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

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_128_fma_single.h"
  34 #include "kernelutil_x86_avx_128_fma_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_VF_avx_128_fma_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_VF_avx_128_fma_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  76     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  77     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  78     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  79     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  80     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  87     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  88     __m128i          ewitab;
  89     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  90     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  91     real             *ewtab;
  92     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  93     real             rswitch_scalar,d_scalar;
  94     __m128           dummy_mask,cutoff_mask;
  95     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  96     __m128           one     = _mm_set1_ps(1.0);
  97     __m128           two     = _mm_set1_ps(2.0);
  98     x                = xx[0];
  99     f                = ff[0];
 100
 101     nri              = nlist->nri;
 102     iinr             = nlist->iinr;
 103     jindex           = nlist->jindex;
 104     jjnr             = nlist->jjnr;
 105     shiftidx         = nlist->shift;
 106     gid              = nlist->gid;
 107     shiftvec         = fr->shift_vec[0];
 108     fshift           = fr->fshift[0];
 109     facel            = _mm_set1_ps(fr->epsfac);
 110     charge           = mdatoms->chargeA;
 111     nvdwtype         = fr->ntype;
 112     vdwparam         = fr->nbfp;
 113     vdwtype          = mdatoms->typeA;
 114
 115     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 116     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
 117     beta2            = _mm_mul_ps(beta,beta);
 118     beta3            = _mm_mul_ps(beta,beta2);
 119     ewtab            = fr->ic->tabq_coul_FDV0;
 120     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 121     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 122
 123     /* Setup water-specific parameters */
 124     inr              = nlist->iinr[0];
 125     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 126     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 127     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 128     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 129
 130     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 131     rcutoff_scalar   = fr->rcoulomb;
 132     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 133     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 134
 135     rswitch_scalar   = fr->rcoulomb_switch;
 136     rswitch          = _mm_set1_ps(rswitch_scalar);
 137     /* Setup switch parameters */
 138     d_scalar         = rcutoff_scalar-rswitch_scalar;
 139     d                = _mm_set1_ps(d_scalar);
 140     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 141     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 142     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 143     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 144     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 145     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 146
 147     /* Avoid stupid compiler warnings */
 148     jnrA = jnrB = jnrC = jnrD = 0;
 149     j_coord_offsetA = 0;
 150     j_coord_offsetB = 0;
 151     j_coord_offsetC = 0;
 152     j_coord_offsetD = 0;
 153
 154     outeriter        = 0;
 155     inneriter        = 0;
 156
 157     for(iidx=0;iidx<4*DIM;iidx++)
 158     {
 159         scratch[iidx] = 0.0;
 160     }
 161
 162     /* Start outer loop over neighborlists */
 163     for(iidx=0; iidx<nri; iidx++)
 164     {
 165         /* Load shift vector for this list */
 166         i_shift_offset   = DIM*shiftidx[iidx];
 167
 168         /* Load limits for loop over neighbors */
 169         j_index_start    = jindex[iidx];
 170         j_index_end      = jindex[iidx+1];
 171
 172         /* Get outer coordinate index */
 173         inr              = iinr[iidx];
 174         i_coord_offset   = DIM*inr;
 175
 176         /* Load i particle coords and add shift vector */
 177         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 178                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 179
 180         fix0             = _mm_setzero_ps();
 181         fiy0             = _mm_setzero_ps();
 182         fiz0             = _mm_setzero_ps();
 183         fix1             = _mm_setzero_ps();
 184         fiy1             = _mm_setzero_ps();
 185         fiz1             = _mm_setzero_ps();
 186         fix2             = _mm_setzero_ps();
 187         fiy2             = _mm_setzero_ps();
 188         fiz2             = _mm_setzero_ps();
 189
 190         /* Reset potential sums */
 191         velecsum         = _mm_setzero_ps();
 192         vvdwsum          = _mm_setzero_ps();
 193
 194         /* Start inner kernel loop */
 195         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 196         {
 197
 198             /* Get j neighbor index, and coordinate index */
 199             jnrA             = jjnr[jidx];
 200             jnrB             = jjnr[jidx+1];
 201             jnrC             = jjnr[jidx+2];
 202             jnrD             = jjnr[jidx+3];
 203             j_coord_offsetA  = DIM*jnrA;
 204             j_coord_offsetB  = DIM*jnrB;
 205             j_coord_offsetC  = DIM*jnrC;
 206             j_coord_offsetD  = DIM*jnrD;
 207
 208             /* load j atom coordinates */
 209             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 210                                               x+j_coord_offsetC,x+j_coord_offsetD,
 211                                               &jx0,&jy0,&jz0);
 212
 213             /* Calculate displacement vector */
 214             dx00             = _mm_sub_ps(ix0,jx0);
 215             dy00             = _mm_sub_ps(iy0,jy0);
 216             dz00             = _mm_sub_ps(iz0,jz0);
 217             dx10             = _mm_sub_ps(ix1,jx0);
 218             dy10             = _mm_sub_ps(iy1,jy0);
 219             dz10             = _mm_sub_ps(iz1,jz0);
 220             dx20             = _mm_sub_ps(ix2,jx0);
 221             dy20             = _mm_sub_ps(iy2,jy0);
 222             dz20             = _mm_sub_ps(iz2,jz0);
 223
 224             /* Calculate squared distance and things based on it */
 225             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 226             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 227             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 228
 229             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 230             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 231             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 232
 233             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 234             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 235             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 236
 237             /* Load parameters for j particles */
 238             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 239                                                               charge+jnrC+0,charge+jnrD+0);
 240             vdwjidx0A        = 2*vdwtype[jnrA+0];
 241             vdwjidx0B        = 2*vdwtype[jnrB+0];
 242             vdwjidx0C        = 2*vdwtype[jnrC+0];
 243             vdwjidx0D        = 2*vdwtype[jnrD+0];
 244
 245             fjx0             = _mm_setzero_ps();
 246             fjy0             = _mm_setzero_ps();
 247             fjz0             = _mm_setzero_ps();
 248
 249             /**************************
 250              * CALCULATE INTERACTIONS *
 251              **************************/
 252
 253             if (gmx_mm_any_lt(rsq00,rcutoff2))
 254             {
 255
 256             r00              = _mm_mul_ps(rsq00,rinv00);
 257
 258             /* Compute parameters for interactions between i and j atoms */
 259             qq00             = _mm_mul_ps(iq0,jq0);
 260             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 261                                          vdwparam+vdwioffset0+vdwjidx0B,
 262                                          vdwparam+vdwioffset0+vdwjidx0C,
 263                                          vdwparam+vdwioffset0+vdwjidx0D,
 264                                          &c6_00,&c12_00);
 265
 266             /* EWALD ELECTROSTATICS */
 267
 268             /* Analytical PME correction */
 269             zeta2            = _mm_mul_ps(beta2,rsq00);
 270             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 271             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 272             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 273             felec            = _mm_mul_ps(qq00,felec);
 274             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 275             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 276             velec            = _mm_mul_ps(qq00,velec);
 277
 278             /* LENNARD-JONES DISPERSION/REPULSION */
 279
 280             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 281             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 282             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 283             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 284             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 285
 286             d                = _mm_sub_ps(r00,rswitch);
 287             d                = _mm_max_ps(d,_mm_setzero_ps());
 288             d2               = _mm_mul_ps(d,d);
 289             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 290
 291             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 292
 293             /* Evaluate switch function */
 294             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 295             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 296             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 297             velec            = _mm_mul_ps(velec,sw);
 298             vvdw             = _mm_mul_ps(vvdw,sw);
 299             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 300
 301             /* Update potential sum for this i atom from the interaction with this j atom. */
 302             velec            = _mm_and_ps(velec,cutoff_mask);
 303             velecsum         = _mm_add_ps(velecsum,velec);
 304             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 305             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 306
 307             fscal            = _mm_add_ps(felec,fvdw);
 308
 309             fscal            = _mm_and_ps(fscal,cutoff_mask);
 310
 311              /* Update vectorial force */
 312             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 313             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 314             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 315
 316             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 317             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 318             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 319
 320             }
 321
 322             /**************************
 323              * CALCULATE INTERACTIONS *
 324              **************************/
 325
 326             if (gmx_mm_any_lt(rsq10,rcutoff2))
 327             {
 328
 329             r10              = _mm_mul_ps(rsq10,rinv10);
 330
 331             /* Compute parameters for interactions between i and j atoms */
 332             qq10             = _mm_mul_ps(iq1,jq0);
 333
 334             /* EWALD ELECTROSTATICS */
 335
 336             /* Analytical PME correction */
 337             zeta2            = _mm_mul_ps(beta2,rsq10);
 338             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 339             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 340             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 341             felec            = _mm_mul_ps(qq10,felec);
 342             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 343             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 344             velec            = _mm_mul_ps(qq10,velec);
 345
 346             d                = _mm_sub_ps(r10,rswitch);
 347             d                = _mm_max_ps(d,_mm_setzero_ps());
 348             d2               = _mm_mul_ps(d,d);
 349             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 350
 351             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 352
 353             /* Evaluate switch function */
 354             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 355             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 356             velec            = _mm_mul_ps(velec,sw);
 357             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 358
 359             /* Update potential sum for this i atom from the interaction with this j atom. */
 360             velec            = _mm_and_ps(velec,cutoff_mask);
 361             velecsum         = _mm_add_ps(velecsum,velec);
 362
 363             fscal            = felec;
 364
 365             fscal            = _mm_and_ps(fscal,cutoff_mask);
 366
 367              /* Update vectorial force */
 368             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 369             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 370             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 371
 372             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 373             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 374             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 375
 376             }
 377
 378             /**************************
 379              * CALCULATE INTERACTIONS *
 380              **************************/
 381
 382             if (gmx_mm_any_lt(rsq20,rcutoff2))
 383             {
 384
 385             r20              = _mm_mul_ps(rsq20,rinv20);
 386
 387             /* Compute parameters for interactions between i and j atoms */
 388             qq20             = _mm_mul_ps(iq2,jq0);
 389
 390             /* EWALD ELECTROSTATICS */
 391
 392             /* Analytical PME correction */
 393             zeta2            = _mm_mul_ps(beta2,rsq20);
 394             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 395             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 396             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 397             felec            = _mm_mul_ps(qq20,felec);
 398             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 399             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 400             velec            = _mm_mul_ps(qq20,velec);
 401
 402             d                = _mm_sub_ps(r20,rswitch);
 403             d                = _mm_max_ps(d,_mm_setzero_ps());
 404             d2               = _mm_mul_ps(d,d);
 405             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 406
 407             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 408
 409             /* Evaluate switch function */
 410             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 411             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
 412             velec            = _mm_mul_ps(velec,sw);
 413             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 414
 415             /* Update potential sum for this i atom from the interaction with this j atom. */
 416             velec            = _mm_and_ps(velec,cutoff_mask);
 417             velecsum         = _mm_add_ps(velecsum,velec);
 418
 419             fscal            = felec;
 420
 421             fscal            = _mm_and_ps(fscal,cutoff_mask);
 422
 423              /* Update vectorial force */
 424             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 425             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 426             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 427
 428             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 429             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 430             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 431
 432             }
 433
 434             fjptrA             = f+j_coord_offsetA;
 435             fjptrB             = f+j_coord_offsetB;
 436             fjptrC             = f+j_coord_offsetC;
 437             fjptrD             = f+j_coord_offsetD;
 438
 439             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 440
 441             /* Inner loop uses 177 flops */
 442         }
 443
 444         if(jidx<j_index_end)
 445         {
 446
 447             /* Get j neighbor index, and coordinate index */
 448             jnrlistA         = jjnr[jidx];
 449             jnrlistB         = jjnr[jidx+1];
 450             jnrlistC         = jjnr[jidx+2];
 451             jnrlistD         = jjnr[jidx+3];
 452             /* Sign of each element will be negative for non-real atoms.
 453              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 454              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 455              */
 456             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 457             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 458             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 459             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 460             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 461             j_coord_offsetA  = DIM*jnrA;
 462             j_coord_offsetB  = DIM*jnrB;
 463             j_coord_offsetC  = DIM*jnrC;
 464             j_coord_offsetD  = DIM*jnrD;
 465
 466             /* load j atom coordinates */
 467             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 468                                               x+j_coord_offsetC,x+j_coord_offsetD,
 469                                               &jx0,&jy0,&jz0);
 470
 471             /* Calculate displacement vector */
 472             dx00             = _mm_sub_ps(ix0,jx0);
 473             dy00             = _mm_sub_ps(iy0,jy0);
 474             dz00             = _mm_sub_ps(iz0,jz0);
 475             dx10             = _mm_sub_ps(ix1,jx0);
 476             dy10             = _mm_sub_ps(iy1,jy0);
 477             dz10             = _mm_sub_ps(iz1,jz0);
 478             dx20             = _mm_sub_ps(ix2,jx0);
 479             dy20             = _mm_sub_ps(iy2,jy0);
 480             dz20             = _mm_sub_ps(iz2,jz0);
 481
 482             /* Calculate squared distance and things based on it */
 483             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 484             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 485             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 486
 487             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 488             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 489             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 490
 491             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 492             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 493             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 494
 495             /* Load parameters for j particles */
 496             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 497                                                               charge+jnrC+0,charge+jnrD+0);
 498             vdwjidx0A        = 2*vdwtype[jnrA+0];
 499             vdwjidx0B        = 2*vdwtype[jnrB+0];
 500             vdwjidx0C        = 2*vdwtype[jnrC+0];
 501             vdwjidx0D        = 2*vdwtype[jnrD+0];
 502
 503             fjx0             = _mm_setzero_ps();
 504             fjy0             = _mm_setzero_ps();
 505             fjz0             = _mm_setzero_ps();
 506
 507             /**************************
 508              * CALCULATE INTERACTIONS *
 509              **************************/
 510
 511             if (gmx_mm_any_lt(rsq00,rcutoff2))
 512             {
 513
 514             r00              = _mm_mul_ps(rsq00,rinv00);
 515             r00              = _mm_andnot_ps(dummy_mask,r00);
 516
 517             /* Compute parameters for interactions between i and j atoms */
 518             qq00             = _mm_mul_ps(iq0,jq0);
 519             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 520                                          vdwparam+vdwioffset0+vdwjidx0B,
 521                                          vdwparam+vdwioffset0+vdwjidx0C,
 522                                          vdwparam+vdwioffset0+vdwjidx0D,
 523                                          &c6_00,&c12_00);
 524
 525             /* EWALD ELECTROSTATICS */
 526
 527             /* Analytical PME correction */
 528             zeta2            = _mm_mul_ps(beta2,rsq00);
 529             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 530             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 531             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 532             felec            = _mm_mul_ps(qq00,felec);
 533             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 534             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 535             velec            = _mm_mul_ps(qq00,velec);
 536
 537             /* LENNARD-JONES DISPERSION/REPULSION */
 538
 539             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 540             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 541             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 542             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 543             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 544
 545             d                = _mm_sub_ps(r00,rswitch);
 546             d                = _mm_max_ps(d,_mm_setzero_ps());
 547             d2               = _mm_mul_ps(d,d);
 548             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 549
 550             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 551
 552             /* Evaluate switch function */
 553             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 554             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 555             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 556             velec            = _mm_mul_ps(velec,sw);
 557             vvdw             = _mm_mul_ps(vvdw,sw);
 558             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 559
 560             /* Update potential sum for this i atom from the interaction with this j atom. */
 561             velec            = _mm_and_ps(velec,cutoff_mask);
 562             velec            = _mm_andnot_ps(dummy_mask,velec);
 563             velecsum         = _mm_add_ps(velecsum,velec);
 564             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 565             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 566             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 567
 568             fscal            = _mm_add_ps(felec,fvdw);
 569
 570             fscal            = _mm_and_ps(fscal,cutoff_mask);
 571
 572             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 573
 574              /* Update vectorial force */
 575             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 576             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 577             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 578
 579             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 580             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 581             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 582
 583             }
 584
 585             /**************************
 586              * CALCULATE INTERACTIONS *
 587              **************************/
 588
 589             if (gmx_mm_any_lt(rsq10,rcutoff2))
 590             {
 591
 592             r10              = _mm_mul_ps(rsq10,rinv10);
 593             r10              = _mm_andnot_ps(dummy_mask,r10);
 594
 595             /* Compute parameters for interactions between i and j atoms */
 596             qq10             = _mm_mul_ps(iq1,jq0);
 597
 598             /* EWALD ELECTROSTATICS */
 599
 600             /* Analytical PME correction */
 601             zeta2            = _mm_mul_ps(beta2,rsq10);
 602             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 603             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 604             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 605             felec            = _mm_mul_ps(qq10,felec);
 606             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 607             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 608             velec            = _mm_mul_ps(qq10,velec);
 609
 610             d                = _mm_sub_ps(r10,rswitch);
 611             d                = _mm_max_ps(d,_mm_setzero_ps());
 612             d2               = _mm_mul_ps(d,d);
 613             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 614
 615             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 616
 617             /* Evaluate switch function */
 618             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 619             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 620             velec            = _mm_mul_ps(velec,sw);
 621             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 622
 623             /* Update potential sum for this i atom from the interaction with this j atom. */
 624             velec            = _mm_and_ps(velec,cutoff_mask);
 625             velec            = _mm_andnot_ps(dummy_mask,velec);
 626             velecsum         = _mm_add_ps(velecsum,velec);
 627
 628             fscal            = felec;
 629
 630             fscal            = _mm_and_ps(fscal,cutoff_mask);
 631
 632             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 633
 634              /* Update vectorial force */
 635             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 636             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 637             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 638
 639             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 640             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 641             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 642
 643             }
 644
 645             /**************************
 646              * CALCULATE INTERACTIONS *
 647              **************************/
 648
 649             if (gmx_mm_any_lt(rsq20,rcutoff2))
 650             {
 651
 652             r20              = _mm_mul_ps(rsq20,rinv20);
 653             r20              = _mm_andnot_ps(dummy_mask,r20);
 654
 655             /* Compute parameters for interactions between i and j atoms */
 656             qq20             = _mm_mul_ps(iq2,jq0);
 657
 658             /* EWALD ELECTROSTATICS */
 659
 660             /* Analytical PME correction */
 661             zeta2            = _mm_mul_ps(beta2,rsq20);
 662             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 663             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 664             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 665             felec            = _mm_mul_ps(qq20,felec);
 666             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 667             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 668             velec            = _mm_mul_ps(qq20,velec);
 669
 670             d                = _mm_sub_ps(r20,rswitch);
 671             d                = _mm_max_ps(d,_mm_setzero_ps());
 672             d2               = _mm_mul_ps(d,d);
 673             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 674
 675             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 676
 677             /* Evaluate switch function */
 678             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 679             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
 680             velec            = _mm_mul_ps(velec,sw);
 681             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 682
 683             /* Update potential sum for this i atom from the interaction with this j atom. */
 684             velec            = _mm_and_ps(velec,cutoff_mask);
 685             velec            = _mm_andnot_ps(dummy_mask,velec);
 686             velecsum         = _mm_add_ps(velecsum,velec);
 687
 688             fscal            = felec;
 689
 690             fscal            = _mm_and_ps(fscal,cutoff_mask);
 691
 692             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 693
 694              /* Update vectorial force */
 695             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 696             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 697             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 698
 699             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 700             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 701             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 702
 703             }
 704
 705             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 706             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 707             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 708             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 709
 710             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 711
 712             /* Inner loop uses 180 flops */
 713         }
 714
 715         /* End of innermost loop */
 716
 717         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 718                                               f+i_coord_offset,fshift+i_shift_offset);
 719
 720         ggid                        = gid[iidx];
 721         /* Update potential energies */
 722         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 723         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 724
 725         /* Increment number of inner iterations */
 726         inneriter                  += j_index_end - j_index_start;
 727
 728         /* Outer loop uses 20 flops */
 729     }
 730
 731     /* Increment number of outer iterations */
 732     outeriter        += nri;
 733
 734     /* Update outer/inner flops */
 735
 736     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*180);
 737 }
 738 /*
 739  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_avx_128_fma_single
 740  * Electrostatics interaction: Ewald
 741  * VdW interaction:            LennardJones
 742  * Geometry:                   Water3-Particle
 743  * Calculate force/pot:        Force
 744  */
 745 void
 746 nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_avx_128_fma_single
 747                     (t_nblist * gmx_restrict                nlist,
 748                      rvec * gmx_restrict                    xx,
 749                      rvec * gmx_restrict                    ff,
 750                      t_forcerec * gmx_restrict              fr,
 751                      t_mdatoms * gmx_restrict               mdatoms,
 752                      nb_kernel_data_t * gmx_restrict        kernel_data,
 753                      t_nrnb * gmx_restrict                  nrnb)
 754 {
 755     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 756      * just 0 for non-waters.
 757      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 758      * jnr indices corresponding to data put in the four positions in the SIMD register.
 759      */
 760     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 761     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 762     int              jnrA,jnrB,jnrC,jnrD;
 763     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 764     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 765     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 766     real             rcutoff_scalar;
 767     real             *shiftvec,*fshift,*x,*f;
 768     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 769     real             scratch[4*DIM];
 770     __m128           fscal,rcutoff,rcutoff2,jidxall;
 771     int              vdwioffset0;
 772     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 773     int              vdwioffset1;
 774     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 775     int              vdwioffset2;
 776     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 777     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 778     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 779     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 780     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 781     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 782     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 783     real             *charge;
 784     int              nvdwtype;
 785     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 786     int              *vdwtype;
 787     real             *vdwparam;
 788     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 789     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 790     __m128i          ewitab;
 791     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 792     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 793     real             *ewtab;
 794     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 795     real             rswitch_scalar,d_scalar;
 796     __m128           dummy_mask,cutoff_mask;
 797     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 798     __m128           one     = _mm_set1_ps(1.0);
 799     __m128           two     = _mm_set1_ps(2.0);
 800     x                = xx[0];
 801     f                = ff[0];
 802
 803     nri              = nlist->nri;
 804     iinr             = nlist->iinr;
 805     jindex           = nlist->jindex;
 806     jjnr             = nlist->jjnr;
 807     shiftidx         = nlist->shift;
 808     gid              = nlist->gid;
 809     shiftvec         = fr->shift_vec[0];
 810     fshift           = fr->fshift[0];
 811     facel            = _mm_set1_ps(fr->epsfac);
 812     charge           = mdatoms->chargeA;
 813     nvdwtype         = fr->ntype;
 814     vdwparam         = fr->nbfp;
 815     vdwtype          = mdatoms->typeA;
 816
 817     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 818     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
 819     beta2            = _mm_mul_ps(beta,beta);
 820     beta3            = _mm_mul_ps(beta,beta2);
 821     ewtab            = fr->ic->tabq_coul_FDV0;
 822     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 823     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 824
 825     /* Setup water-specific parameters */
 826     inr              = nlist->iinr[0];
 827     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 828     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 829     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 830     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 831
 832     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 833     rcutoff_scalar   = fr->rcoulomb;
 834     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 835     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 836
 837     rswitch_scalar   = fr->rcoulomb_switch;
 838     rswitch          = _mm_set1_ps(rswitch_scalar);
 839     /* Setup switch parameters */
 840     d_scalar         = rcutoff_scalar-rswitch_scalar;
 841     d                = _mm_set1_ps(d_scalar);
 842     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 843     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 844     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 845     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 846     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 847     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 848
 849     /* Avoid stupid compiler warnings */
 850     jnrA = jnrB = jnrC = jnrD = 0;
 851     j_coord_offsetA = 0;
 852     j_coord_offsetB = 0;
 853     j_coord_offsetC = 0;
 854     j_coord_offsetD = 0;
 855
 856     outeriter        = 0;
 857     inneriter        = 0;
 858
 859     for(iidx=0;iidx<4*DIM;iidx++)
 860     {
 861         scratch[iidx] = 0.0;
 862     }
 863
 864     /* Start outer loop over neighborlists */
 865     for(iidx=0; iidx<nri; iidx++)
 866     {
 867         /* Load shift vector for this list */
 868         i_shift_offset   = DIM*shiftidx[iidx];
 869
 870         /* Load limits for loop over neighbors */
 871         j_index_start    = jindex[iidx];
 872         j_index_end      = jindex[iidx+1];
 873
 874         /* Get outer coordinate index */
 875         inr              = iinr[iidx];
 876         i_coord_offset   = DIM*inr;
 877
 878         /* Load i particle coords and add shift vector */
 879         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 880                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 881
 882         fix0             = _mm_setzero_ps();
 883         fiy0             = _mm_setzero_ps();
 884         fiz0             = _mm_setzero_ps();
 885         fix1             = _mm_setzero_ps();
 886         fiy1             = _mm_setzero_ps();
 887         fiz1             = _mm_setzero_ps();
 888         fix2             = _mm_setzero_ps();
 889         fiy2             = _mm_setzero_ps();
 890         fiz2             = _mm_setzero_ps();
 891
 892         /* Start inner kernel loop */
 893         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 894         {
 895
 896             /* Get j neighbor index, and coordinate index */
 897             jnrA             = jjnr[jidx];
 898             jnrB             = jjnr[jidx+1];
 899             jnrC             = jjnr[jidx+2];
 900             jnrD             = jjnr[jidx+3];
 901             j_coord_offsetA  = DIM*jnrA;
 902             j_coord_offsetB  = DIM*jnrB;
 903             j_coord_offsetC  = DIM*jnrC;
 904             j_coord_offsetD  = DIM*jnrD;
 905
 906             /* load j atom coordinates */
 907             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 908                                               x+j_coord_offsetC,x+j_coord_offsetD,
 909                                               &jx0,&jy0,&jz0);
 910
 911             /* Calculate displacement vector */
 912             dx00             = _mm_sub_ps(ix0,jx0);
 913             dy00             = _mm_sub_ps(iy0,jy0);
 914             dz00             = _mm_sub_ps(iz0,jz0);
 915             dx10             = _mm_sub_ps(ix1,jx0);
 916             dy10             = _mm_sub_ps(iy1,jy0);
 917             dz10             = _mm_sub_ps(iz1,jz0);
 918             dx20             = _mm_sub_ps(ix2,jx0);
 919             dy20             = _mm_sub_ps(iy2,jy0);
 920             dz20             = _mm_sub_ps(iz2,jz0);
 921
 922             /* Calculate squared distance and things based on it */
 923             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 924             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 925             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 926
 927             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 928             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 929             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 930
 931             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 932             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 933             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 934
 935             /* Load parameters for j particles */
 936             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 937                                                               charge+jnrC+0,charge+jnrD+0);
 938             vdwjidx0A        = 2*vdwtype[jnrA+0];
 939             vdwjidx0B        = 2*vdwtype[jnrB+0];
 940             vdwjidx0C        = 2*vdwtype[jnrC+0];
 941             vdwjidx0D        = 2*vdwtype[jnrD+0];
 942
 943             fjx0             = _mm_setzero_ps();
 944             fjy0             = _mm_setzero_ps();
 945             fjz0             = _mm_setzero_ps();
 946
 947             /**************************
 948              * CALCULATE INTERACTIONS *
 949              **************************/
 950
 951             if (gmx_mm_any_lt(rsq00,rcutoff2))
 952             {
 953
 954             r00              = _mm_mul_ps(rsq00,rinv00);
 955
 956             /* Compute parameters for interactions between i and j atoms */
 957             qq00             = _mm_mul_ps(iq0,jq0);
 958             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 959                                          vdwparam+vdwioffset0+vdwjidx0B,
 960                                          vdwparam+vdwioffset0+vdwjidx0C,
 961                                          vdwparam+vdwioffset0+vdwjidx0D,
 962                                          &c6_00,&c12_00);
 963
 964             /* EWALD ELECTROSTATICS */
 965
 966             /* Analytical PME correction */
 967             zeta2            = _mm_mul_ps(beta2,rsq00);
 968             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 969             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 970             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 971             felec            = _mm_mul_ps(qq00,felec);
 972             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 973             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 974             velec            = _mm_mul_ps(qq00,velec);
 975
 976             /* LENNARD-JONES DISPERSION/REPULSION */
 977
 978             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 979             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 980             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 981             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 982             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 983
 984             d                = _mm_sub_ps(r00,rswitch);
 985             d                = _mm_max_ps(d,_mm_setzero_ps());
 986             d2               = _mm_mul_ps(d,d);
 987             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 988
 989             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 990
 991             /* Evaluate switch function */
 992             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 993             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 994             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 995             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 996
 997             fscal            = _mm_add_ps(felec,fvdw);
 998
 999             fscal            = _mm_and_ps(fscal,cutoff_mask);
1000
1001              /* Update vectorial force */
1002             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1003             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1004             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1005
1006             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1007             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1008             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1009
1010             }
1011
1012             /**************************
1013              * CALCULATE INTERACTIONS *
1014              **************************/
1015
1016             if (gmx_mm_any_lt(rsq10,rcutoff2))
1017             {
1018
1019             r10              = _mm_mul_ps(rsq10,rinv10);
1020
1021             /* Compute parameters for interactions between i and j atoms */
1022             qq10             = _mm_mul_ps(iq1,jq0);
1023
1024             /* EWALD ELECTROSTATICS */
1025
1026             /* Analytical PME correction */
1027             zeta2            = _mm_mul_ps(beta2,rsq10);
1028             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1029             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1030             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1031             felec            = _mm_mul_ps(qq10,felec);
1032             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1033             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
1034             velec            = _mm_mul_ps(qq10,velec);
1035
1036             d                = _mm_sub_ps(r10,rswitch);
1037             d                = _mm_max_ps(d,_mm_setzero_ps());
1038             d2               = _mm_mul_ps(d,d);
1039             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1040
1041             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1042
1043             /* Evaluate switch function */
1044             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1045             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
1046             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1047
1048             fscal            = felec;
1049
1050             fscal            = _mm_and_ps(fscal,cutoff_mask);
1051
1052              /* Update vectorial force */
1053             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1054             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1055             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1056
1057             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1058             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1059             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1060
1061             }
1062
1063             /**************************
1064              * CALCULATE INTERACTIONS *
1065              **************************/
1066
1067             if (gmx_mm_any_lt(rsq20,rcutoff2))
1068             {
1069
1070             r20              = _mm_mul_ps(rsq20,rinv20);
1071
1072             /* Compute parameters for interactions between i and j atoms */
1073             qq20             = _mm_mul_ps(iq2,jq0);
1074
1075             /* EWALD ELECTROSTATICS */
1076
1077             /* Analytical PME correction */
1078             zeta2            = _mm_mul_ps(beta2,rsq20);
1079             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1080             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1081             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1082             felec            = _mm_mul_ps(qq20,felec);
1083             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1084             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
1085             velec            = _mm_mul_ps(qq20,velec);
1086
1087             d                = _mm_sub_ps(r20,rswitch);
1088             d                = _mm_max_ps(d,_mm_setzero_ps());
1089             d2               = _mm_mul_ps(d,d);
1090             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1091
1092             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1093
1094             /* Evaluate switch function */
1095             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1096             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1097             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1098
1099             fscal            = felec;
1100
1101             fscal            = _mm_and_ps(fscal,cutoff_mask);
1102
1103              /* Update vectorial force */
1104             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1105             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1106             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1107
1108             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1109             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1110             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1111
1112             }
1113
1114             fjptrA             = f+j_coord_offsetA;
1115             fjptrB             = f+j_coord_offsetB;
1116             fjptrC             = f+j_coord_offsetC;
1117             fjptrD             = f+j_coord_offsetD;
1118
1119             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1120
1121             /* Inner loop uses 165 flops */
1122         }
1123
1124         if(jidx<j_index_end)
1125         {
1126
1127             /* Get j neighbor index, and coordinate index */
1128             jnrlistA         = jjnr[jidx];
1129             jnrlistB         = jjnr[jidx+1];
1130             jnrlistC         = jjnr[jidx+2];
1131             jnrlistD         = jjnr[jidx+3];
1132             /* Sign of each element will be negative for non-real atoms.
1133              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1134              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1135              */
1136             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1137             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1138             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1139             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1140             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1141             j_coord_offsetA  = DIM*jnrA;
1142             j_coord_offsetB  = DIM*jnrB;
1143             j_coord_offsetC  = DIM*jnrC;
1144             j_coord_offsetD  = DIM*jnrD;
1145
1146             /* load j atom coordinates */
1147             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1148                                               x+j_coord_offsetC,x+j_coord_offsetD,
1149                                               &jx0,&jy0,&jz0);
1150
1151             /* Calculate displacement vector */
1152             dx00             = _mm_sub_ps(ix0,jx0);
1153             dy00             = _mm_sub_ps(iy0,jy0);
1154             dz00             = _mm_sub_ps(iz0,jz0);
1155             dx10             = _mm_sub_ps(ix1,jx0);
1156             dy10             = _mm_sub_ps(iy1,jy0);
1157             dz10             = _mm_sub_ps(iz1,jz0);
1158             dx20             = _mm_sub_ps(ix2,jx0);
1159             dy20             = _mm_sub_ps(iy2,jy0);
1160             dz20             = _mm_sub_ps(iz2,jz0);
1161
1162             /* Calculate squared distance and things based on it */
1163             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1164             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1165             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1166
1167             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1168             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1169             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1170
1171             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1172             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1173             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1174
1175             /* Load parameters for j particles */
1176             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1177                                                               charge+jnrC+0,charge+jnrD+0);
1178             vdwjidx0A        = 2*vdwtype[jnrA+0];
1179             vdwjidx0B        = 2*vdwtype[jnrB+0];
1180             vdwjidx0C        = 2*vdwtype[jnrC+0];
1181             vdwjidx0D        = 2*vdwtype[jnrD+0];
1182
1183             fjx0             = _mm_setzero_ps();
1184             fjy0             = _mm_setzero_ps();
1185             fjz0             = _mm_setzero_ps();
1186
1187             /**************************
1188              * CALCULATE INTERACTIONS *
1189              **************************/
1190
1191             if (gmx_mm_any_lt(rsq00,rcutoff2))
1192             {
1193
1194             r00              = _mm_mul_ps(rsq00,rinv00);
1195             r00              = _mm_andnot_ps(dummy_mask,r00);
1196
1197             /* Compute parameters for interactions between i and j atoms */
1198             qq00             = _mm_mul_ps(iq0,jq0);
1199             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1200                                          vdwparam+vdwioffset0+vdwjidx0B,
1201                                          vdwparam+vdwioffset0+vdwjidx0C,
1202                                          vdwparam+vdwioffset0+vdwjidx0D,
1203                                          &c6_00,&c12_00);
1204
1205             /* EWALD ELECTROSTATICS */
1206
1207             /* Analytical PME correction */
1208             zeta2            = _mm_mul_ps(beta2,rsq00);
1209             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
1210             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1211             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1212             felec            = _mm_mul_ps(qq00,felec);
1213             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1214             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
1215             velec            = _mm_mul_ps(qq00,velec);
1216
1217             /* LENNARD-JONES DISPERSION/REPULSION */
1218
1219             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1220             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1221             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1222             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
1223             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1224
1225             d                = _mm_sub_ps(r00,rswitch);
1226             d                = _mm_max_ps(d,_mm_setzero_ps());
1227             d2               = _mm_mul_ps(d,d);
1228             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1229
1230             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1231
1232             /* Evaluate switch function */
1233             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1234             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
1235             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1236             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1237
1238             fscal            = _mm_add_ps(felec,fvdw);
1239
1240             fscal            = _mm_and_ps(fscal,cutoff_mask);
1241
1242             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1243
1244              /* Update vectorial force */
1245             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1246             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1247             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1248
1249             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1250             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1251             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1252
1253             }
1254
1255             /**************************
1256              * CALCULATE INTERACTIONS *
1257              **************************/
1258
1259             if (gmx_mm_any_lt(rsq10,rcutoff2))
1260             {
1261
1262             r10              = _mm_mul_ps(rsq10,rinv10);
1263             r10              = _mm_andnot_ps(dummy_mask,r10);
1264
1265             /* Compute parameters for interactions between i and j atoms */
1266             qq10             = _mm_mul_ps(iq1,jq0);
1267
1268             /* EWALD ELECTROSTATICS */
1269
1270             /* Analytical PME correction */
1271             zeta2            = _mm_mul_ps(beta2,rsq10);
1272             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1273             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1274             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1275             felec            = _mm_mul_ps(qq10,felec);
1276             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1277             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
1278             velec            = _mm_mul_ps(qq10,velec);
1279
1280             d                = _mm_sub_ps(r10,rswitch);
1281             d                = _mm_max_ps(d,_mm_setzero_ps());
1282             d2               = _mm_mul_ps(d,d);
1283             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1284
1285             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1286
1287             /* Evaluate switch function */
1288             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1289             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
1290             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1291
1292             fscal            = felec;
1293
1294             fscal            = _mm_and_ps(fscal,cutoff_mask);
1295
1296             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1297
1298              /* Update vectorial force */
1299             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1300             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1301             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1302
1303             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1304             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1305             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1306
1307             }
1308
1309             /**************************
1310              * CALCULATE INTERACTIONS *
1311              **************************/
1312
1313             if (gmx_mm_any_lt(rsq20,rcutoff2))
1314             {
1315
1316             r20              = _mm_mul_ps(rsq20,rinv20);
1317             r20              = _mm_andnot_ps(dummy_mask,r20);
1318
1319             /* Compute parameters for interactions between i and j atoms */
1320             qq20             = _mm_mul_ps(iq2,jq0);
1321
1322             /* EWALD ELECTROSTATICS */
1323
1324             /* Analytical PME correction */
1325             zeta2            = _mm_mul_ps(beta2,rsq20);
1326             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1327             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1328             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1329             felec            = _mm_mul_ps(qq20,felec);
1330             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1331             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
1332             velec            = _mm_mul_ps(qq20,velec);
1333
1334             d                = _mm_sub_ps(r20,rswitch);
1335             d                = _mm_max_ps(d,_mm_setzero_ps());
1336             d2               = _mm_mul_ps(d,d);
1337             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1338
1339             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1340
1341             /* Evaluate switch function */
1342             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1343             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1344             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1345
1346             fscal            = felec;
1347
1348             fscal            = _mm_and_ps(fscal,cutoff_mask);
1349
1350             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1351
1352              /* Update vectorial force */
1353             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1354             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1355             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1356
1357             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1358             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1359             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1360
1361             }
1362
1363             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1364             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1365             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1366             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1367
1368             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1369
1370             /* Inner loop uses 168 flops */
1371         }
1372
1373         /* End of innermost loop */
1374
1375         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1376                                               f+i_coord_offset,fshift+i_shift_offset);
1377
1378         /* Increment number of inner iterations */
1379         inneriter                  += j_index_end - j_index_start;
1380
1381         /* Outer loop uses 18 flops */
1382     }
1383
1384     /* Increment number of outer iterations */
1385     outeriter        += nri;
1386
1387     /* Update outer/inner flops */
1388
1389     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*168);
1390 }