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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  50 #include "kernelutil_x86_avx_128_fma_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_VF_avx_128_fma_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_VF_avx_128_fma_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwioffset3;
  92     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  94     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  97     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  99     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 106     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 107     __m128           c6grid_00;
 108     __m128           c6grid_10;
 109     __m128           c6grid_20;
 110     __m128           c6grid_30;
 111     real             *vdwgridparam;
 112     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 113     __m128           one_half = _mm_set1_ps(0.5);
 114     __m128           minus_one = _mm_set1_ps(-1.0);
 115     __m128i          ewitab;
 116     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 117     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 118     real             *ewtab;
 119     __m128           dummy_mask,cutoff_mask;
 120     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 121     __m128           one     = _mm_set1_ps(1.0);
 122     __m128           two     = _mm_set1_ps(2.0);
 123     x                = xx[0];
 124     f                = ff[0];
 125
 126     nri              = nlist->nri;
 127     iinr             = nlist->iinr;
 128     jindex           = nlist->jindex;
 129     jjnr             = nlist->jjnr;
 130     shiftidx         = nlist->shift;
 131     gid              = nlist->gid;
 132     shiftvec         = fr->shift_vec[0];
 133     fshift           = fr->fshift[0];
 134     facel            = _mm_set1_ps(fr->epsfac);
 135     charge           = mdatoms->chargeA;
 136     nvdwtype         = fr->ntype;
 137     vdwparam         = fr->nbfp;
 138     vdwtype          = mdatoms->typeA;
 139     vdwgridparam     = fr->ljpme_c6grid;
 140     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 141     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 142     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 143
 144     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 145     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 146     beta2            = _mm_mul_ps(beta,beta);
 147     beta3            = _mm_mul_ps(beta,beta2);
 148     ewtab            = fr->ic->tabq_coul_FDV0;
 149     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 150     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 151
 152     /* Setup water-specific parameters */
 153     inr              = nlist->iinr[0];
 154     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 155     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 156     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 157     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 158
 159     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 160     rcutoff_scalar   = fr->rcoulomb;
 161     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 162     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 163
 164     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 165     rvdw             = _mm_set1_ps(fr->rvdw);
 166
 167     /* Avoid stupid compiler warnings */
 168     jnrA = jnrB = jnrC = jnrD = 0;
 169     j_coord_offsetA = 0;
 170     j_coord_offsetB = 0;
 171     j_coord_offsetC = 0;
 172     j_coord_offsetD = 0;
 173
 174     outeriter        = 0;
 175     inneriter        = 0;
 176
 177     for(iidx=0;iidx<4*DIM;iidx++)
 178     {
 179         scratch[iidx] = 0.0;
 180     }
 181
 182     /* Start outer loop over neighborlists */
 183     for(iidx=0; iidx<nri; iidx++)
 184     {
 185         /* Load shift vector for this list */
 186         i_shift_offset   = DIM*shiftidx[iidx];
 187
 188         /* Load limits for loop over neighbors */
 189         j_index_start    = jindex[iidx];
 190         j_index_end      = jindex[iidx+1];
 191
 192         /* Get outer coordinate index */
 193         inr              = iinr[iidx];
 194         i_coord_offset   = DIM*inr;
 195
 196         /* Load i particle coords and add shift vector */
 197         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 198                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 199
 200         fix0             = _mm_setzero_ps();
 201         fiy0             = _mm_setzero_ps();
 202         fiz0             = _mm_setzero_ps();
 203         fix1             = _mm_setzero_ps();
 204         fiy1             = _mm_setzero_ps();
 205         fiz1             = _mm_setzero_ps();
 206         fix2             = _mm_setzero_ps();
 207         fiy2             = _mm_setzero_ps();
 208         fiz2             = _mm_setzero_ps();
 209         fix3             = _mm_setzero_ps();
 210         fiy3             = _mm_setzero_ps();
 211         fiz3             = _mm_setzero_ps();
 212
 213         /* Reset potential sums */
 214         velecsum         = _mm_setzero_ps();
 215         vvdwsum          = _mm_setzero_ps();
 216
 217         /* Start inner kernel loop */
 218         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 219         {
 220
 221             /* Get j neighbor index, and coordinate index */
 222             jnrA             = jjnr[jidx];
 223             jnrB             = jjnr[jidx+1];
 224             jnrC             = jjnr[jidx+2];
 225             jnrD             = jjnr[jidx+3];
 226             j_coord_offsetA  = DIM*jnrA;
 227             j_coord_offsetB  = DIM*jnrB;
 228             j_coord_offsetC  = DIM*jnrC;
 229             j_coord_offsetD  = DIM*jnrD;
 230
 231             /* load j atom coordinates */
 232             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 233                                               x+j_coord_offsetC,x+j_coord_offsetD,
 234                                               &jx0,&jy0,&jz0);
 235
 236             /* Calculate displacement vector */
 237             dx00             = _mm_sub_ps(ix0,jx0);
 238             dy00             = _mm_sub_ps(iy0,jy0);
 239             dz00             = _mm_sub_ps(iz0,jz0);
 240             dx10             = _mm_sub_ps(ix1,jx0);
 241             dy10             = _mm_sub_ps(iy1,jy0);
 242             dz10             = _mm_sub_ps(iz1,jz0);
 243             dx20             = _mm_sub_ps(ix2,jx0);
 244             dy20             = _mm_sub_ps(iy2,jy0);
 245             dz20             = _mm_sub_ps(iz2,jz0);
 246             dx30             = _mm_sub_ps(ix3,jx0);
 247             dy30             = _mm_sub_ps(iy3,jy0);
 248             dz30             = _mm_sub_ps(iz3,jz0);
 249
 250             /* Calculate squared distance and things based on it */
 251             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 252             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 253             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 254             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 255
 256             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 257             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 258             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 259             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 260
 261             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 262             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 263             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 264             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 265
 266             /* Load parameters for j particles */
 267             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 268                                                               charge+jnrC+0,charge+jnrD+0);
 269             vdwjidx0A        = 2*vdwtype[jnrA+0];
 270             vdwjidx0B        = 2*vdwtype[jnrB+0];
 271             vdwjidx0C        = 2*vdwtype[jnrC+0];
 272             vdwjidx0D        = 2*vdwtype[jnrD+0];
 273
 274             fjx0             = _mm_setzero_ps();
 275             fjy0             = _mm_setzero_ps();
 276             fjz0             = _mm_setzero_ps();
 277
 278             /**************************
 279              * CALCULATE INTERACTIONS *
 280              **************************/
 281
 282             if (gmx_mm_any_lt(rsq00,rcutoff2))
 283             {
 284
 285             r00              = _mm_mul_ps(rsq00,rinv00);
 286
 287             /* Compute parameters for interactions between i and j atoms */
 288             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 289                                          vdwparam+vdwioffset0+vdwjidx0B,
 290                                          vdwparam+vdwioffset0+vdwjidx0C,
 291                                          vdwparam+vdwioffset0+vdwjidx0D,
 292                                          &c6_00,&c12_00);
 293
 294             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 295                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 296                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 297                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 298
 299             /* Analytical LJ-PME */
 300             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 301             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 302             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 303             exponent         = gmx_simd_exp_r(ewcljrsq);
 304             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 305             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
 306             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 307             vvdw6            = _mm_mul_ps(_mm_macc_ps(-c6grid_00,_mm_sub_ps(one,poly),c6_00),rinvsix);
 308             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 309             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 310                                           _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));
 311             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 312             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);
 313
 314             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 315
 316             /* Update potential sum for this i atom from the interaction with this j atom. */
 317             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 318             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 319
 320             fscal            = fvdw;
 321
 322             fscal            = _mm_and_ps(fscal,cutoff_mask);
 323
 324              /* Update vectorial force */
 325             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 326             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 327             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 328
 329             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 330             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 331             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 332
 333             }
 334
 335             /**************************
 336              * CALCULATE INTERACTIONS *
 337              **************************/
 338
 339             if (gmx_mm_any_lt(rsq10,rcutoff2))
 340             {
 341
 342             r10              = _mm_mul_ps(rsq10,rinv10);
 343
 344             /* Compute parameters for interactions between i and j atoms */
 345             qq10             = _mm_mul_ps(iq1,jq0);
 346
 347             /* EWALD ELECTROSTATICS */
 348
 349             /* Analytical PME correction */
 350             zeta2            = _mm_mul_ps(beta2,rsq10);
 351             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 352             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 353             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 354             felec            = _mm_mul_ps(qq10,felec);
 355             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 356             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv10,sh_ewald));
 357             velec            = _mm_mul_ps(qq10,velec);
 358
 359             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 360
 361             /* Update potential sum for this i atom from the interaction with this j atom. */
 362             velec            = _mm_and_ps(velec,cutoff_mask);
 363             velecsum         = _mm_add_ps(velecsum,velec);
 364
 365             fscal            = felec;
 366
 367             fscal            = _mm_and_ps(fscal,cutoff_mask);
 368
 369              /* Update vectorial force */
 370             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 371             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 372             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 373
 374             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 375             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 376             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 377
 378             }
 379
 380             /**************************
 381              * CALCULATE INTERACTIONS *
 382              **************************/
 383
 384             if (gmx_mm_any_lt(rsq20,rcutoff2))
 385             {
 386
 387             r20              = _mm_mul_ps(rsq20,rinv20);
 388
 389             /* Compute parameters for interactions between i and j atoms */
 390             qq20             = _mm_mul_ps(iq2,jq0);
 391
 392             /* EWALD ELECTROSTATICS */
 393
 394             /* Analytical PME correction */
 395             zeta2            = _mm_mul_ps(beta2,rsq20);
 396             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 397             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 398             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 399             felec            = _mm_mul_ps(qq20,felec);
 400             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 401             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv20,sh_ewald));
 402             velec            = _mm_mul_ps(qq20,velec);
 403
 404             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 405
 406             /* Update potential sum for this i atom from the interaction with this j atom. */
 407             velec            = _mm_and_ps(velec,cutoff_mask);
 408             velecsum         = _mm_add_ps(velecsum,velec);
 409
 410             fscal            = felec;
 411
 412             fscal            = _mm_and_ps(fscal,cutoff_mask);
 413
 414              /* Update vectorial force */
 415             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 416             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 417             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 418
 419             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 420             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 421             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 422
 423             }
 424
 425             /**************************
 426              * CALCULATE INTERACTIONS *
 427              **************************/
 428
 429             if (gmx_mm_any_lt(rsq30,rcutoff2))
 430             {
 431
 432             r30              = _mm_mul_ps(rsq30,rinv30);
 433
 434             /* Compute parameters for interactions between i and j atoms */
 435             qq30             = _mm_mul_ps(iq3,jq0);
 436
 437             /* EWALD ELECTROSTATICS */
 438
 439             /* Analytical PME correction */
 440             zeta2            = _mm_mul_ps(beta2,rsq30);
 441             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 442             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 443             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 444             felec            = _mm_mul_ps(qq30,felec);
 445             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 446             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv30,sh_ewald));
 447             velec            = _mm_mul_ps(qq30,velec);
 448
 449             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 450
 451             /* Update potential sum for this i atom from the interaction with this j atom. */
 452             velec            = _mm_and_ps(velec,cutoff_mask);
 453             velecsum         = _mm_add_ps(velecsum,velec);
 454
 455             fscal            = felec;
 456
 457             fscal            = _mm_and_ps(fscal,cutoff_mask);
 458
 459              /* Update vectorial force */
 460             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 461             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 462             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 463
 464             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 465             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 466             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 467
 468             }
 469
 470             fjptrA             = f+j_coord_offsetA;
 471             fjptrB             = f+j_coord_offsetB;
 472             fjptrC             = f+j_coord_offsetC;
 473             fjptrD             = f+j_coord_offsetD;
 474
 475             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 476
 477             /* Inner loop uses 158 flops */
 478         }
 479
 480         if(jidx<j_index_end)
 481         {
 482
 483             /* Get j neighbor index, and coordinate index */
 484             jnrlistA         = jjnr[jidx];
 485             jnrlistB         = jjnr[jidx+1];
 486             jnrlistC         = jjnr[jidx+2];
 487             jnrlistD         = jjnr[jidx+3];
 488             /* Sign of each element will be negative for non-real atoms.
 489              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 490              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 491              */
 492             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 493             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 494             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 495             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 496             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 497             j_coord_offsetA  = DIM*jnrA;
 498             j_coord_offsetB  = DIM*jnrB;
 499             j_coord_offsetC  = DIM*jnrC;
 500             j_coord_offsetD  = DIM*jnrD;
 501
 502             /* load j atom coordinates */
 503             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 504                                               x+j_coord_offsetC,x+j_coord_offsetD,
 505                                               &jx0,&jy0,&jz0);
 506
 507             /* Calculate displacement vector */
 508             dx00             = _mm_sub_ps(ix0,jx0);
 509             dy00             = _mm_sub_ps(iy0,jy0);
 510             dz00             = _mm_sub_ps(iz0,jz0);
 511             dx10             = _mm_sub_ps(ix1,jx0);
 512             dy10             = _mm_sub_ps(iy1,jy0);
 513             dz10             = _mm_sub_ps(iz1,jz0);
 514             dx20             = _mm_sub_ps(ix2,jx0);
 515             dy20             = _mm_sub_ps(iy2,jy0);
 516             dz20             = _mm_sub_ps(iz2,jz0);
 517             dx30             = _mm_sub_ps(ix3,jx0);
 518             dy30             = _mm_sub_ps(iy3,jy0);
 519             dz30             = _mm_sub_ps(iz3,jz0);
 520
 521             /* Calculate squared distance and things based on it */
 522             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 523             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 524             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 525             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 526
 527             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 528             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 529             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 530             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 531
 532             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 533             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 534             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 535             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 536
 537             /* Load parameters for j particles */
 538             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 539                                                               charge+jnrC+0,charge+jnrD+0);
 540             vdwjidx0A        = 2*vdwtype[jnrA+0];
 541             vdwjidx0B        = 2*vdwtype[jnrB+0];
 542             vdwjidx0C        = 2*vdwtype[jnrC+0];
 543             vdwjidx0D        = 2*vdwtype[jnrD+0];
 544
 545             fjx0             = _mm_setzero_ps();
 546             fjy0             = _mm_setzero_ps();
 547             fjz0             = _mm_setzero_ps();
 548
 549             /**************************
 550              * CALCULATE INTERACTIONS *
 551              **************************/
 552
 553             if (gmx_mm_any_lt(rsq00,rcutoff2))
 554             {
 555
 556             r00              = _mm_mul_ps(rsq00,rinv00);
 557             r00              = _mm_andnot_ps(dummy_mask,r00);
 558
 559             /* Compute parameters for interactions between i and j atoms */
 560             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 561                                          vdwparam+vdwioffset0+vdwjidx0B,
 562                                          vdwparam+vdwioffset0+vdwjidx0C,
 563                                          vdwparam+vdwioffset0+vdwjidx0D,
 564                                          &c6_00,&c12_00);
 565
 566             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 567                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 568                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 569                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 570
 571             /* Analytical LJ-PME */
 572             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 573             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 574             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 575             exponent         = gmx_simd_exp_r(ewcljrsq);
 576             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 577             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
 578             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 579             vvdw6            = _mm_mul_ps(_mm_macc_ps(-c6grid_00,_mm_sub_ps(one,poly),c6_00),rinvsix);
 580             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 581             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 582                                           _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));
 583             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 584             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);
 585
 586             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 587
 588             /* Update potential sum for this i atom from the interaction with this j atom. */
 589             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 590             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 591             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 592
 593             fscal            = fvdw;
 594
 595             fscal            = _mm_and_ps(fscal,cutoff_mask);
 596
 597             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 598
 599              /* Update vectorial force */
 600             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 601             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 602             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 603
 604             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 605             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 606             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 607
 608             }
 609
 610             /**************************
 611              * CALCULATE INTERACTIONS *
 612              **************************/
 613
 614             if (gmx_mm_any_lt(rsq10,rcutoff2))
 615             {
 616
 617             r10              = _mm_mul_ps(rsq10,rinv10);
 618             r10              = _mm_andnot_ps(dummy_mask,r10);
 619
 620             /* Compute parameters for interactions between i and j atoms */
 621             qq10             = _mm_mul_ps(iq1,jq0);
 622
 623             /* EWALD ELECTROSTATICS */
 624
 625             /* Analytical PME correction */
 626             zeta2            = _mm_mul_ps(beta2,rsq10);
 627             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 628             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 629             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 630             felec            = _mm_mul_ps(qq10,felec);
 631             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 632             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv10,sh_ewald));
 633             velec            = _mm_mul_ps(qq10,velec);
 634
 635             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 636
 637             /* Update potential sum for this i atom from the interaction with this j atom. */
 638             velec            = _mm_and_ps(velec,cutoff_mask);
 639             velec            = _mm_andnot_ps(dummy_mask,velec);
 640             velecsum         = _mm_add_ps(velecsum,velec);
 641
 642             fscal            = felec;
 643
 644             fscal            = _mm_and_ps(fscal,cutoff_mask);
 645
 646             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 647
 648              /* Update vectorial force */
 649             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 650             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 651             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 652
 653             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 654             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 655             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 656
 657             }
 658
 659             /**************************
 660              * CALCULATE INTERACTIONS *
 661              **************************/
 662
 663             if (gmx_mm_any_lt(rsq20,rcutoff2))
 664             {
 665
 666             r20              = _mm_mul_ps(rsq20,rinv20);
 667             r20              = _mm_andnot_ps(dummy_mask,r20);
 668
 669             /* Compute parameters for interactions between i and j atoms */
 670             qq20             = _mm_mul_ps(iq2,jq0);
 671
 672             /* EWALD ELECTROSTATICS */
 673
 674             /* Analytical PME correction */
 675             zeta2            = _mm_mul_ps(beta2,rsq20);
 676             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 677             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 678             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 679             felec            = _mm_mul_ps(qq20,felec);
 680             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 681             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv20,sh_ewald));
 682             velec            = _mm_mul_ps(qq20,velec);
 683
 684             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 685
 686             /* Update potential sum for this i atom from the interaction with this j atom. */
 687             velec            = _mm_and_ps(velec,cutoff_mask);
 688             velec            = _mm_andnot_ps(dummy_mask,velec);
 689             velecsum         = _mm_add_ps(velecsum,velec);
 690
 691             fscal            = felec;
 692
 693             fscal            = _mm_and_ps(fscal,cutoff_mask);
 694
 695             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 696
 697              /* Update vectorial force */
 698             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 699             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 700             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 701
 702             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 703             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 704             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 705
 706             }
 707
 708             /**************************
 709              * CALCULATE INTERACTIONS *
 710              **************************/
 711
 712             if (gmx_mm_any_lt(rsq30,rcutoff2))
 713             {
 714
 715             r30              = _mm_mul_ps(rsq30,rinv30);
 716             r30              = _mm_andnot_ps(dummy_mask,r30);
 717
 718             /* Compute parameters for interactions between i and j atoms */
 719             qq30             = _mm_mul_ps(iq3,jq0);
 720
 721             /* EWALD ELECTROSTATICS */
 722
 723             /* Analytical PME correction */
 724             zeta2            = _mm_mul_ps(beta2,rsq30);
 725             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 726             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 727             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 728             felec            = _mm_mul_ps(qq30,felec);
 729             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 730             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv30,sh_ewald));
 731             velec            = _mm_mul_ps(qq30,velec);
 732
 733             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 734
 735             /* Update potential sum for this i atom from the interaction with this j atom. */
 736             velec            = _mm_and_ps(velec,cutoff_mask);
 737             velec            = _mm_andnot_ps(dummy_mask,velec);
 738             velecsum         = _mm_add_ps(velecsum,velec);
 739
 740             fscal            = felec;
 741
 742             fscal            = _mm_and_ps(fscal,cutoff_mask);
 743
 744             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 745
 746              /* Update vectorial force */
 747             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 748             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 749             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 750
 751             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 752             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 753             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 754
 755             }
 756
 757             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 758             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 759             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 760             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 761
 762             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 763
 764             /* Inner loop uses 162 flops */
 765         }
 766
 767         /* End of innermost loop */
 768
 769         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 770                                               f+i_coord_offset,fshift+i_shift_offset);
 771
 772         ggid                        = gid[iidx];
 773         /* Update potential energies */
 774         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 775         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 776
 777         /* Increment number of inner iterations */
 778         inneriter                  += j_index_end - j_index_start;
 779
 780         /* Outer loop uses 26 flops */
 781     }
 782
 783     /* Increment number of outer iterations */
 784     outeriter        += nri;
 785
 786     /* Update outer/inner flops */
 787
 788     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*162);
 789 }
 790 /*
 791  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_avx_128_fma_single
 792  * Electrostatics interaction: Ewald
 793  * VdW interaction:            LJEwald
 794  * Geometry:                   Water4-Particle
 795  * Calculate force/pot:        Force
 796  */
 797 void
 798 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_avx_128_fma_single
 799                     (t_nblist                    * gmx_restrict       nlist,
 800                      rvec                        * gmx_restrict          xx,
 801                      rvec                        * gmx_restrict          ff,
 802                      t_forcerec                  * gmx_restrict          fr,
 803                      t_mdatoms                   * gmx_restrict     mdatoms,
 804                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 805                      t_nrnb                      * gmx_restrict        nrnb)
 806 {
 807     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 808      * just 0 for non-waters.
 809      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 810      * jnr indices corresponding to data put in the four positions in the SIMD register.
 811      */
 812     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 813     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 814     int              jnrA,jnrB,jnrC,jnrD;
 815     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 816     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 817     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 818     real             rcutoff_scalar;
 819     real             *shiftvec,*fshift,*x,*f;
 820     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 821     real             scratch[4*DIM];
 822     __m128           fscal,rcutoff,rcutoff2,jidxall;
 823     int              vdwioffset0;
 824     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 825     int              vdwioffset1;
 826     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 827     int              vdwioffset2;
 828     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 829     int              vdwioffset3;
 830     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 831     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 832     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 833     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 834     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 835     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 836     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 837     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 838     real             *charge;
 839     int              nvdwtype;
 840     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 841     int              *vdwtype;
 842     real             *vdwparam;
 843     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 844     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 845     __m128           c6grid_00;
 846     __m128           c6grid_10;
 847     __m128           c6grid_20;
 848     __m128           c6grid_30;
 849     real             *vdwgridparam;
 850     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 851     __m128           one_half = _mm_set1_ps(0.5);
 852     __m128           minus_one = _mm_set1_ps(-1.0);
 853     __m128i          ewitab;
 854     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 855     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 856     real             *ewtab;
 857     __m128           dummy_mask,cutoff_mask;
 858     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 859     __m128           one     = _mm_set1_ps(1.0);
 860     __m128           two     = _mm_set1_ps(2.0);
 861     x                = xx[0];
 862     f                = ff[0];
 863
 864     nri              = nlist->nri;
 865     iinr             = nlist->iinr;
 866     jindex           = nlist->jindex;
 867     jjnr             = nlist->jjnr;
 868     shiftidx         = nlist->shift;
 869     gid              = nlist->gid;
 870     shiftvec         = fr->shift_vec[0];
 871     fshift           = fr->fshift[0];
 872     facel            = _mm_set1_ps(fr->epsfac);
 873     charge           = mdatoms->chargeA;
 874     nvdwtype         = fr->ntype;
 875     vdwparam         = fr->nbfp;
 876     vdwtype          = mdatoms->typeA;
 877     vdwgridparam     = fr->ljpme_c6grid;
 878     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 879     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 880     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 881
 882     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 883     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 884     beta2            = _mm_mul_ps(beta,beta);
 885     beta3            = _mm_mul_ps(beta,beta2);
 886     ewtab            = fr->ic->tabq_coul_F;
 887     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 888     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 889
 890     /* Setup water-specific parameters */
 891     inr              = nlist->iinr[0];
 892     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 893     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 894     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 895     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 896
 897     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 898     rcutoff_scalar   = fr->rcoulomb;
 899     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 900     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 901
 902     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 903     rvdw             = _mm_set1_ps(fr->rvdw);
 904
 905     /* Avoid stupid compiler warnings */
 906     jnrA = jnrB = jnrC = jnrD = 0;
 907     j_coord_offsetA = 0;
 908     j_coord_offsetB = 0;
 909     j_coord_offsetC = 0;
 910     j_coord_offsetD = 0;
 911
 912     outeriter        = 0;
 913     inneriter        = 0;
 914
 915     for(iidx=0;iidx<4*DIM;iidx++)
 916     {
 917         scratch[iidx] = 0.0;
 918     }
 919
 920     /* Start outer loop over neighborlists */
 921     for(iidx=0; iidx<nri; iidx++)
 922     {
 923         /* Load shift vector for this list */
 924         i_shift_offset   = DIM*shiftidx[iidx];
 925
 926         /* Load limits for loop over neighbors */
 927         j_index_start    = jindex[iidx];
 928         j_index_end      = jindex[iidx+1];
 929
 930         /* Get outer coordinate index */
 931         inr              = iinr[iidx];
 932         i_coord_offset   = DIM*inr;
 933
 934         /* Load i particle coords and add shift vector */
 935         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 936                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 937
 938         fix0             = _mm_setzero_ps();
 939         fiy0             = _mm_setzero_ps();
 940         fiz0             = _mm_setzero_ps();
 941         fix1             = _mm_setzero_ps();
 942         fiy1             = _mm_setzero_ps();
 943         fiz1             = _mm_setzero_ps();
 944         fix2             = _mm_setzero_ps();
 945         fiy2             = _mm_setzero_ps();
 946         fiz2             = _mm_setzero_ps();
 947         fix3             = _mm_setzero_ps();
 948         fiy3             = _mm_setzero_ps();
 949         fiz3             = _mm_setzero_ps();
 950
 951         /* Start inner kernel loop */
 952         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 953         {
 954
 955             /* Get j neighbor index, and coordinate index */
 956             jnrA             = jjnr[jidx];
 957             jnrB             = jjnr[jidx+1];
 958             jnrC             = jjnr[jidx+2];
 959             jnrD             = jjnr[jidx+3];
 960             j_coord_offsetA  = DIM*jnrA;
 961             j_coord_offsetB  = DIM*jnrB;
 962             j_coord_offsetC  = DIM*jnrC;
 963             j_coord_offsetD  = DIM*jnrD;
 964
 965             /* load j atom coordinates */
 966             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 967                                               x+j_coord_offsetC,x+j_coord_offsetD,
 968                                               &jx0,&jy0,&jz0);
 969
 970             /* Calculate displacement vector */
 971             dx00             = _mm_sub_ps(ix0,jx0);
 972             dy00             = _mm_sub_ps(iy0,jy0);
 973             dz00             = _mm_sub_ps(iz0,jz0);
 974             dx10             = _mm_sub_ps(ix1,jx0);
 975             dy10             = _mm_sub_ps(iy1,jy0);
 976             dz10             = _mm_sub_ps(iz1,jz0);
 977             dx20             = _mm_sub_ps(ix2,jx0);
 978             dy20             = _mm_sub_ps(iy2,jy0);
 979             dz20             = _mm_sub_ps(iz2,jz0);
 980             dx30             = _mm_sub_ps(ix3,jx0);
 981             dy30             = _mm_sub_ps(iy3,jy0);
 982             dz30             = _mm_sub_ps(iz3,jz0);
 983
 984             /* Calculate squared distance and things based on it */
 985             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 986             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 987             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 988             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 989
 990             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 991             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 992             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 993             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 994
 995             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 996             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 997             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 998             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 999
1000             /* Load parameters for j particles */
1001             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1002                                                               charge+jnrC+0,charge+jnrD+0);
1003             vdwjidx0A        = 2*vdwtype[jnrA+0];
1004             vdwjidx0B        = 2*vdwtype[jnrB+0];
1005             vdwjidx0C        = 2*vdwtype[jnrC+0];
1006             vdwjidx0D        = 2*vdwtype[jnrD+0];
1007
1008             fjx0             = _mm_setzero_ps();
1009             fjy0             = _mm_setzero_ps();
1010             fjz0             = _mm_setzero_ps();
1011
1012             /**************************
1013              * CALCULATE INTERACTIONS *
1014              **************************/
1015
1016             if (gmx_mm_any_lt(rsq00,rcutoff2))
1017             {
1018
1019             r00              = _mm_mul_ps(rsq00,rinv00);
1020
1021             /* Compute parameters for interactions between i and j atoms */
1022             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1023                                          vdwparam+vdwioffset0+vdwjidx0B,
1024                                          vdwparam+vdwioffset0+vdwjidx0C,
1025                                          vdwparam+vdwioffset0+vdwjidx0D,
1026                                          &c6_00,&c12_00);
1027
1028             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1029                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1030                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1031                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1032
1033             /* Analytical LJ-PME */
1034             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1035             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1036             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1037             exponent         = gmx_simd_exp_r(ewcljrsq);
1038             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1039             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
1040             /* f6A = 6 * C6grid * (1 - poly) */
1041             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1042             /* f6B = C6grid * exponent * beta^6 */
1043             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1044             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1045             fvdw              = _mm_mul_ps(_mm_macc_ps(_mm_msub_ps(c12_00,rinvsix,_mm_sub_ps(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1046
1047             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1048
1049             fscal            = fvdw;
1050
1051             fscal            = _mm_and_ps(fscal,cutoff_mask);
1052
1053              /* Update vectorial force */
1054             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1055             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1056             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1057
1058             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1059             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1060             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1061
1062             }
1063
1064             /**************************
1065              * CALCULATE INTERACTIONS *
1066              **************************/
1067
1068             if (gmx_mm_any_lt(rsq10,rcutoff2))
1069             {
1070
1071             r10              = _mm_mul_ps(rsq10,rinv10);
1072
1073             /* Compute parameters for interactions between i and j atoms */
1074             qq10             = _mm_mul_ps(iq1,jq0);
1075
1076             /* EWALD ELECTROSTATICS */
1077
1078             /* Analytical PME correction */
1079             zeta2            = _mm_mul_ps(beta2,rsq10);
1080             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1081             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1082             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1083             felec            = _mm_mul_ps(qq10,felec);
1084
1085             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1086
1087             fscal            = felec;
1088
1089             fscal            = _mm_and_ps(fscal,cutoff_mask);
1090
1091              /* Update vectorial force */
1092             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1093             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1094             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1095
1096             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1097             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1098             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1099
1100             }
1101
1102             /**************************
1103              * CALCULATE INTERACTIONS *
1104              **************************/
1105
1106             if (gmx_mm_any_lt(rsq20,rcutoff2))
1107             {
1108
1109             r20              = _mm_mul_ps(rsq20,rinv20);
1110
1111             /* Compute parameters for interactions between i and j atoms */
1112             qq20             = _mm_mul_ps(iq2,jq0);
1113
1114             /* EWALD ELECTROSTATICS */
1115
1116             /* Analytical PME correction */
1117             zeta2            = _mm_mul_ps(beta2,rsq20);
1118             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1119             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1120             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1121             felec            = _mm_mul_ps(qq20,felec);
1122
1123             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1124
1125             fscal            = felec;
1126
1127             fscal            = _mm_and_ps(fscal,cutoff_mask);
1128
1129              /* Update vectorial force */
1130             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1131             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1132             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1133
1134             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1135             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1136             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1137
1138             }
1139
1140             /**************************
1141              * CALCULATE INTERACTIONS *
1142              **************************/
1143
1144             if (gmx_mm_any_lt(rsq30,rcutoff2))
1145             {
1146
1147             r30              = _mm_mul_ps(rsq30,rinv30);
1148
1149             /* Compute parameters for interactions between i and j atoms */
1150             qq30             = _mm_mul_ps(iq3,jq0);
1151
1152             /* EWALD ELECTROSTATICS */
1153
1154             /* Analytical PME correction */
1155             zeta2            = _mm_mul_ps(beta2,rsq30);
1156             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
1157             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1158             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1159             felec            = _mm_mul_ps(qq30,felec);
1160
1161             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1162
1163             fscal            = felec;
1164
1165             fscal            = _mm_and_ps(fscal,cutoff_mask);
1166
1167              /* Update vectorial force */
1168             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1169             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1170             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1171
1172             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1173             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1174             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1175
1176             }
1177
1178             fjptrA             = f+j_coord_offsetA;
1179             fjptrB             = f+j_coord_offsetB;
1180             fjptrC             = f+j_coord_offsetC;
1181             fjptrD             = f+j_coord_offsetD;
1182
1183             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1184
1185             /* Inner loop uses 143 flops */
1186         }
1187
1188         if(jidx<j_index_end)
1189         {
1190
1191             /* Get j neighbor index, and coordinate index */
1192             jnrlistA         = jjnr[jidx];
1193             jnrlistB         = jjnr[jidx+1];
1194             jnrlistC         = jjnr[jidx+2];
1195             jnrlistD         = jjnr[jidx+3];
1196             /* Sign of each element will be negative for non-real atoms.
1197              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1198              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1199              */
1200             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1201             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1202             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1203             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1204             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1205             j_coord_offsetA  = DIM*jnrA;
1206             j_coord_offsetB  = DIM*jnrB;
1207             j_coord_offsetC  = DIM*jnrC;
1208             j_coord_offsetD  = DIM*jnrD;
1209
1210             /* load j atom coordinates */
1211             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1212                                               x+j_coord_offsetC,x+j_coord_offsetD,
1213                                               &jx0,&jy0,&jz0);
1214
1215             /* Calculate displacement vector */
1216             dx00             = _mm_sub_ps(ix0,jx0);
1217             dy00             = _mm_sub_ps(iy0,jy0);
1218             dz00             = _mm_sub_ps(iz0,jz0);
1219             dx10             = _mm_sub_ps(ix1,jx0);
1220             dy10             = _mm_sub_ps(iy1,jy0);
1221             dz10             = _mm_sub_ps(iz1,jz0);
1222             dx20             = _mm_sub_ps(ix2,jx0);
1223             dy20             = _mm_sub_ps(iy2,jy0);
1224             dz20             = _mm_sub_ps(iz2,jz0);
1225             dx30             = _mm_sub_ps(ix3,jx0);
1226             dy30             = _mm_sub_ps(iy3,jy0);
1227             dz30             = _mm_sub_ps(iz3,jz0);
1228
1229             /* Calculate squared distance and things based on it */
1230             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1231             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1232             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1233             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1234
1235             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1236             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1237             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1238             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1239
1240             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1241             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1242             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1243             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1244
1245             /* Load parameters for j particles */
1246             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1247                                                               charge+jnrC+0,charge+jnrD+0);
1248             vdwjidx0A        = 2*vdwtype[jnrA+0];
1249             vdwjidx0B        = 2*vdwtype[jnrB+0];
1250             vdwjidx0C        = 2*vdwtype[jnrC+0];
1251             vdwjidx0D        = 2*vdwtype[jnrD+0];
1252
1253             fjx0             = _mm_setzero_ps();
1254             fjy0             = _mm_setzero_ps();
1255             fjz0             = _mm_setzero_ps();
1256
1257             /**************************
1258              * CALCULATE INTERACTIONS *
1259              **************************/
1260
1261             if (gmx_mm_any_lt(rsq00,rcutoff2))
1262             {
1263
1264             r00              = _mm_mul_ps(rsq00,rinv00);
1265             r00              = _mm_andnot_ps(dummy_mask,r00);
1266
1267             /* Compute parameters for interactions between i and j atoms */
1268             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1269                                          vdwparam+vdwioffset0+vdwjidx0B,
1270                                          vdwparam+vdwioffset0+vdwjidx0C,
1271                                          vdwparam+vdwioffset0+vdwjidx0D,
1272                                          &c6_00,&c12_00);
1273
1274             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1275                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1276                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1277                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1278
1279             /* Analytical LJ-PME */
1280             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1281             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1282             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1283             exponent         = gmx_simd_exp_r(ewcljrsq);
1284             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1285             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
1286             /* f6A = 6 * C6grid * (1 - poly) */
1287             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1288             /* f6B = C6grid * exponent * beta^6 */
1289             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1290             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1291             fvdw              = _mm_mul_ps(_mm_macc_ps(_mm_msub_ps(c12_00,rinvsix,_mm_sub_ps(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1292
1293             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1294
1295             fscal            = fvdw;
1296
1297             fscal            = _mm_and_ps(fscal,cutoff_mask);
1298
1299             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1300
1301              /* Update vectorial force */
1302             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1303             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1304             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1305
1306             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1307             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1308             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1309
1310             }
1311
1312             /**************************
1313              * CALCULATE INTERACTIONS *
1314              **************************/
1315
1316             if (gmx_mm_any_lt(rsq10,rcutoff2))
1317             {
1318
1319             r10              = _mm_mul_ps(rsq10,rinv10);
1320             r10              = _mm_andnot_ps(dummy_mask,r10);
1321
1322             /* Compute parameters for interactions between i and j atoms */
1323             qq10             = _mm_mul_ps(iq1,jq0);
1324
1325             /* EWALD ELECTROSTATICS */
1326
1327             /* Analytical PME correction */
1328             zeta2            = _mm_mul_ps(beta2,rsq10);
1329             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1330             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1331             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1332             felec            = _mm_mul_ps(qq10,felec);
1333
1334             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1335
1336             fscal            = felec;
1337
1338             fscal            = _mm_and_ps(fscal,cutoff_mask);
1339
1340             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1341
1342              /* Update vectorial force */
1343             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1344             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1345             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1346
1347             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1348             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1349             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1350
1351             }
1352
1353             /**************************
1354              * CALCULATE INTERACTIONS *
1355              **************************/
1356
1357             if (gmx_mm_any_lt(rsq20,rcutoff2))
1358             {
1359
1360             r20              = _mm_mul_ps(rsq20,rinv20);
1361             r20              = _mm_andnot_ps(dummy_mask,r20);
1362
1363             /* Compute parameters for interactions between i and j atoms */
1364             qq20             = _mm_mul_ps(iq2,jq0);
1365
1366             /* EWALD ELECTROSTATICS */
1367
1368             /* Analytical PME correction */
1369             zeta2            = _mm_mul_ps(beta2,rsq20);
1370             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1371             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1372             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1373             felec            = _mm_mul_ps(qq20,felec);
1374
1375             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1376
1377             fscal            = felec;
1378
1379             fscal            = _mm_and_ps(fscal,cutoff_mask);
1380
1381             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1382
1383              /* Update vectorial force */
1384             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1385             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1386             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1387
1388             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1389             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1390             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1391
1392             }
1393
1394             /**************************
1395              * CALCULATE INTERACTIONS *
1396              **************************/
1397
1398             if (gmx_mm_any_lt(rsq30,rcutoff2))
1399             {
1400
1401             r30              = _mm_mul_ps(rsq30,rinv30);
1402             r30              = _mm_andnot_ps(dummy_mask,r30);
1403
1404             /* Compute parameters for interactions between i and j atoms */
1405             qq30             = _mm_mul_ps(iq3,jq0);
1406
1407             /* EWALD ELECTROSTATICS */
1408
1409             /* Analytical PME correction */
1410             zeta2            = _mm_mul_ps(beta2,rsq30);
1411             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
1412             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1413             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1414             felec            = _mm_mul_ps(qq30,felec);
1415
1416             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1417
1418             fscal            = felec;
1419
1420             fscal            = _mm_and_ps(fscal,cutoff_mask);
1421
1422             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1423
1424              /* Update vectorial force */
1425             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1426             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1427             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1428
1429             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1430             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1431             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1432
1433             }
1434
1435             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1436             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1437             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1438             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1439
1440             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1441
1442             /* Inner loop uses 147 flops */
1443         }
1444
1445         /* End of innermost loop */
1446
1447         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1448                                               f+i_coord_offset,fshift+i_shift_offset);
1449
1450         /* Increment number of inner iterations */
1451         inneriter                  += j_index_end - j_index_start;
1452
1453         /* Outer loop uses 24 flops */
1454     }
1455
1456     /* Increment number of outer iterations */
1457     outeriter        += nri;
1458
1459     /* Update outer/inner flops */
1460
1461     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*147);
1462 }