta2            = _mm_mul_ps(beta2,rsq20);
1093             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1094             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1095             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1096             felec            = _mm_mul_ps(qq20,felec);
1097             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1098             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
1099             velec            = _mm_mul_ps(qq20,velec);
1100
1101             d                = _mm_sub_ps(r20,rswitch);
1102             d                = _mm_max_ps(d,_mm_setzero_ps());
1103             d2               = _mm_mul_ps(d,d);
1104             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1105
1106             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1107
1108             /* Evaluate switch function */
1109             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1110             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1111             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1112
1113             fscal            = felec;
1114
1115             fscal            = _mm_and_ps(fscal,cutoff_mask);
1116
1117              /* Update vectorial force */
1118             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1119             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1120             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1121
1122             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1123             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1124             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1125
1126             }
1127
1128             fjptrA             = f+j_coord_offsetA;
1129             fjptrB             = f+j_coord_offsetB;
1130             fjptrC             = f+j_coord_offsetC;
1131             fjptrD             = f+j_coord_offsetD;
1132
1133             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1134
1135             /* Inner loop uses 165 flops */
1136         }
1137
1138         if(jidx<j_index_end)
1139         {
1140
1141             /* Get j neighbor index, and coordinate index */
1142             jnrlistA         = jjnr[jidx];
1143             jnrlistB         = jjnr[jidx+1];
1144             jnrlistC         = jjnr[jidx+2];
1145             jnrlistD         = jjnr[jidx+3];
1146             /* Sign of each element will be negative for non-real atoms.
1147              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1148              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1149              */
1150             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1151             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1152             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1153             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1154             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1155             j_coord_offsetA  = DIM*jnrA;
1156             j_coord_offsetB  = DIM*jnrB;
1157             j_coord_offsetC  = DIM*jnrC;
1158             j_coord_offsetD  = DIM*jnrD;
1159
1160             /* load j atom coordinates */
1161             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1162                                               x+j_coord_offsetC,x+j_coord_offsetD,
1163                                               &jx0,&jy0,&jz0);
1164
1165             /* Calculate displacement vector */
1166             dx00             = _mm_sub_ps(ix0,jx0);
1167             dy00             = _mm_sub_ps(iy0,jy0);
1168             dz00             = _mm_sub_ps(iz0,jz0);
1169             dx10             = _mm_sub_ps(ix1,jx0);
1170             dy10             = _mm_sub_ps(iy1,jy0);
1171             dz10             = _mm_sub_ps(iz1,jz0);
1172             dx20             = _mm_sub_ps(ix2,jx0);
1173             dy20             = _mm_sub_ps(iy2,jy0);
1174             dz20             = _mm_sub_ps(iz2,jz0);
1175
1176             /* Calculate squared distance and things based on it */
1177             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1178             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1179             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1180
1181             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1182             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1183             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1184
1185             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1186             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1187             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1188
1189             /* Load parameters for j particles */
1190             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1191                                                               charge+jnrC+0,charge+jnrD+0);
1192             vdwjidx0A        = 2*vdwtype[jnrA+0];
1193             vdwjidx0B        = 2*vdwtype[jnrB+0];
1194             vdwjidx0C        = 2*vdwtype[jnrC+0];
1195             vdwjidx0D        = 2*vdwtype[jnrD+0];
1196
1197             fjx0             = _mm_setzero_ps();
1198             fjy0             = _mm_setzero_ps();
1199             fjz0             = _mm_setzero_ps();
1200
1201             /**************************
1202              * CALCULATE INTERACTIONS *
1203              **************************/
1204
1205             if (gmx_mm_any_lt(rsq00,rcutoff2))
1206             {
1207
1208             r00              = _mm_mul_ps(rsq00,rinv00);
1209             r00              = _mm_andnot_ps(dummy_mask,r00);
1210
1211             /* Compute parameters for interactions between i and j atoms */
1212             qq00             = _mm_mul_ps(iq0,jq0);
1213             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1214                                          vdwparam+vdwioffset0+vdwjidx0B,
1215                                          vdwparam+vdwioffset0+vdwjidx0C,
1216                                          vdwparam+vdwioffset0+vdwjidx0D,
1217                                          &c6_00,&c12_00);
1218
1219             /* EWALD ELECTROSTATICS */
1220
1221             /* Analytical PME correction */
1222             zeta2            = _mm_mul_ps(beta2,rsq00);
1223             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
1224             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1225             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1226             felec            = _mm_mul_ps(qq00,felec);
1227             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1228             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
1229             velec            = _mm_mul_ps(qq00,velec);
1230
1231             /* LENNARD-JONES DISPERSION/REPULSION */
1232
1233             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1234             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1235             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1236             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
1237             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1238
1239             d                = _mm_sub_ps(r00,rswitch);
1240             d                = _mm_max_ps(d,_mm_setzero_ps());
1241             d2               = _mm_mul_ps(d,d);
1242             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1243
1244             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1245
1246             /* Evaluate switch function */
1247             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1248             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
1249             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1250             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1251
1252             fscal            = _mm_add_ps(felec,fvdw);
1253
1254             fscal            = _mm_and_ps(fscal,cutoff_mask);
1255
1256             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1257
1258              /* Update vectorial force */
1259             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1260             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1261             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1262
1263             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1264             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1265             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1266
1267             }
1268
1269             /**************************
1270              * CALCULATE INTERACTIONS *
1271              **************************/
1272
1273             if (gmx_mm_any_lt(rsq10,rcutoff2))
1274             {
1275
1276             r10              = _mm_mul_ps(rsq10,rinv10);
1277             r10              = _mm_andnot_ps(dummy_mask,r10);
1278
1279             /* Compute parameters for interactions between i and j atoms */
1280             qq10             = _mm_mul_ps(iq1,jq0);
1281
1282             /* EWALD ELECTROSTATICS */
1283
1284             /* Analytical PME correction */
1285             zeta2            = _mm_mul_ps(beta2,rsq10);
1286             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1287             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1288             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1289             felec            = _mm_mul_ps(qq10,felec);
1290             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1291             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
1292             velec            = _mm_mul_ps(qq10,velec);
1293
1294             d                = _mm_sub_ps(r10,rswitch);
1295             d                = _mm_max_ps(d,_mm_setzero_ps());
1296             d2               = _mm_mul_ps(d,d);
1297             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1298
1299             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1300
1301             /* Evaluate switch function */
1302             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1303             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
1304             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1305
1306             fscal            = felec;
1307
1308             fscal            = _mm_and_ps(fscal,cutoff_mask);
1309
1310             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1311
1312              /* Update vectorial force */
1313             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1314             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1315             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1316
1317             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1318             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1319             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1320
1321             }
1322
1323             /**************************
1324              * CALCULATE INTERACTIONS *
1325              **************************/
1326
1327             if (gmx_mm_any_lt(rsq20,rcutoff2))
1328             {
1329
1330             r20              = _mm_mul_ps(rsq20,rinv20);
1331             r20              = _mm_andnot_ps(dummy_mask,r20);
1332
1333             /* Compute parameters for interactions between i and j atoms */
1334             qq20             = _mm_mul_ps(iq2,jq0);
1335
1336             /* EWALD ELECTROSTATICS */
1337
1338             /* Analytical PME correction */
1339             zeta2            = _mm_mul_ps(beta2,rsq20);
1340             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1341             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1342             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1343             felec            = _mm_mul_ps(qq20,felec);
1344             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1345             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
1346             velec            = _mm_mul_ps(qq20,velec);
1347
1348             d                = _mm_sub_ps(r20,rswitch);
1349             d                = _mm_max_ps(d,_mm_setzero_ps());
1350             d2               = _mm_mul_ps(d,d);
1351             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1352
1353             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1354
1355             /* Evaluate switch function */
1356             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1357             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1358             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1359
1360             fscal            = felec;
1361
1362             fscal            = _mm_and_ps(fscal,cutoff_mask);
1363
1364             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1365
1366              /* Update vectorial force */
1367             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1368             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1369             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1370
1371             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1372             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1373             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1374
1375             }
1376
1377             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1378             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1379             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1380             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1381
1382             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1383
1384             /* Inner loop uses 168 flops */
1385         }
1386
1387         /* End of innermost loop */
1388
1389         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1390                                               f+i_coord_offset,fshift+i_shift_offset);
1391
1392         /* Increment number of inner iterations */
1393         inneriter                  += j_index_end - j_index_start;
1394
1395         /* Outer loop uses 18 flops */
1396     }
1397
1398     /* Increment number of outer iterations */
1399     outeriter        += nri;
1400
1401     /* Update outer/inner flops */
1402
1403     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*168);
1404 }