src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_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 "gmx_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_GeomW4P1_VF_avx_128_fma_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_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              vdwioffset3;
  92     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  94     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  97     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  99     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 106     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 107     __m128i          ewitab;
 108     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 109     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 110     real             *ewtab;
 111     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 112     real             rswitch_scalar,d_scalar;
 113     __m128           dummy_mask,cutoff_mask;
 114     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 115     __m128           one     = _mm_set1_ps(1.0);
 116     __m128           two     = _mm_set1_ps(2.0);
 117     x                = xx[0];
 118     f                = ff[0];
 119
 120     nri              = nlist->nri;
 121     iinr             = nlist->iinr;
 122     jindex           = nlist->jindex;
 123     jjnr             = nlist->jjnr;
 124     shiftidx         = nlist->shift;
 125     gid              = nlist->gid;
 126     shiftvec         = fr->shift_vec[0];
 127     fshift           = fr->fshift[0];
 128     facel            = _mm_set1_ps(fr->epsfac);
 129     charge           = mdatoms->chargeA;
 130     nvdwtype         = fr->ntype;
 131     vdwparam         = fr->nbfp;
 132     vdwtype          = mdatoms->typeA;
 133
 134     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 135     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
 136     beta2            = _mm_mul_ps(beta,beta);
 137     beta3            = _mm_mul_ps(beta,beta2);
 138     ewtab            = fr->ic->tabq_coul_FDV0;
 139     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 140     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 141
 142     /* Setup water-specific parameters */
 143     inr              = nlist->iinr[0];
 144     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 145     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 146     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 147     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 148
 149     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 150     rcutoff_scalar   = fr->rcoulomb;
 151     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 152     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 153
 154     rswitch_scalar   = fr->rcoulomb_switch;
 155     rswitch          = _mm_set1_ps(rswitch_scalar);
 156     /* Setup switch parameters */
 157     d_scalar         = rcutoff_scalar-rswitch_scalar;
 158     d                = _mm_set1_ps(d_scalar);
 159     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 160     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 161     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 162     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 163     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 164     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 165
 166     /* Avoid stupid compiler warnings */
 167     jnrA = jnrB = jnrC = jnrD = 0;
 168     j_coord_offsetA = 0;
 169     j_coord_offsetB = 0;
 170     j_coord_offsetC = 0;
 171     j_coord_offsetD = 0;
 172
 173     outeriter        = 0;
 174     inneriter        = 0;
 175
 176     for(iidx=0;iidx<4*DIM;iidx++)
 177     {
 178         scratch[iidx] = 0.0;
 179     }
 180
 181     /* Start outer loop over neighborlists */
 182     for(iidx=0; iidx<nri; iidx++)
 183     {
 184         /* Load shift vector for this list */
 185         i_shift_offset   = DIM*shiftidx[iidx];
 186
 187         /* Load limits for loop over neighbors */
 188         j_index_start    = jindex[iidx];
 189         j_index_end      = jindex[iidx+1];
 190
 191         /* Get outer coordinate index */
 192         inr              = iinr[iidx];
 193         i_coord_offset   = DIM*inr;
 194
 195         /* Load i particle coords and add shift vector */
 196         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 197                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 198
 199         fix0             = _mm_setzero_ps();
 200         fiy0             = _mm_setzero_ps();
 201         fiz0             = _mm_setzero_ps();
 202         fix1             = _mm_setzero_ps();
 203         fiy1             = _mm_setzero_ps();
 204         fiz1             = _mm_setzero_ps();
 205         fix2             = _mm_setzero_ps();
 206         fiy2             = _mm_setzero_ps();
 207         fiz2             = _mm_setzero_ps();
 208         fix3             = _mm_setzero_ps();
 209         fiy3             = _mm_setzero_ps();
 210         fiz3             = _mm_setzero_ps();
 211
 212         /* Reset potential sums */
 213         velecsum         = _mm_setzero_ps();
 214         vvdwsum          = _mm_setzero_ps();
 215
 216         /* Start inner kernel loop */
 217         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 218         {
 219
 220             /* Get j neighbor index, and coordinate index */
 221             jnrA             = jjnr[jidx];
 222             jnrB             = jjnr[jidx+1];
 223             jnrC             = jjnr[jidx+2];
 224             jnrD             = jjnr[jidx+3];
 225             j_coord_offsetA  = DIM*jnrA;
 226             j_coord_offsetB  = DIM*jnrB;
 227             j_coord_offsetC  = DIM*jnrC;
 228             j_coord_offsetD  = DIM*jnrD;
 229
 230             /* load j atom coordinates */
 231             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 232                                               x+j_coord_offsetC,x+j_coord_offsetD,
 233                                               &jx0,&jy0,&jz0);
 234
 235             /* Calculate displacement vector */
 236             dx00             = _mm_sub_ps(ix0,jx0);
 237             dy00             = _mm_sub_ps(iy0,jy0);
 238             dz00             = _mm_sub_ps(iz0,jz0);
 239             dx10             = _mm_sub_ps(ix1,jx0);
 240             dy10             = _mm_sub_ps(iy1,jy0);
 241             dz10             = _mm_sub_ps(iz1,jz0);
 242             dx20             = _mm_sub_ps(ix2,jx0);
 243             dy20             = _mm_sub_ps(iy2,jy0);
 244             dz20             = _mm_sub_ps(iz2,jz0);
 245             dx30             = _mm_sub_ps(ix3,jx0);
 246             dy30             = _mm_sub_ps(iy3,jy0);
 247             dz30             = _mm_sub_ps(iz3,jz0);
 248
 249             /* Calculate squared distance and things based on it */
 250             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 251             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 252             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 253             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 254
 255             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 256             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 257             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 258             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 259
 260             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 261             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 262             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 263             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 264
 265             /* Load parameters for j particles */
 266             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 267                                                               charge+jnrC+0,charge+jnrD+0);
 268             vdwjidx0A        = 2*vdwtype[jnrA+0];
 269             vdwjidx0B        = 2*vdwtype[jnrB+0];
 270             vdwjidx0C        = 2*vdwtype[jnrC+0];
 271             vdwjidx0D        = 2*vdwtype[jnrD+0];
 272
 273             fjx0             = _mm_setzero_ps();
 274             fjy0             = _mm_setzero_ps();
 275             fjz0             = _mm_setzero_ps();
 276
 277             /**************************
 278              * CALCULATE INTERACTIONS *
 279              **************************/
 280
 281             if (gmx_mm_any_lt(rsq00,rcutoff2))
 282             {
 283
 284             r00              = _mm_mul_ps(rsq00,rinv00);
 285
 286             /* Compute parameters for interactions between i and j atoms */
 287             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 288                                          vdwparam+vdwioffset0+vdwjidx0B,
 289                                          vdwparam+vdwioffset0+vdwjidx0C,
 290                                          vdwparam+vdwioffset0+vdwjidx0D,
 291                                          &c6_00,&c12_00);
 292
 293             /* LENNARD-JONES DISPERSION/REPULSION */
 294
 295             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 296             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 297             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 298             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 299             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 300
 301             d                = _mm_sub_ps(r00,rswitch);
 302             d                = _mm_max_ps(d,_mm_setzero_ps());
 303             d2               = _mm_mul_ps(d,d);
 304             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 305
 306             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 307
 308             /* Evaluate switch function */
 309             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 310             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 311             vvdw             = _mm_mul_ps(vvdw,sw);
 312             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 313
 314             /* Update potential sum for this i atom from the interaction with this j atom. */
 315             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 316             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 317
 318             fscal            = fvdw;
 319
 320             fscal            = _mm_and_ps(fscal,cutoff_mask);
 321
 322              /* Update vectorial force */
 323             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 324             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 325             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 326
 327             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 328             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 329             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 330
 331             }
 332
 333             /**************************
 334              * CALCULATE INTERACTIONS *
 335              **************************/
 336
 337             if (gmx_mm_any_lt(rsq10,rcutoff2))
 338             {
 339
 340             r10              = _mm_mul_ps(rsq10,rinv10);
 341
 342             /* Compute parameters for interactions between i and j atoms */
 343             qq10             = _mm_mul_ps(iq1,jq0);
 344
 345             /* EWALD ELECTROSTATICS */
 346
 347             /* Analytical PME correction */
 348             zeta2            = _mm_mul_ps(beta2,rsq10);
 349             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 350             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 351             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 352             felec            = _mm_mul_ps(qq10,felec);
 353             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 354             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 355             velec            = _mm_mul_ps(qq10,velec);
 356
 357             d                = _mm_sub_ps(r10,rswitch);
 358             d                = _mm_max_ps(d,_mm_setzero_ps());
 359             d2               = _mm_mul_ps(d,d);
 360             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 361
 362             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 363
 364             /* Evaluate switch function */
 365             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 366             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 367             velec            = _mm_mul_ps(velec,sw);
 368             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 369
 370             /* Update potential sum for this i atom from the interaction with this j atom. */
 371             velec            = _mm_and_ps(velec,cutoff_mask);
 372             velecsum         = _mm_add_ps(velecsum,velec);
 373
 374             fscal            = felec;
 375
 376             fscal            = _mm_and_ps(fscal,cutoff_mask);
 377
 378              /* Update vectorial force */
 379             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 380             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 381             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 382
 383             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 384             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 385             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 386
 387             }
 388
 389             /**************************
 390              * CALCULATE INTERACTIONS *
 391              **************************/
 392
 393             if (gmx_mm_any_lt(rsq20,rcutoff2))
 394             {
 395
 396             r20              = _mm_mul_ps(rsq20,rinv20);
 397
 398             /* Compute parameters for interactions between i and j atoms */
 399             qq20             = _mm_mul_ps(iq2,jq0);
 400
 401             /* EWALD ELECTROSTATICS */
 402
 403             /* Analytical PME correction */
 404             zeta2            = _mm_mul_ps(beta2,rsq20);
 405             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 406             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 407             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 408             felec            = _mm_mul_ps(qq20,felec);
 409             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 410             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 411             velec            = _mm_mul_ps(qq20,velec);
 412
 413             d                = _mm_sub_ps(r20,rswitch);
 414             d                = _mm_max_ps(d,_mm_setzero_ps());
 415             d2               = _mm_mul_ps(d,d);
 416             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 417
 418             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 419
 420             /* Evaluate switch function */
 421             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 422             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
 423             velec            = _mm_mul_ps(velec,sw);
 424             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 425
 426             /* Update potential sum for this i atom from the interaction with this j atom. */
 427             velec            = _mm_and_ps(velec,cutoff_mask);
 428             velecsum         = _mm_add_ps(velecsum,velec);
 429
 430             fscal            = felec;
 431
 432             fscal            = _mm_and_ps(fscal,cutoff_mask);
 433
 434              /* Update vectorial force */
 435             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 436             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 437             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 438
 439             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 440             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 441             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 442
 443             }
 444
 445             /**************************
 446              * CALCULATE INTERACTIONS *
 447              **************************/
 448
 449             if (gmx_mm_any_lt(rsq30,rcutoff2))
 450             {
 451
 452             r30              = _mm_mul_ps(rsq30,rinv30);
 453
 454             /* Compute parameters for interactions between i and j atoms */
 455             qq30             = _mm_mul_ps(iq3,jq0);
 456
 457             /* EWALD ELECTROSTATICS */
 458
 459             /* Analytical PME correction */
 460             zeta2            = _mm_mul_ps(beta2,rsq30);
 461             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 462             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 463             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 464             felec            = _mm_mul_ps(qq30,felec);
 465             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 466             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv30);
 467             velec            = _mm_mul_ps(qq30,velec);
 468
 469             d                = _mm_sub_ps(r30,rswitch);
 470             d                = _mm_max_ps(d,_mm_setzero_ps());
 471             d2               = _mm_mul_ps(d,d);
 472             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 473
 474             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 475
 476             /* Evaluate switch function */
 477             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 478             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv30,_mm_mul_ps(velec,dsw)) );
 479             velec            = _mm_mul_ps(velec,sw);
 480             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 481
 482             /* Update potential sum for this i atom from the interaction with this j atom. */
 483             velec            = _mm_and_ps(velec,cutoff_mask);
 484             velecsum         = _mm_add_ps(velecsum,velec);
 485
 486             fscal            = felec;
 487
 488             fscal            = _mm_and_ps(fscal,cutoff_mask);
 489
 490              /* Update vectorial force */
 491             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 492             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 493             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 494
 495             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 496             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 497             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 498
 499             }
 500
 501             fjptrA             = f+j_coord_offsetA;
 502             fjptrB             = f+j_coord_offsetB;
 503             fjptrC             = f+j_coord_offsetC;
 504             fjptrD             = f+j_coord_offsetD;
 505
 506             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 507
 508             /* Inner loop uses 221 flops */
 509         }
 510
 511         if(jidx<j_index_end)
 512         {
 513
 514             /* Get j neighbor index, and coordinate index */
 515             jnrlistA         = jjnr[jidx];
 516             jnrlistB         = jjnr[jidx+1];
 517             jnrlistC         = jjnr[jidx+2];
 518             jnrlistD         = jjnr[jidx+3];
 519             /* Sign of each element will be negative for non-real atoms.
 520              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 521              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 522              */
 523             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 524             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 525             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 526             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 527             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 528             j_coord_offsetA  = DIM*jnrA;
 529             j_coord_offsetB  = DIM*jnrB;
 530             j_coord_offsetC  = DIM*jnrC;
 531             j_coord_offsetD  = DIM*jnrD;
 532
 533             /* load j atom coordinates */
 534             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 535                                               x+j_coord_offsetC,x+j_coord_offsetD,
 536                                               &jx0,&jy0,&jz0);
 537
 538             /* Calculate displacement vector */
 539             dx00             = _mm_sub_ps(ix0,jx0);
 540             dy00             = _mm_sub_ps(iy0,jy0);
 541             dz00             = _mm_sub_ps(iz0,jz0);
 542             dx10             = _mm_sub_ps(ix1,jx0);
 543             dy10             = _mm_sub_ps(iy1,jy0);
 544             dz10             = _mm_sub_ps(iz1,jz0);
 545             dx20             = _mm_sub_ps(ix2,jx0);
 546             dy20             = _mm_sub_ps(iy2,jy0);
 547             dz20             = _mm_sub_ps(iz2,jz0);
 548             dx30             = _mm_sub_ps(ix3,jx0);
 549             dy30             = _mm_sub_ps(iy3,jy0);
 550             dz30             = _mm_sub_ps(iz3,jz0);
 551
 552             /* Calculate squared distance and things based on it */
 553             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 554             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 555             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 556             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 557
 558             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 559             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 560             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 561             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 562
 563             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 564             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 565             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 566             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 567
 568             /* Load parameters for j particles */
 569             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 570                                                               charge+jnrC+0,charge+jnrD+0);
 571             vdwjidx0A        = 2*vdwtype[jnrA+0];
 572             vdwjidx0B        = 2*vdwtype[jnrB+0];
 573             vdwjidx0C        = 2*vdwtype[jnrC+0];
 574             vdwjidx0D        = 2*vdwtype[jnrD+0];
 575
 576             fjx0             = _mm_setzero_ps();
 577             fjy0             = _mm_setzero_ps();
 578             fjz0             = _mm_setzero_ps();
 579
 580             /**************************
 581              * CALCULATE INTERACTIONS *
 582              **************************/
 583
 584             if (gmx_mm_any_lt(rsq00,rcutoff2))
 585             {
 586
 587             r00              = _mm_mul_ps(rsq00,rinv00);
 588             r00              = _mm_andnot_ps(dummy_mask,r00);
 589
 590             /* Compute parameters for interactions between i and j atoms */
 591             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 592                                          vdwparam+vdwioffset0+vdwjidx0B,
 593                                          vdwparam+vdwioffset0+vdwjidx0C,
 594                                          vdwparam+vdwioffset0+vdwjidx0D,
 595                                          &c6_00,&c12_00);
 596
 597             /* LENNARD-JONES DISPERSION/REPULSION */
 598
 599             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 600             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 601             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 602             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 603             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 604
 605             d                = _mm_sub_ps(r00,rswitch);
 606             d                = _mm_max_ps(d,_mm_setzero_ps());
 607             d2               = _mm_mul_ps(d,d);
 608             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 609
 610             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 611
 612             /* Evaluate switch function */
 613             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 614             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 615             vvdw             = _mm_mul_ps(vvdw,sw);
 616             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 617
 618             /* Update potential sum for this i atom from the interaction with this j atom. */
 619             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 620             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 621             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 622
 623             fscal            = fvdw;
 624
 625             fscal            = _mm_and_ps(fscal,cutoff_mask);
 626
 627             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 628
 629              /* Update vectorial force */
 630             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 631             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 632             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 633
 634             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 635             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 636             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 637
 638             }
 639
 640             /**************************
 641              * CALCULATE INTERACTIONS *
 642              **************************/
 643
 644             if (gmx_mm_any_lt(rsq10,rcutoff2))
 645             {
 646
 647             r10              = _mm_mul_ps(rsq10,rinv10);
 648             r10              = _mm_andnot_ps(dummy_mask,r10);
 649
 650             /* Compute parameters for interactions between i and j atoms */
 651             qq10             = _mm_mul_ps(iq1,jq0);
 652
 653             /* EWALD ELECTROSTATICS */
 654
 655             /* Analytical PME correction */
 656             zeta2            = _mm_mul_ps(beta2,rsq10);
 657             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 658             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 659             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 660             felec            = _mm_mul_ps(qq10,felec);
 661             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 662             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 663             velec            = _mm_mul_ps(qq10,velec);
 664
 665             d                = _mm_sub_ps(r10,rswitch);
 666             d                = _mm_max_ps(d,_mm_setzero_ps());
 667             d2               = _mm_mul_ps(d,d);
 668             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 669
 670             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 671
 672             /* Evaluate switch function */
 673             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 674             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 675             velec            = _mm_mul_ps(velec,sw);
 676             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 677
 678             /* Update potential sum for this i atom from the interaction with this j atom. */
 679             velec            = _mm_and_ps(velec,cutoff_mask);
 680             velec            = _mm_andnot_ps(dummy_mask,velec);
 681             velecsum         = _mm_add_ps(velecsum,velec);
 682
 683             fscal            = felec;
 684
 685             fscal            = _mm_and_ps(fscal,cutoff_mask);
 686
 687             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 688
 689              /* Update vectorial force */
 690             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 691             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 692             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 693
 694             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 695             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 696             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 697
 698             }
 699
 700             /**************************
 701              * CALCULATE INTERACTIONS *
 702              **************************/
 703
 704             if (gmx_mm_any_lt(rsq20,rcutoff2))
 705             {
 706
 707             r20              = _mm_mul_ps(rsq20,rinv20);
 708             r20              = _mm_andnot_ps(dummy_mask,r20);
 709
 710             /* Compute parameters for interactions between i and j atoms */
 711             qq20             = _mm_mul_ps(iq2,jq0);
 712
 713             /* EWALD ELECTROSTATICS */
 714
 715             /* Analytical PME correction */
 716             zeta2            = _mm_mul_ps(beta2,rsq20);
 717             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 718             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 719             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 720             felec            = _mm_mul_ps(qq20,felec);
 721             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 722             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 723             velec            = _mm_mul_ps(qq20,velec);
 724
 725             d                = _mm_sub_ps(r20,rswitch);
 726             d                = _mm_max_ps(d,_mm_setzero_ps());
 727             d2               = _mm_mul_ps(d,d);
 728             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 729
 730             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 731
 732             /* Evaluate switch function */
 733             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 734             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
 735             velec            = _mm_mul_ps(velec,sw);
 736             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 737
 738             /* Update potential sum for this i atom from the interaction with this j atom. */
 739             velec            = _mm_and_ps(velec,cutoff_mask);
 740             velec            = _mm_andnot_ps(dummy_mask,velec);
 741             velecsum         = _mm_add_ps(velecsum,velec);
 742
 743             fscal            = felec;
 744
 745             fscal            = _mm_and_ps(fscal,cutoff_mask);
 746
 747             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 748
 749              /* Update vectorial force */
 750             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 751             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 752             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 753
 754             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 755             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 756             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 757
 758             }
 759
 760             /**************************
 761              * CALCULATE INTERACTIONS *
 762              **************************/
 763
 764             if (gmx_mm_any_lt(rsq30,rcutoff2))
 765             {
 766
 767             r30              = _mm_mul_ps(rsq30,rinv30);
 768             r30              = _mm_andnot_ps(dummy_mask,r30);
 769
 770             /* Compute parameters for interactions between i and j atoms */
 771             qq30             = _mm_mul_ps(iq3,jq0);
 772
 773             /* EWALD ELECTROSTATICS */
 774
 775             /* Analytical PME correction */
 776             zeta2            = _mm_mul_ps(beta2,rsq30);
 777             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 778             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 779             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 780             felec            = _mm_mul_ps(qq30,felec);
 781             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 782             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv30);
 783             velec            = _mm_mul_ps(qq30,velec);
 784
 785             d                = _mm_sub_ps(r30,rswitch);
 786             d                = _mm_max_ps(d,_mm_setzero_ps());
 787             d2               = _mm_mul_ps(d,d);
 788             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 789
 790             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 791
 792             /* Evaluate switch function */
 793             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 794             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv30,_mm_mul_ps(velec,dsw)) );
 795             velec            = _mm_mul_ps(velec,sw);
 796             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 797
 798             /* Update potential sum for this i atom from the interaction with this j atom. */
 799             velec            = _mm_and_ps(velec,cutoff_mask);
 800             velec            = _mm_andnot_ps(dummy_mask,velec);
 801             velecsum         = _mm_add_ps(velecsum,velec);
 802
 803             fscal            = felec;
 804
 805             fscal            = _mm_and_ps(fscal,cutoff_mask);
 806
 807             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 808
 809              /* Update vectorial force */
 810             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 811             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 812             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 813
 814             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 815             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 816             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 817
 818             }
 819
 820             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 821             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 822             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 823             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 824
 825             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 826
 827             /* Inner loop uses 225 flops */
 828         }
 829
 830         /* End of innermost loop */
 831
 832         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 833                                               f+i_coord_offset,fshift+i_shift_offset);
 834
 835         ggid                        = gid[iidx];
 836         /* Update potential energies */
 837         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 838         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 839
 840         /* Increment number of inner iterations */
 841         inneriter                  += j_index_end - j_index_start;
 842
 843         /* Outer loop uses 26 flops */
 844     }
 845
 846     /* Increment number of outer iterations */
 847     outeriter        += nri;
 848
 849     /* Update outer/inner flops */
 850
 851     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*225);
 852 }
 853 /*
 854  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_F_avx_128_fma_single
 855  * Electrostatics interaction: Ewald
 856  * VdW interaction:            LennardJones
 857  * Geometry:                   Water4-Particle
 858  * Calculate force/pot:        Force
 859  */
 860 void
 861 nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_F_avx_128_fma_single
 862                     (t_nblist                    * gmx_restrict       nlist,
 863                      rvec                        * gmx_restrict          xx,
 864                      rvec                        * gmx_restrict          ff,
 865                      t_forcerec                  * gmx_restrict          fr,
 866                      t_mdatoms                   * gmx_restrict     mdatoms,
 867                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 868                      t_nrnb                      * gmx_restrict        nrnb)
 869 {
 870     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 871      * just 0 for non-waters.
 872      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 873      * jnr indices corresponding to data put in the four positions in the SIMD register.
 874      */
 875     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 876     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 877     int              jnrA,jnrB,jnrC,jnrD;
 878     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 879     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 880     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 881     real             rcutoff_scalar;
 882     real             *shiftvec,*fshift,*x,*f;
 883     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 884     real             scratch[4*DIM];
 885     __m128           fscal,rcutoff,rcutoff2,jidxall;
 886     int              vdwioffset0;
 887     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 888     int              vdwioffset1;
 889     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 890     int              vdwioffset2;
 891     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 892     int              vdwioffset3;
 893     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 894     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 895     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 896     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 897     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 898     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 899     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 900     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 901     real             *charge;
 902     int              nvdwtype;
 903     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 904     int              *vdwtype;
 905     real             *vdwparam;
 906     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 907     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 908     __m128i          ewitab;
 909     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 910     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 911     real             *ewtab;
 912     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 913     real             rswitch_scalar,d_scalar;
 914     __m128           dummy_mask,cutoff_mask;
 915     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 916     __m128           one     = _mm_set1_ps(1.0);
 917     __m128           two     = _mm_set1_ps(2.0);
 918     x                = xx[0];
 919     f                = ff[0];
 920
 921     nri              = nlist->nri;
 922     iinr             = nlist->iinr;
 923     jindex           = nlist->jindex;
 924     jjnr             = nlist->jjnr;
 925     shiftidx         = nlist->shift;
 926     gid              = nlist->gid;
 927     shiftvec         = fr->shift_vec[0];
 928     fshift           = fr->fshift[0];
 929     facel            = _mm_set1_ps(fr->epsfac);
 930     charge           = mdatoms->chargeA;
 931     nvdwtype         = fr->ntype;
 932     vdwparam         = fr->nbfp;
 933     vdwtype          = mdatoms->typeA;
 934
 935     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 936     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
 937     beta2            = _mm_mul_ps(beta,beta);
 938     beta3            = _mm_mul_ps(beta,beta2);
 939     ewtab            = fr->ic->tabq_coul_FDV0;
 940     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 941     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 942
 943     /* Setup water-specific parameters */
 944     inr              = nlist->iinr[0];
 945     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 946     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 947     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 948     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 949
 950     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 951     rcutoff_scalar   = fr->rcoulomb;
 952     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 953     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 954
 955     rswitch_scalar   = fr->rcoulomb_switch;
 956     rswitch          = _mm_set1_ps(rswitch_scalar);
 957     /* Setup switch parameters */
 958     d_scalar         = rcutoff_scalar-rswitch_scalar;
 959     d                = _mm_set1_ps(d_scalar);
 960     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 961     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 962     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 963     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 964     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 965     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 966
 967     /* Avoid stupid compiler warnings */
 968     jnrA = jnrB = jnrC = jnrD = 0;
 969     j_coord_offsetA = 0;
 970     j_coord_offsetB = 0;
 971     j_coord_offsetC = 0;
 972     j_coord_offsetD = 0;
 973
 974     outeriter        = 0;
 975     inneriter        = 0;
 976
 977     for(iidx=0;iidx<4*DIM;iidx++)
 978     {
 979         scratch[iidx] = 0.0;
 980     }
 981
 982     /* Start outer loop over neighborlists */
 983     for(iidx=0; iidx<nri; iidx++)
 984     {
 985         /* Load shift vector for this list */
 986         i_shift_offset   = DIM*shiftidx[iidx];
 987
 988         /* Load limits for loop over neighbors */
 989         j_index_start    = jindex[iidx];
 990         j_index_end      = jindex[iidx+1];
 991
 992         /* Get outer coordinate index */
 993         inr              = iinr[iidx];
 994         i_coord_offset   = DIM*inr;
 995
 996         /* Load i particle coords and add shift vector */
 997         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 998                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 999
1000         fix0             = _mm_setzero_ps();
1001         fiy0             = _mm_setzero_ps();
1002         fiz0             = _mm_setzero_ps();
1003         fix1             = _mm_setzero_ps();
1004         fiy1             = _mm_setzero_ps();
1005         fiz1             = _mm_setzero_ps();
1006         fix2             = _mm_setzero_ps();
1007         fiy2             = _mm_setzero_ps();
1008         fiz2             = _mm_setzero_ps();
1009         fix3             = _mm_setzero_ps();
1010         fiy3             = _mm_setzero_ps();
1011         fiz3             = _mm_setzero_ps();
1012
1013         /* Start inner kernel loop */
1014         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1015         {
1016
1017             /* Get j neighbor index, and coordinate index */
1018             jnrA             = jjnr[jidx];
1019             jnrB             = jjnr[jidx+1];
1020             jnrC             = jjnr[jidx+2];
1021             jnrD             = jjnr[jidx+3];
1022             j_coord_offsetA  = DIM*jnrA;
1023             j_coord_offsetB  = DIM*jnrB;
1024             j_coord_offsetC  = DIM*jnrC;
1025             j_coord_offsetD  = DIM*jnrD;
1026
1027             /* load j atom coordinates */
1028             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1029                                               x+j_coord_offsetC,x+j_coord_offsetD,
1030                                               &jx0,&jy0,&jz0);
1031
1032             /* Calculate displacement vector */
1033             dx00             = _mm_sub_ps(ix0,jx0);
1034             dy00             = _mm_sub_ps(iy0,jy0);
1035             dz00             = _mm_sub_ps(iz0,jz0);
1036             dx10             = _mm_sub_ps(ix1,jx0);
1037             dy10             = _mm_sub_ps(iy1,jy0);
1038             dz10             = _mm_sub_ps(iz1,jz0);
1039             dx20             = _mm_sub_ps(ix2,jx0);
1040             dy20             = _mm_sub_ps(iy2,jy0);
1041             dz20             = _mm_sub_ps(iz2,jz0);
1042             dx30             = _mm_sub_ps(ix3,jx0);
1043             dy30             = _mm_sub_ps(iy3,jy0);
1044             dz30             = _mm_sub_ps(iz3,jz0);
1045
1046             /* Calculate squared distance and things based on it */
1047             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1048             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1049             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1050             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1051
1052             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1053             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1054             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1055             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1056
1057             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1058             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1059             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1060             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1061
1062             /* Load parameters for j particles */
1063             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1064                                                               charge+jnrC+0,charge+jnrD+0);
1065             vdwjidx0A        = 2*vdwtype[jnrA+0];
1066             vdwjidx0B        = 2*vdwtype[jnrB+0];
1067             vdwjidx0C        = 2*vdwtype[jnrC+0];
1068             vdwjidx0D        = 2*vdwtype[jnrD+0];
1069
1070             fjx0             = _mm_setzero_ps();
1071             fjy0             = _mm_setzero_ps();
1072             fjz0             = _mm_setzero_ps();
1073
1074             /**************************
1075              * CALCULATE INTERACTIONS *
1076              **************************/
1077
1078             if (gmx_mm_any_lt(rsq00,rcutoff2))
1079             {
1080
1081             r00              = _mm_mul_ps(rsq00,rinv00);
1082
1083             /* Compute parameters for interactions between i and j atoms */
1084             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1085                                          vdwparam+vdwioffset0+vdwjidx0B,
1086                                          vdwparam+vdwioffset0+vdwjidx0C,
1087                                          vdwparam+vdwioffset0+vdwjidx0D,
1088                                          &c6_00,&c12_00);
1089
1090             /* LENNARD-JONES DISPERSION/REPULSION */
1091
1092             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1093             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1094             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1095             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
1096             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1097
1098             d                = _mm_sub_ps(r00,rswitch);
1099             d                = _mm_max_ps(d,_mm_setzero_ps());
1100             d2               = _mm_mul_ps(d,d);
1101             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1102
1103             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1104
1105             /* Evaluate switch function */
1106             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1107             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1108             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1109
1110             fscal            = fvdw;
1111
1112             fscal            = _mm_and_ps(fscal,cutoff_mask);
1113
1114              /* Update vectorial force */
1115             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1116             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1117             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1118
1119             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1120             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1121             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1122
1123             }
1124
1125             /**************************
1126              * CALCULATE INTERACTIONS *
1127              **************************/
1128
1129             if (gmx_mm_any_lt(rsq10,rcutoff2))
1130             {
1131
1132             r10              = _mm_mul_ps(rsq10,rinv10);
1133
1134             /* Compute parameters for interactions between i and j atoms */
1135             qq10             = _mm_mul_ps(iq1,jq0);
1136
1137             /* EWALD ELECTROSTATICS */
1138
1139             /* Analytical PME correction */
1140             zeta2            = _mm_mul_ps(beta2,rsq10);
1141             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1142             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1143             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1144             felec            = _mm_mul_ps(qq10,felec);
1145             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1146             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
1147             velec            = _mm_mul_ps(qq10,velec);
1148
1149             d                = _mm_sub_ps(r10,rswitch);
1150             d                = _mm_max_ps(d,_mm_setzero_ps());
1151             d2               = _mm_mul_ps(d,d);
1152             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1153
1154             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1155
1156             /* Evaluate switch function */
1157             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1158             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
1159             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1160
1161             fscal            = felec;
1162
1163             fscal            = _mm_and_ps(fscal,cutoff_mask);
1164
1165              /* Update vectorial force */
1166             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1167             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1168             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1169
1170             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1171             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1172             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1173
1174             }
1175
1176             /**************************
1177              * CALCULATE INTERACTIONS *
1178              **************************/
1179
1180             if (gmx_mm_any_lt(rsq20,rcutoff2))
1181             {
1182
1183             r20              = _mm_mul_ps(rsq20,rinv20);
1184
1185             /* Compute parameters for interactions between i and j atoms */
1186             qq20             = _mm_mul_ps(iq2,jq0);
1187
1188             /* EWALD ELECTROSTATICS */
1189
1190             /* Analytical PME correction */
1191             zeta2            = _mm_mul_ps(beta2,rsq20);
1192             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1193             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1194             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1195             felec            = _mm_mul_ps(qq20,felec);
1196             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1197             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
1198             velec            = _mm_mul_ps(qq20,velec);
1199
1200             d                = _mm_sub_ps(r20,rswitch);
1201             d                = _mm_max_ps(d,_mm_setzero_ps());
1202             d2               = _mm_mul_ps(d,d);
1203             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1204
1205             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1206
1207             /* Evaluate switch function */
1208             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1209             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1210             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1211
1212             fscal            = felec;
1213
1214             fscal            = _mm_and_ps(fscal,cutoff_mask);
1215
1216              /* Update vectorial force */
1217             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1218             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1219             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1220
1221             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1222             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1223             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1224
1225             }
1226
1227             /**************************
1228              * CALCULATE INTERACTIONS *
1229              **************************/
1230
1231             if (gmx_mm_any_lt(rsq30,rcutoff2))
1232             {
1233
1234             r30              = _mm_mul_ps(rsq30,rinv30);
1235
1236             /* Compute parameters for interactions between i and j atoms */
1237             qq30             = _mm_mul_ps(iq3,jq0);
1238
1239             /* EWALD ELECTROSTATICS */
1240
1241             /* Analytical PME correction */
1242             zeta2            = _mm_mul_ps(beta2,rsq30);
1243             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
1244             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1245             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1246             felec            = _mm_mul_ps(qq30,felec);
1247             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1248             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv30);
1249             velec            = _mm_mul_ps(qq30,velec);
1250
1251             d                = _mm_sub_ps(r30,rswitch);
1252             d                = _mm_max_ps(d,_mm_setzero_ps());
1253             d2               = _mm_mul_ps(d,d);
1254             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1255
1256             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1257
1258             /* Evaluate switch function */
1259             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1260             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv30,_mm_mul_ps(velec,dsw)) );
1261             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1262
1263             fscal            = felec;
1264
1265             fscal            = _mm_and_ps(fscal,cutoff_mask);
1266
1267              /* Update vectorial force */
1268             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1269             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1270             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1271
1272             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1273             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1274             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1275
1276             }
1277
1278             fjptrA             = f+j_coord_offsetA;
1279             fjptrB             = f+j_coord_offsetB;
1280             fjptrC             = f+j_coord_offsetC;
1281             fjptrD             = f+j_coord_offsetD;
1282
1283             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1284
1285             /* Inner loop uses 209 flops */
1286         }
1287
1288         if(jidx<j_index_end)
1289         {
1290
1291             /* Get j neighbor index, and coordinate index */
1292             jnrlistA         = jjnr[jidx];
1293             jnrlistB         = jjnr[jidx+1];
1294             jnrlistC         = jjnr[jidx+2];
1295             jnrlistD         = jjnr[jidx+3];
1296             /* Sign of each element will be negative for non-real atoms.
1297              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1298              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1299              */
1300             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1301             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1302             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1303             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1304             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1305             j_coord_offsetA  = DIM*jnrA;
1306             j_coord_offsetB  = DIM*jnrB;
1307             j_coord_offsetC  = DIM*jnrC;
1308             j_coord_offsetD  = DIM*jnrD;
1309
1310             /* load j atom coordinates */
1311             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1312                                               x+j_coord_offsetC,x+j_coord_offsetD,
1313                                               &jx0,&jy0,&jz0);
1314
1315             /* Calculate displacement vector */
1316             dx00             = _mm_sub_ps(ix0,jx0);
1317             dy00             = _mm_sub_ps(iy0,jy0);
1318             dz00             = _mm_sub_ps(iz0,jz0);
1319             dx10             = _mm_sub_ps(ix1,jx0);
1320             dy10             = _mm_sub_ps(iy1,jy0);
1321             dz10             = _mm_sub_ps(iz1,jz0);
1322             dx20             = _mm_sub_ps(ix2,jx0);
1323             dy20             = _mm_sub_ps(iy2,jy0);
1324             dz20             = _mm_sub_ps(iz2,jz0);
1325             dx30             = _mm_sub_ps(ix3,jx0);
1326             dy30             = _mm_sub_ps(iy3,jy0);
1327             dz30             = _mm_sub_ps(iz3,jz0);
1328
1329             /* Calculate squared distance and things based on it */
1330             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1331             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1332             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1333             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1334
1335             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1336             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1337             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1338             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1339
1340             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1341             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1342             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1343             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1344
1345             /* Load parameters for j particles */
1346             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1347                                                               charge+jnrC+0,charge+jnrD+0);
1348             vdwjidx0A        = 2*vdwtype[jnrA+0];
1349             vdwjidx0B        = 2*vdwtype[jnrB+0];
1350             vdwjidx0C        = 2*vdwtype[jnrC+0];
1351             vdwjidx0D        = 2*vdwtype[jnrD+0];
1352
1353             fjx0             = _mm_setzero_ps();
1354             fjy0             = _mm_setzero_ps();
1355             fjz0             = _mm_setzero_ps();
1356
1357             /**************************
1358              * CALCULATE INTERACTIONS *
1359              **************************/
1360
1361             if (gmx_mm_any_lt(rsq00,rcutoff2))
1362             {
1363
1364             r00              = _mm_mul_ps(rsq00,rinv00);
1365             r00              = _mm_andnot_ps(dummy_mask,r00);
1366
1367             /* Compute parameters for interactions between i and j atoms */
1368             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1369                                          vdwparam+vdwioffset0+vdwjidx0B,
1370                                          vdwparam+vdwioffset0+vdwjidx0C,
1371                                          vdwparam+vdwioffset0+vdwjidx0D,
1372                                          &c6_00,&c12_00);
1373
1374             /* LENNARD-JONES DISPERSION/REPULSION */
1375
1376             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1377             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1378             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1379             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
1380             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1381
1382             d                = _mm_sub_ps(r00,rswitch);
1383             d                = _mm_max_ps(d,_mm_setzero_ps());
1384             d2               = _mm_mul_ps(d,d);
1385             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1386
1387             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1388
1389             /* Evaluate switch function */
1390             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1391             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1392             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1393
1394             fscal            = fvdw;
1395
1396             fscal            = _mm_and_ps(fscal,cutoff_mask);
1397
1398             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1399
1400              /* Update vectorial force */
1401             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1402             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1403             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1404
1405             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1406             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1407             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1408
1409             }
1410
1411             /**************************
1412              * CALCULATE INTERACTIONS *
1413              **************************/
1414
1415             if (gmx_mm_any_lt(rsq10,rcutoff2))
1416             {
1417
1418             r10              = _mm_mul_ps(rsq10,rinv10);
1419             r10              = _mm_andnot_ps(dummy_mask,r10);
1420
1421             /* Compute parameters for interactions between i and j atoms */
1422             qq10             = _mm_mul_ps(iq1,jq0);
1423
1424             /* EWALD ELECTROSTATICS */
1425
1426             /* Analytical PME correction */
1427             zeta2            = _mm_mul_ps(beta2,rsq10);
1428             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1429             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1430             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1431             felec            = _mm_mul_ps(qq10,felec);
1432             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1433             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
1434             velec            = _mm_mul_ps(qq10,velec);
1435
1436             d                = _mm_sub_ps(r10,rswitch);
1437             d                = _mm_max_ps(d,_mm_setzero_ps());
1438             d2               = _mm_mul_ps(d,d);
1439             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1440
1441             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1442
1443             /* Evaluate switch function */
1444             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1445             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
1446             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1447
1448             fscal            = felec;
1449
1450             fscal            = _mm_and_ps(fscal,cutoff_mask);
1451
1452             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1453
1454              /* Update vectorial force */
1455             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1456             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1457             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1458
1459             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1460             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1461             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1462
1463             }
1464
1465             /**************************
1466              * CALCULATE INTERACTIONS *
1467              **************************/
1468
1469             if (gmx_mm_any_lt(rsq20,rcutoff2))
1470             {
1471
1472             r20              = _mm_mul_ps(rsq20,rinv20);
1473             r20              = _mm_andnot_ps(dummy_mask,r20);
1474
1475             /* Compute parameters for interactions between i and j atoms */
1476             qq20             = _mm_mul_ps(iq2,jq0);
1477
1478             /* EWALD ELECTROSTATICS */
1479
1480             /* Analytical PME correction */
1481             zeta2            = _mm_mul_ps(beta2,rsq20);
1482             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1483             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1484             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1485             felec            = _mm_mul_ps(qq20,felec);
1486             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1487             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
1488             velec            = _mm_mul_ps(qq20,velec);
1489
1490             d                = _mm_sub_ps(r20,rswitch);
1491             d                = _mm_max_ps(d,_mm_setzero_ps());
1492             d2               = _mm_mul_ps(d,d);
1493             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1494
1495             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1496
1497             /* Evaluate switch function */
1498             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1499             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1500             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1501
1502             fscal            = felec;
1503
1504             fscal            = _mm_and_ps(fscal,cutoff_mask);
1505
1506             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1507
1508              /* Update vectorial force */
1509             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1510             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1511             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1512
1513             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1514             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1515             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1516
1517             }
1518
1519             /**************************
1520              * CALCULATE INTERACTIONS *
1521              **************************/
1522
1523             if (gmx_mm_any_lt(rsq30,rcutoff2))
1524             {
1525
1526             r30              = _mm_mul_ps(rsq30,rinv30);
1527             r30              = _mm_andnot_ps(dummy_mask,r30);
1528
1529             /* Compute parameters for interactions between i and j atoms */
1530             qq30             = _mm_mul_ps(iq3,jq0);
1531
1532             /* EWALD ELECTROSTATICS */
1533
1534             /* Analytical PME correction */
1535             zeta2            = _mm_mul_ps(beta2,rsq30);
1536             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
1537             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1538             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1539             felec            = _mm_mul_ps(qq30,felec);
1540             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1541             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv30);
1542             velec            = _mm_mul_ps(qq30,velec);
1543
1544             d                = _mm_sub_ps(r30,rswitch);
1545             d                = _mm_max_ps(d,_mm_setzero_ps());
1546             d2               = _mm_mul_ps(d,d);
1547             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1548
1549             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1550
1551             /* Evaluate switch function */
1552             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1553             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv30,_mm_mul_ps(velec,dsw)) );
1554             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1555
1556             fscal            = felec;
1557
1558             fscal            = _mm_and_ps(fscal,cutoff_mask);
1559
1560             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1561
1562              /* Update vectorial force */
1563             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1564             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1565             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1566
1567             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1568             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1569             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1570
1571             }
1572
1573             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1574             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1575             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1576             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1577
1578             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1579
1580             /* Inner loop uses 213 flops */
1581         }
1582
1583         /* End of innermost loop */
1584
1585         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1586                                               f+i_coord_offset,fshift+i_shift_offset);
1587
1588         /* Increment number of inner iterations */
1589         inneriter                  += j_index_end - j_index_start;
1590
1591         /* Outer loop uses 24 flops */
1592     }
1593
1594     /* Increment number of outer iterations */
1595     outeriter        += nri;
1596
1597     /* Update outer/inner flops */
1598
1599     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*213);
1600 }