src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_VF_avx_128_fma_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwioffset3;
  90     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  97     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 104     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 105     __m128           c6grid_00;
 106     __m128           c6grid_10;
 107     __m128           c6grid_20;
 108     __m128           c6grid_30;
 109     real             *vdwgridparam;
 110     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 111     __m128           one_half = _mm_set1_ps(0.5);
 112     __m128           minus_one = _mm_set1_ps(-1.0);
 113     __m128i          ewitab;
 114     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 115     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 116     real             *ewtab;
 117     __m128           dummy_mask,cutoff_mask;
 118     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 119     __m128           one     = _mm_set1_ps(1.0);
 120     __m128           two     = _mm_set1_ps(2.0);
 121     x                = xx[0];
 122     f                = ff[0];
 123
 124     nri              = nlist->nri;
 125     iinr             = nlist->iinr;
 126     jindex           = nlist->jindex;
 127     jjnr             = nlist->jjnr;
 128     shiftidx         = nlist->shift;
 129     gid              = nlist->gid;
 130     shiftvec         = fr->shift_vec[0];
 131     fshift           = fr->fshift[0];
 132     facel            = _mm_set1_ps(fr->epsfac);
 133     charge           = mdatoms->chargeA;
 134     nvdwtype         = fr->ntype;
 135     vdwparam         = fr->nbfp;
 136     vdwtype          = mdatoms->typeA;
 137     vdwgridparam     = fr->ljpme_c6grid;
 138     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 139     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 140     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 141
 142     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 143     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 144     beta2            = _mm_mul_ps(beta,beta);
 145     beta3            = _mm_mul_ps(beta,beta2);
 146     ewtab            = fr->ic->tabq_coul_FDV0;
 147     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 148     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 149
 150     /* Setup water-specific parameters */
 151     inr              = nlist->iinr[0];
 152     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 153     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 154     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 155     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 156
 157     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 158     rcutoff_scalar   = fr->rcoulomb;
 159     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 160     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 161
 162     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 163     rvdw             = _mm_set1_ps(fr->rvdw);
 164
 165     /* Avoid stupid compiler warnings */
 166     jnrA = jnrB = jnrC = jnrD = 0;
 167     j_coord_offsetA = 0;
 168     j_coord_offsetB = 0;
 169     j_coord_offsetC = 0;
 170     j_coord_offsetD = 0;
 171
 172     outeriter        = 0;
 173     inneriter        = 0;
 174
 175     for(iidx=0;iidx<4*DIM;iidx++)
 176     {
 177         scratch[iidx] = 0.0;
 178     }
 179
 180     /* Start outer loop over neighborlists */
 181     for(iidx=0; iidx<nri; iidx++)
 182     {
 183         /* Load shift vector for this list */
 184         i_shift_offset   = DIM*shiftidx[iidx];
 185
 186         /* Load limits for loop over neighbors */
 187         j_index_start    = jindex[iidx];
 188         j_index_end      = jindex[iidx+1];
 189
 190         /* Get outer coordinate index */
 191         inr              = iinr[iidx];
 192         i_coord_offset   = DIM*inr;
 193
 194         /* Load i particle coords and add shift vector */
 195         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 196                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 197
 198         fix0             = _mm_setzero_ps();
 199         fiy0             = _mm_setzero_ps();
 200         fiz0             = _mm_setzero_ps();
 201         fix1             = _mm_setzero_ps();
 202         fiy1             = _mm_setzero_ps();
 203         fiz1             = _mm_setzero_ps();
 204         fix2             = _mm_setzero_ps();
 205         fiy2             = _mm_setzero_ps();
 206         fiz2             = _mm_setzero_ps();
 207         fix3             = _mm_setzero_ps();
 208         fiy3             = _mm_setzero_ps();
 209         fiz3             = _mm_setzero_ps();
 210
 211         /* Reset potential sums */
 212         velecsum         = _mm_setzero_ps();
 213         vvdwsum          = _mm_setzero_ps();
 214
 215         /* Start inner kernel loop */
 216         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 217         {
 218
 219             /* Get j neighbor index, and coordinate index */
 220             jnrA             = jjnr[jidx];
 221             jnrB             = jjnr[jidx+1];
 222             jnrC             = jjnr[jidx+2];
 223             jnrD             = jjnr[jidx+3];
 224             j_coord_offsetA  = DIM*jnrA;
 225             j_coord_offsetB  = DIM*jnrB;
 226             j_coord_offsetC  = DIM*jnrC;
 227             j_coord_offsetD  = DIM*jnrD;
 228
 229             /* load j atom coordinates */
 230             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 231                                               x+j_coord_offsetC,x+j_coord_offsetD,
 232                                               &jx0,&jy0,&jz0);
 233
 234             /* Calculate displacement vector */
 235             dx00             = _mm_sub_ps(ix0,jx0);
 236             dy00             = _mm_sub_ps(iy0,jy0);
 237             dz00             = _mm_sub_ps(iz0,jz0);
 238             dx10             = _mm_sub_ps(ix1,jx0);
 239             dy10             = _mm_sub_ps(iy1,jy0);
 240             dz10             = _mm_sub_ps(iz1,jz0);
 241             dx20             = _mm_sub_ps(ix2,jx0);
 242             dy20             = _mm_sub_ps(iy2,jy0);
 243             dz20             = _mm_sub_ps(iz2,jz0);
 244             dx30             = _mm_sub_ps(ix3,jx0);
 245             dy30             = _mm_sub_ps(iy3,jy0);
 246             dz30             = _mm_sub_ps(iz3,jz0);
 247
 248             /* Calculate squared distance and things based on it */
 249             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 250             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 251             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 252             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 253
 254             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 255             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 256             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 257             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 258
 259             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 260             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 261             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 262             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 263
 264             /* Load parameters for j particles */
 265             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 266                                                               charge+jnrC+0,charge+jnrD+0);
 267             vdwjidx0A        = 2*vdwtype[jnrA+0];
 268             vdwjidx0B        = 2*vdwtype[jnrB+0];
 269             vdwjidx0C        = 2*vdwtype[jnrC+0];
 270             vdwjidx0D        = 2*vdwtype[jnrD+0];
 271
 272             fjx0             = _mm_setzero_ps();
 273             fjy0             = _mm_setzero_ps();
 274             fjz0             = _mm_setzero_ps();
 275
 276             /**************************
 277              * CALCULATE INTERACTIONS *
 278              **************************/
 279
 280             if (gmx_mm_any_lt(rsq00,rcutoff2))
 281             {
 282
 283             r00              = _mm_mul_ps(rsq00,rinv00);
 284
 285             /* Compute parameters for interactions between i and j atoms */
 286             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 287                                          vdwparam+vdwioffset0+vdwjidx0B,
 288                                          vdwparam+vdwioffset0+vdwjidx0C,
 289                                          vdwparam+vdwioffset0+vdwjidx0D,
 290                                          &c6_00,&c12_00);
 291
 292             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 293                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 294                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 295                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 296
 297             /* Analytical LJ-PME */
 298             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 299             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 300             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 301             exponent         = gmx_simd_exp_r(ewcljrsq);
 302             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 303             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
 304             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 305             vvdw6            = _mm_mul_ps(_mm_macc_ps(-c6grid_00,_mm_sub_ps(one,poly),c6_00),rinvsix);
 306             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 307             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 308                                           _mm_mul_ps(_mm_sub_ps(vvdw6,_mm_macc_ps(c6grid_00,sh_lj_ewald,_mm_mul_ps(c6_00,sh_vdw_invrcut6))),one_sixth));
 309             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 310             fvdw             = _mm_mul_ps(_mm_add_ps(vvdw12,_mm_msub_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6),vvdw6)),rinvsq00);
 311
 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,_mm_sub_ps(rinv10,sh_ewald));
 355             velec            = _mm_mul_ps(qq10,velec);
 356
 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,_mm_sub_ps(rinv20,sh_ewald));
 400             velec            = _mm_mul_ps(qq20,velec);
 401
 402             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 403
 404             /* Update potential sum for this i atom from the interaction with this j atom. */
 405             velec            = _mm_and_ps(velec,cutoff_mask);
 406             velecsum         = _mm_add_ps(velecsum,velec);
 407
 408             fscal            = felec;
 409
 410             fscal            = _mm_and_ps(fscal,cutoff_mask);
 411
 412              /* Update vectorial force */
 413             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 414             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 415             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 416
 417             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 418             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 419             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 420
 421             }
 422
 423             /**************************
 424              * CALCULATE INTERACTIONS *
 425              **************************/
 426
 427             if (gmx_mm_any_lt(rsq30,rcutoff2))
 428             {
 429
 430             r30              = _mm_mul_ps(rsq30,rinv30);
 431
 432             /* Compute parameters for interactions between i and j atoms */
 433             qq30             = _mm_mul_ps(iq3,jq0);
 434
 435             /* EWALD ELECTROSTATICS */
 436
 437             /* Analytical PME correction */
 438             zeta2            = _mm_mul_ps(beta2,rsq30);
 439             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 440             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 441             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 442             felec            = _mm_mul_ps(qq30,felec);
 443             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 444             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv30,sh_ewald));
 445             velec            = _mm_mul_ps(qq30,velec);
 446
 447             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 448
 449             /* Update potential sum for this i atom from the interaction with this j atom. */
 450             velec            = _mm_and_ps(velec,cutoff_mask);
 451             velecsum         = _mm_add_ps(velecsum,velec);
 452
 453             fscal            = felec;
 454
 455             fscal            = _mm_and_ps(fscal,cutoff_mask);
 456
 457              /* Update vectorial force */
 458             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 459             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 460             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 461
 462             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 463             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 464             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 465
 466             }
 467
 468             fjptrA             = f+j_coord_offsetA;
 469             fjptrB             = f+j_coord_offsetB;
 470             fjptrC             = f+j_coord_offsetC;
 471             fjptrD             = f+j_coord_offsetD;
 472
 473             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 474
 475             /* Inner loop uses 158 flops */
 476         }
 477
 478         if(jidx<j_index_end)
 479         {
 480
 481             /* Get j neighbor index, and coordinate index */
 482             jnrlistA         = jjnr[jidx];
 483             jnrlistB         = jjnr[jidx+1];
 484             jnrlistC         = jjnr[jidx+2];
 485             jnrlistD         = jjnr[jidx+3];
 486             /* Sign of each element will be negative for non-real atoms.
 487              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 488              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 489              */
 490             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 491             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 492             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 493             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 494             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 495             j_coord_offsetA  = DIM*jnrA;
 496             j_coord_offsetB  = DIM*jnrB;
 497             j_coord_offsetC  = DIM*jnrC;
 498             j_coord_offsetD  = DIM*jnrD;
 499
 500             /* load j atom coordinates */
 501             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 502                                               x+j_coord_offsetC,x+j_coord_offsetD,
 503                                               &jx0,&jy0,&jz0);
 504
 505             /* Calculate displacement vector */
 506             dx00             = _mm_sub_ps(ix0,jx0);
 507             dy00             = _mm_sub_ps(iy0,jy0);
 508             dz00             = _mm_sub_ps(iz0,jz0);
 509             dx10             = _mm_sub_ps(ix1,jx0);
 510             dy10             = _mm_sub_ps(iy1,jy0);
 511             dz10             = _mm_sub_ps(iz1,jz0);
 512             dx20             = _mm_sub_ps(ix2,jx0);
 513             dy20             = _mm_sub_ps(iy2,jy0);
 514             dz20             = _mm_sub_ps(iz2,jz0);
 515             dx30             = _mm_sub_ps(ix3,jx0);
 516             dy30             = _mm_sub_ps(iy3,jy0);
 517             dz30             = _mm_sub_ps(iz3,jz0);
 518
 519             /* Calculate squared distance and things based on it */
 520             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 521             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 522             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 523             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 524
 525             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 526             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 527             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 528             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 529
 530             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 531             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 532             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 533             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 534
 535             /* Load parameters for j particles */
 536             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 537                                                               charge+jnrC+0,charge+jnrD+0);
 538             vdwjidx0A        = 2*vdwtype[jnrA+0];
 539             vdwjidx0B        = 2*vdwtype[jnrB+0];
 540             vdwjidx0C        = 2*vdwtype[jnrC+0];
 541             vdwjidx0D        = 2*vdwtype[jnrD+0];
 542
 543             fjx0             = _mm_setzero_ps();
 544             fjy0             = _mm_setzero_ps();
 545             fjz0             = _mm_setzero_ps();
 546
 547             /**************************
 548              * CALCULATE INTERACTIONS *
 549              **************************/
 550
 551             if (gmx_mm_any_lt(rsq00,rcutoff2))
 552             {
 553
 554             r00              = _mm_mul_ps(rsq00,rinv00);
 555             r00              = _mm_andnot_ps(dummy_mask,r00);
 556
 557             /* Compute parameters for interactions between i and j atoms */
 558             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 559                                          vdwparam+vdwioffset0+vdwjidx0B,
 560                                          vdwparam+vdwioffset0+vdwjidx0C,
 561                                          vdwparam+vdwioffset0+vdwjidx0D,
 562                                          &c6_00,&c12_00);
 563
 564             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 565                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 566                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 567                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 568
 569             /* Analytical LJ-PME */
 570             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 571             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 572             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 573             exponent         = gmx_simd_exp_r(ewcljrsq);
 574             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 575             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
 576             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 577             vvdw6            = _mm_mul_ps(_mm_macc_ps(-c6grid_00,_mm_sub_ps(one,poly),c6_00),rinvsix);
 578             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 579             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 580                                           _mm_mul_ps(_mm_sub_ps(vvdw6,_mm_macc_ps(c6grid_00,sh_lj_ewald,_mm_mul_ps(c6_00,sh_vdw_invrcut6))),one_sixth));
 581             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 582             fvdw             = _mm_mul_ps(_mm_add_ps(vvdw12,_mm_msub_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6),vvdw6)),rinvsq00);
 583
 584             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 585
 586             /* Update potential sum for this i atom from the interaction with this j atom. */
 587             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 588             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 589             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 590
 591             fscal            = fvdw;
 592
 593             fscal            = _mm_and_ps(fscal,cutoff_mask);
 594
 595             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 596
 597              /* Update vectorial force */
 598             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 599             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 600             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 601
 602             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 603             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 604             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 605
 606             }
 607
 608             /**************************
 609              * CALCULATE INTERACTIONS *
 610              **************************/
 611
 612             if (gmx_mm_any_lt(rsq10,rcutoff2))
 613             {
 614
 615             r10              = _mm_mul_ps(rsq10,rinv10);
 616             r10              = _mm_andnot_ps(dummy_mask,r10);
 617
 618             /* Compute parameters for interactions between i and j atoms */
 619             qq10             = _mm_mul_ps(iq1,jq0);
 620
 621             /* EWALD ELECTROSTATICS */
 622
 623             /* Analytical PME correction */
 624             zeta2            = _mm_mul_ps(beta2,rsq10);
 625             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 626             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 627             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 628             felec            = _mm_mul_ps(qq10,felec);
 629             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 630             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv10,sh_ewald));
 631             velec            = _mm_mul_ps(qq10,velec);
 632
 633             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 634
 635             /* Update potential sum for this i atom from the interaction with this j atom. */
 636             velec            = _mm_and_ps(velec,cutoff_mask);
 637             velec            = _mm_andnot_ps(dummy_mask,velec);
 638             velecsum         = _mm_add_ps(velecsum,velec);
 639
 640             fscal            = felec;
 641
 642             fscal            = _mm_and_ps(fscal,cutoff_mask);
 643
 644             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 645
 646              /* Update vectorial force */
 647             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 648             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 649             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 650
 651             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 652             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 653             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 654
 655             }
 656
 657             /**************************
 658              * CALCULATE INTERACTIONS *
 659              **************************/
 660
 661             if (gmx_mm_any_lt(rsq20,rcutoff2))
 662             {
 663
 664             r20              = _mm_mul_ps(rsq20,rinv20);
 665             r20              = _mm_andnot_ps(dummy_mask,r20);
 666
 667             /* Compute parameters for interactions between i and j atoms */
 668             qq20             = _mm_mul_ps(iq2,jq0);
 669
 670             /* EWALD ELECTROSTATICS */
 671
 672             /* Analytical PME correction */
 673             zeta2            = _mm_mul_ps(beta2,rsq20);
 674             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 675             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 676             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 677             felec            = _mm_mul_ps(qq20,felec);
 678             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 679             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv20,sh_ewald));
 680             velec            = _mm_mul_ps(qq20,velec);
 681
 682             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 683
 684             /* Update potential sum for this i atom from the interaction with this j atom. */
 685             velec            = _mm_and_ps(velec,cutoff_mask);
 686             velec            = _mm_andnot_ps(dummy_mask,velec);
 687             velecsum         = _mm_add_ps(velecsum,velec);
 688
 689             fscal            = felec;
 690
 691             fscal            = _mm_and_ps(fscal,cutoff_mask);
 692
 693             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 694
 695              /* Update vectorial force */
 696             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 697             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 698             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 699
 700             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 701             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 702             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 703
 704             }
 705
 706             /**************************
 707              * CALCULATE INTERACTIONS *
 708              **************************/
 709
 710             if (gmx_mm_any_lt(rsq30,rcutoff2))
 711             {
 712
 713             r30              = _mm_mul_ps(rsq30,rinv30);
 714             r30              = _mm_andnot_ps(dummy_mask,r30);
 715
 716             /* Compute parameters for interactions between i and j atoms */
 717             qq30             = _mm_mul_ps(iq3,jq0);
 718
 719             /* EWALD ELECTROSTATICS */
 720
 721             /* Analytical PME correction */
 722             zeta2            = _mm_mul_ps(beta2,rsq30);
 723             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 724             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 725             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 726             felec            = _mm_mul_ps(qq30,felec);
 727             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 728             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv30,sh_ewald));
 729             velec            = _mm_mul_ps(qq30,velec);
 730
 731             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 732
 733             /* Update potential sum for this i atom from the interaction with this j atom. */
 734             velec            = _mm_and_ps(velec,cutoff_mask);
 735             velec            = _mm_andnot_ps(dummy_mask,velec);
 736             velecsum         = _mm_add_ps(velecsum,velec);
 737
 738             fscal            = felec;
 739
 740             fscal            = _mm_and_ps(fscal,cutoff_mask);
 741
 742             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 743
 744              /* Update vectorial force */
 745             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 746             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 747             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 748
 749             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 750             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 751             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 752
 753             }
 754
 755             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 756             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 757             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 758             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 759
 760             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 761
 762             /* Inner loop uses 162 flops */
 763         }
 764
 765         /* End of innermost loop */
 766
 767         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 768                                               f+i_coord_offset,fshift+i_shift_offset);
 769
 770         ggid                        = gid[iidx];
 771         /* Update potential energies */
 772         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 773         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 774
 775         /* Increment number of inner iterations */
 776         inneriter                  += j_index_end - j_index_start;
 777
 778         /* Outer loop uses 26 flops */
 779     }
 780
 781     /* Increment number of outer iterations */
 782     outeriter        += nri;
 783
 784     /* Update outer/inner flops */
 785
 786     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*162);
 787 }
 788 /*
 789  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_avx_128_fma_single
 790  * Electrostatics interaction: Ewald
 791  * VdW interaction:            LJEwald
 792  * Geometry:                   Water4-Particle
 793  * Calculate force/pot:        Force
 794  */
 795 void
 796 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_avx_128_fma_single
 797                     (t_nblist                    * gmx_restrict       nlist,
 798                      rvec                        * gmx_restrict          xx,
 799                      rvec                        * gmx_restrict          ff,
 800                      t_forcerec                  * gmx_restrict          fr,
 801                      t_mdatoms                   * gmx_restrict     mdatoms,
 802                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 803                      t_nrnb                      * gmx_restrict        nrnb)
 804 {
 805     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 806      * just 0 for non-waters.
 807      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 808      * jnr indices corresponding to data put in the four positions in the SIMD register.
 809      */
 810     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 811     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 812     int              jnrA,jnrB,jnrC,jnrD;
 813     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 814     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 815     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 816     real             rcutoff_scalar;
 817     real             *shiftvec,*fshift,*x,*f;
 818     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 819     real             scratch[4*DIM];
 820     __m128           fscal,rcutoff,rcutoff2,jidxall;
 821     int              vdwioffset0;
 822     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 823     int              vdwioffset1;
 824     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 825     int              vdwioffset2;
 826     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 827     int              vdwioffset3;
 828     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 829     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 830     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 831     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 832     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 833     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 834     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 835     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 836     real             *charge;
 837     int              nvdwtype;
 838     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 839     int              *vdwtype;
 840     real             *vdwparam;
 841     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 842     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 843     __m128           c6grid_00;
 844     __m128           c6grid_10;
 845     __m128           c6grid_20;
 846     __m128           c6grid_30;
 847     real             *vdwgridparam;
 848     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 849     __m128           one_half = _mm_set1_ps(0.5);
 850     __m128           minus_one = _mm_set1_ps(-1.0);
 851     __m128i          ewitab;
 852     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 853     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 854     real             *ewtab;
 855     __m128           dummy_mask,cutoff_mask;
 856     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 857     __m128           one     = _mm_set1_ps(1.0);
 858     __m128           two     = _mm_set1_ps(2.0);
 859     x                = xx[0];
 860     f                = ff[0];
 861
 862     nri              = nlist->nri;
 863     iinr             = nlist->iinr;
 864     jindex           = nlist->jindex;
 865     jjnr             = nlist->jjnr;
 866     shiftidx         = nlist->shift;
 867     gid              = nlist->gid;
 868     shiftvec         = fr->shift_vec[0];
 869     fshift           = fr->fshift[0];
 870     facel            = _mm_set1_ps(fr->epsfac);
 871     charge           = mdatoms->chargeA;
 872     nvdwtype         = fr->ntype;
 873     vdwparam         = fr->nbfp;
 874     vdwtype          = mdatoms->typeA;
 875     vdwgridparam     = fr->ljpme_c6grid;
 876     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 877     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 878     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 879
 880     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 881     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 882     beta2            = _mm_mul_ps(beta,beta);
 883     beta3            = _mm_mul_ps(beta,beta2);
 884     ewtab            = fr->ic->tabq_coul_F;
 885     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 886     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 887
 888     /* Setup water-specific parameters */
 889     inr              = nlist->iinr[0];
 890     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 891     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 892     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 893     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 894
 895     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 896     rcutoff_scalar   = fr->rcoulomb;
 897     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 898     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 899
 900     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 901     rvdw             = _mm_set1_ps(fr->rvdw);
 902
 903     /* Avoid stupid compiler warnings */
 904     jnrA = jnrB = jnrC = jnrD = 0;
 905     j_coord_offsetA = 0;
 906     j_coord_offsetB = 0;
 907     j_coord_offsetC = 0;
 908     j_coord_offsetD = 0;
 909
 910     outeriter        = 0;
 911     inneriter        = 0;
 912
 913     for(iidx=0;iidx<4*DIM;iidx++)
 914     {
 915         scratch[iidx] = 0.0;
 916     }
 917
 918     /* Start outer loop over neighborlists */
 919     for(iidx=0; iidx<nri; iidx++)
 920     {
 921         /* Load shift vector for this list */
 922         i_shift_offset   = DIM*shiftidx[iidx];
 923
 924         /* Load limits for loop over neighbors */
 925         j_index_start    = jindex[iidx];
 926         j_index_end      = jindex[iidx+1];
 927
 928         /* Get outer coordinate index */
 929         inr              = iinr[iidx];
 930         i_coord_offset   = DIM*inr;
 931
 932         /* Load i particle coords and add shift vector */
 933         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 934                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 935
 936         fix0             = _mm_setzero_ps();
 937         fiy0             = _mm_setzero_ps();
 938         fiz0             = _mm_setzero_ps();
 939         fix1             = _mm_setzero_ps();
 940         fiy1             = _mm_setzero_ps();
 941         fiz1             = _mm_setzero_ps();
 942         fix2             = _mm_setzero_ps();
 943         fiy2             = _mm_setzero_ps();
 944         fiz2             = _mm_setzero_ps();
 945         fix3             = _mm_setzero_ps();
 946         fiy3             = _mm_setzero_ps();
 947         fiz3             = _mm_setzero_ps();
 948
 949         /* Start inner kernel loop */
 950         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 951         {
 952
 953             /* Get j neighbor index, and coordinate index */
 954             jnrA             = jjnr[jidx];
 955             jnrB             = jjnr[jidx+1];
 956             jnrC             = jjnr[jidx+2];
 957             jnrD             = jjnr[jidx+3];
 958             j_coord_offsetA  = DIM*jnrA;
 959             j_coord_offsetB  = DIM*jnrB;
 960             j_coord_offsetC  = DIM*jnrC;
 961             j_coord_offsetD  = DIM*jnrD;
 962
 963             /* load j atom coordinates */
 964             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 965                                               x+j_coord_offsetC,x+j_coord_offsetD,
 966                                               &jx0,&jy0,&jz0);
 967
 968             /* Calculate displacement vector */
 969             dx00             = _mm_sub_ps(ix0,jx0);
 970             dy00             = _mm_sub_ps(iy0,jy0);
 971             dz00             = _mm_sub_ps(iz0,jz0);
 972             dx10             = _mm_sub_ps(ix1,jx0);
 973             dy10             = _mm_sub_ps(iy1,jy0);
 974             dz10             = _mm_sub_ps(iz1,jz0);
 975             dx20             = _mm_sub_ps(ix2,jx0);
 976             dy20             = _mm_sub_ps(iy2,jy0);
 977             dz20             = _mm_sub_ps(iz2,jz0);
 978             dx30             = _mm_sub_ps(ix3,jx0);
 979             dy30             = _mm_sub_ps(iy3,jy0);
 980             dz30             = _mm_sub_ps(iz3,jz0);
 981
 982             /* Calculate squared distance and things based on it */
 983             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 984             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 985             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 986             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 987
 988             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 989             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 990             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 991             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 992
 993             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 994             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 995             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 996             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 997
 998             /* Load parameters for j particles */
 999             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1000                                                               charge+jnrC+0,charge+jnrD+0);
1001             vdwjidx0A        = 2*vdwtype[jnrA+0];
1002             vdwjidx0B        = 2*vdwtype[jnrB+0];
1003             vdwjidx0C        = 2*vdwtype[jnrC+0];
1004             vdwjidx0D        = 2*vdwtype[jnrD+0];
1005
1006             fjx0             = _mm_setzero_ps();
1007             fjy0             = _mm_setzero_ps();
1008             fjz0             = _mm_setzero_ps();
1009
1010             /**************************
1011              * CALCULATE INTERACTIONS *
1012              **************************/
1013
1014             if (gmx_mm_any_lt(rsq00,rcutoff2))
1015             {
1016
1017             r00              = _mm_mul_ps(rsq00,rinv00);
1018
1019             /* Compute parameters for interactions between i and j atoms */
1020             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1021                                          vdwparam+vdwioffset0+vdwjidx0B,
1022                                          vdwparam+vdwioffset0+vdwjidx0C,
1023                                          vdwparam+vdwioffset0+vdwjidx0D,
1024                                          &c6_00,&c12_00);
1025
1026             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1027                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1028                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1029                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1030
1031             /* Analytical LJ-PME */
1032             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1033             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1034             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1035             exponent         = gmx_simd_exp_r(ewcljrsq);
1036             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1037             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
1038             /* f6A = 6 * C6grid * (1 - poly) */
1039             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1040             /* f6B = C6grid * exponent * beta^6 */
1041             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1042             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1043             fvdw              = _mm_mul_ps(_mm_macc_ps(_mm_msub_ps(c12_00,rinvsix,_mm_sub_ps(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1044
1045             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1046
1047             fscal            = fvdw;
1048
1049             fscal            = _mm_and_ps(fscal,cutoff_mask);
1050
1051              /* Update vectorial force */
1052             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1053             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1054             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1055
1056             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1057             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1058             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1059
1060             }
1061
1062             /**************************
1063              * CALCULATE INTERACTIONS *
1064              **************************/
1065
1066             if (gmx_mm_any_lt(rsq10,rcutoff2))
1067             {
1068
1069             r10              = _mm_mul_ps(rsq10,rinv10);
1070
1071             /* Compute parameters for interactions between i and j atoms */
1072             qq10             = _mm_mul_ps(iq1,jq0);
1073
1074             /* EWALD ELECTROSTATICS */
1075
1076             /* Analytical PME correction */
1077             zeta2            = _mm_mul_ps(beta2,rsq10);
1078             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1079             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1080             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1081             felec            = _mm_mul_ps(qq10,felec);
1082
1083             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1084
1085             fscal            = felec;
1086
1087             fscal            = _mm_and_ps(fscal,cutoff_mask);
1088
1089              /* Update vectorial force */
1090             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1091             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1092             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1093
1094             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1095             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1096             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1097
1098             }
1099
1100             /**************************
1101              * CALCULATE INTERACTIONS *
1102              **************************/
1103
1104             if (gmx_mm_any_lt(rsq20,rcutoff2))
1105             {
1106
1107             r20              = _mm_mul_ps(rsq20,rinv20);
1108
1109             /* Compute parameters for interactions between i and j atoms */
1110             qq20             = _mm_mul_ps(iq2,jq0);
1111
1112             /* EWALD ELECTROSTATICS */
1113
1114             /* Analytical PME correction */
1115             zeta2            = _mm_mul_ps(beta2,rsq20);
1116             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1117             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1118             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1119             felec            = _mm_mul_ps(qq20,felec);
1120
1121             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1122
1123             fscal            = felec;
1124
1125             fscal            = _mm_and_ps(fscal,cutoff_mask);
1126
1127              /* Update vectorial force */
1128             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1129             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1130             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1131
1132             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1133             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1134             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1135
1136             }
1137
1138             /**************************
1139              * CALCULATE INTERACTIONS *
1140              **************************/
1141
1142             if (gmx_mm_any_lt(rsq30,rcutoff2))
1143             {
1144
1145             r30              = _mm_mul_ps(rsq30,rinv30);
1146
1147             /* Compute parameters for interactions between i and j atoms */
1148             qq30             = _mm_mul_ps(iq3,jq0);
1149
1150             /* EWALD ELECTROSTATICS */
1151
1152             /* Analytical PME correction */
1153             zeta2            = _mm_mul_ps(beta2,rsq30);
1154             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
1155             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1156             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1157             felec            = _mm_mul_ps(qq30,felec);
1158
1159             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1160
1161             fscal            = felec;
1162
1163             fscal            = _mm_and_ps(fscal,cutoff_mask);
1164
1165              /* Update vectorial force */
1166             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1167             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1168             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1169
1170             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1171             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1172             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1173
1174             }
1175
1176             fjptrA             = f+j_coord_offsetA;
1177             fjptrB             = f+j_coord_offsetB;
1178             fjptrC             = f+j_coord_offsetC;
1179             fjptrD             = f+j_coord_offsetD;
1180
1181             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1182
1183             /* Inner loop uses 143 flops */
1184         }
1185
1186         if(jidx<j_index_end)
1187         {
1188
1189             /* Get j neighbor index, and coordinate index */
1190             jnrlistA         = jjnr[jidx];
1191             jnrlistB         = jjnr[jidx+1];
1192             jnrlistC         = jjnr[jidx+2];
1193             jnrlistD         = jjnr[jidx+3];
1194             /* Sign of each element will be negative for non-real atoms.
1195              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1196              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1197              */
1198             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1199             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1200             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1201             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1202             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1203             j_coord_offsetA  = DIM*jnrA;
1204             j_coord_offsetB  = DIM*jnrB;
1205             j_coord_offsetC  = DIM*jnrC;
1206             j_coord_offsetD  = DIM*jnrD;
1207
1208             /* load j atom coordinates */
1209             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1210                                               x+j_coord_offsetC,x+j_coord_offsetD,
1211                                               &jx0,&jy0,&jz0);
1212
1213             /* Calculate displacement vector */
1214             dx00             = _mm_sub_ps(ix0,jx0);
1215             dy00             = _mm_sub_ps(iy0,jy0);
1216             dz00             = _mm_sub_ps(iz0,jz0);
1217             dx10             = _mm_sub_ps(ix1,jx0);
1218             dy10             = _mm_sub_ps(iy1,jy0);
1219             dz10             = _mm_sub_ps(iz1,jz0);
1220             dx20             = _mm_sub_ps(ix2,jx0);
1221             dy20             = _mm_sub_ps(iy2,jy0);
1222             dz20             = _mm_sub_ps(iz2,jz0);
1223             dx30             = _mm_sub_ps(ix3,jx0);
1224             dy30             = _mm_sub_ps(iy3,jy0);
1225             dz30             = _mm_sub_ps(iz3,jz0);
1226
1227             /* Calculate squared distance and things based on it */
1228             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1229             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1230             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1231             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1232
1233             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1234             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1235             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1236             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1237
1238             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1239             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1240             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1241             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1242
1243             /* Load parameters for j particles */
1244             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1245                                                               charge+jnrC+0,charge+jnrD+0);
1246             vdwjidx0A        = 2*vdwtype[jnrA+0];
1247             vdwjidx0B        = 2*vdwtype[jnrB+0];
1248             vdwjidx0C        = 2*vdwtype[jnrC+0];
1249             vdwjidx0D        = 2*vdwtype[jnrD+0];
1250
1251             fjx0             = _mm_setzero_ps();
1252             fjy0             = _mm_setzero_ps();
1253             fjz0             = _mm_setzero_ps();
1254
1255             /**************************
1256              * CALCULATE INTERACTIONS *
1257              **************************/
1258
1259             if (gmx_mm_any_lt(rsq00,rcutoff2))
1260             {
1261
1262             r00              = _mm_mul_ps(rsq00,rinv00);
1263             r00              = _mm_andnot_ps(dummy_mask,r00);
1264
1265             /* Compute parameters for interactions between i and j atoms */
1266             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1267                                          vdwparam+vdwioffset0+vdwjidx0B,
1268                                          vdwparam+vdwioffset0+vdwjidx0C,
1269                                          vdwparam+vdwioffset0+vdwjidx0D,
1270                                          &c6_00,&c12_00);
1271
1272             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1273                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1274                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1275                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1276
1277             /* Analytical LJ-PME */
1278             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1279             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1280             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1281             exponent         = gmx_simd_exp_r(ewcljrsq);
1282             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1283             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
1284             /* f6A = 6 * C6grid * (1 - poly) */
1285             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1286             /* f6B = C6grid * exponent * beta^6 */
1287             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1288             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1289             fvdw              = _mm_mul_ps(_mm_macc_ps(_mm_msub_ps(c12_00,rinvsix,_mm_sub_ps(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1290
1291             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1292
1293             fscal            = fvdw;
1294
1295             fscal            = _mm_and_ps(fscal,cutoff_mask);
1296
1297             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1298
1299              /* Update vectorial force */
1300             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1301             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1302             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1303
1304             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1305             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1306             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1307
1308             }
1309
1310             /**************************
1311              * CALCULATE INTERACTIONS *
1312              **************************/
1313
1314             if (gmx_mm_any_lt(rsq10,rcutoff2))
1315             {
1316
1317             r10              = _mm_mul_ps(rsq10,rinv10);
1318             r10              = _mm_andnot_ps(dummy_mask,r10);
1319
1320             /* Compute parameters for interactions between i and j atoms */
1321             qq10             = _mm_mul_ps(iq1,jq0);
1322
1323             /* EWALD ELECTROSTATICS */
1324
1325             /* Analytical PME correction */
1326             zeta2            = _mm_mul_ps(beta2,rsq10);
1327             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1328             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1329             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1330             felec            = _mm_mul_ps(qq10,felec);
1331
1332             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1333
1334             fscal            = felec;
1335
1336             fscal            = _mm_and_ps(fscal,cutoff_mask);
1337
1338             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1339
1340              /* Update vectorial force */
1341             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1342             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1343             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1344
1345             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1346             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1347             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1348
1349             }
1350
1351             /**************************
1352              * CALCULATE INTERACTIONS *
1353              **************************/
1354
1355             if (gmx_mm_any_lt(rsq20,rcutoff2))
1356             {
1357
1358             r20              = _mm_mul_ps(rsq20,rinv20);
1359             r20              = _mm_andnot_ps(dummy_mask,r20);
1360
1361             /* Compute parameters for interactions between i and j atoms */
1362             qq20             = _mm_mul_ps(iq2,jq0);
1363
1364             /* EWALD ELECTROSTATICS */
1365
1366             /* Analytical PME correction */
1367             zeta2            = _mm_mul_ps(beta2,rsq20);
1368             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1369             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1370             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1371             felec            = _mm_mul_ps(qq20,felec);
1372
1373             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1374
1375             fscal            = felec;
1376
1377             fscal            = _mm_and_ps(fscal,cutoff_mask);
1378
1379             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1380
1381              /* Update vectorial force */
1382             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1383             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1384             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1385
1386             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1387             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1388             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1389
1390             }
1391
1392             /**************************
1393              * CALCULATE INTERACTIONS *
1394              **************************/
1395
1396             if (gmx_mm_any_lt(rsq30,rcutoff2))
1397             {
1398
1399             r30              = _mm_mul_ps(rsq30,rinv30);
1400             r30              = _mm_andnot_ps(dummy_mask,r30);
1401
1402             /* Compute parameters for interactions between i and j atoms */
1403             qq30             = _mm_mul_ps(iq3,jq0);
1404
1405             /* EWALD ELECTROSTATICS */
1406
1407             /* Analytical PME correction */
1408             zeta2            = _mm_mul_ps(beta2,rsq30);
1409             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
1410             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1411             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1412             felec            = _mm_mul_ps(qq30,felec);
1413
1414             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1415
1416             fscal            = felec;
1417
1418             fscal            = _mm_and_ps(fscal,cutoff_mask);
1419
1420             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1421
1422              /* Update vectorial force */
1423             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1424             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1425             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1426
1427             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1428             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1429             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1430
1431             }
1432
1433             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1434             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1435             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1436             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1437
1438             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1439
1440             /* Inner loop uses 147 flops */
1441         }
1442
1443         /* End of innermost loop */
1444
1445         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1446                                               f+i_coord_offset,fshift+i_shift_offset);
1447
1448         /* Increment number of inner iterations */
1449         inneriter                  += j_index_end - j_index_start;
1450
1451         /* Outer loop uses 24 flops */
1452     }
1453
1454     /* Increment number of outer iterations */
1455     outeriter        += nri;
1456
1457     /* Update outer/inner flops */
1458
1459     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*147);
1460 }