src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_sse4_1_double.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 sse4_1_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_double.h"
  48 #include "kernelutil_x86_sse4_1_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_VF_sse4_1_double
  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_sse4_1_double
  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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwioffset3;
  87     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  94     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 101     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 102     __m128d           c6grid_00;
 103     __m128d           c6grid_10;
 104     __m128d           c6grid_20;
 105     __m128d           c6grid_30;
 106     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 107     real             *vdwgridparam;
 108     __m128d           one_half = _mm_set1_pd(0.5);
 109     __m128d           minus_one = _mm_set1_pd(-1.0);
 110     __m128i          ewitab;
 111     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 112     real             *ewtab;
 113     __m128d          dummy_mask,cutoff_mask;
 114     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 115     __m128d          one     = _mm_set1_pd(1.0);
 116     __m128d          two     = _mm_set1_pd(2.0);
 117     x                = xx[0];
 118     f                = ff[0];
 119
 120     nri              = nlist->nri;
 121     iinr             = nlist->iinr;
 122     jindex           = nlist->jindex;
 123     jjnr             = nlist->jjnr;
 124     shiftidx         = nlist->shift;
 125     gid              = nlist->gid;
 126     shiftvec         = fr->shift_vec[0];
 127     fshift           = fr->fshift[0];
 128     facel            = _mm_set1_pd(fr->epsfac);
 129     charge           = mdatoms->chargeA;
 130     nvdwtype         = fr->ntype;
 131     vdwparam         = fr->nbfp;
 132     vdwtype          = mdatoms->typeA;
 133     vdwgridparam     = fr->ljpme_c6grid;
 134     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 135     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 136     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 137
 138     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 139     ewtab            = fr->ic->tabq_coul_FDV0;
 140     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 141     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 142
 143     /* Setup water-specific parameters */
 144     inr              = nlist->iinr[0];
 145     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 146     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 147     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 148     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 149
 150     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 151     rcutoff_scalar   = fr->rcoulomb;
 152     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 153     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 154
 155     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 156     rvdw             = _mm_set1_pd(fr->rvdw);
 157
 158     /* Avoid stupid compiler warnings */
 159     jnrA = jnrB = 0;
 160     j_coord_offsetA = 0;
 161     j_coord_offsetB = 0;
 162
 163     outeriter        = 0;
 164     inneriter        = 0;
 165
 166     /* Start outer loop over neighborlists */
 167     for(iidx=0; iidx<nri; iidx++)
 168     {
 169         /* Load shift vector for this list */
 170         i_shift_offset   = DIM*shiftidx[iidx];
 171
 172         /* Load limits for loop over neighbors */
 173         j_index_start    = jindex[iidx];
 174         j_index_end      = jindex[iidx+1];
 175
 176         /* Get outer coordinate index */
 177         inr              = iinr[iidx];
 178         i_coord_offset   = DIM*inr;
 179
 180         /* Load i particle coords and add shift vector */
 181         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 182                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 183
 184         fix0             = _mm_setzero_pd();
 185         fiy0             = _mm_setzero_pd();
 186         fiz0             = _mm_setzero_pd();
 187         fix1             = _mm_setzero_pd();
 188         fiy1             = _mm_setzero_pd();
 189         fiz1             = _mm_setzero_pd();
 190         fix2             = _mm_setzero_pd();
 191         fiy2             = _mm_setzero_pd();
 192         fiz2             = _mm_setzero_pd();
 193         fix3             = _mm_setzero_pd();
 194         fiy3             = _mm_setzero_pd();
 195         fiz3             = _mm_setzero_pd();
 196
 197         /* Reset potential sums */
 198         velecsum         = _mm_setzero_pd();
 199         vvdwsum          = _mm_setzero_pd();
 200
 201         /* Start inner kernel loop */
 202         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 203         {
 204
 205             /* Get j neighbor index, and coordinate index */
 206             jnrA             = jjnr[jidx];
 207             jnrB             = jjnr[jidx+1];
 208             j_coord_offsetA  = DIM*jnrA;
 209             j_coord_offsetB  = DIM*jnrB;
 210
 211             /* load j atom coordinates */
 212             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 213                                               &jx0,&jy0,&jz0);
 214
 215             /* Calculate displacement vector */
 216             dx00             = _mm_sub_pd(ix0,jx0);
 217             dy00             = _mm_sub_pd(iy0,jy0);
 218             dz00             = _mm_sub_pd(iz0,jz0);
 219             dx10             = _mm_sub_pd(ix1,jx0);
 220             dy10             = _mm_sub_pd(iy1,jy0);
 221             dz10             = _mm_sub_pd(iz1,jz0);
 222             dx20             = _mm_sub_pd(ix2,jx0);
 223             dy20             = _mm_sub_pd(iy2,jy0);
 224             dz20             = _mm_sub_pd(iz2,jz0);
 225             dx30             = _mm_sub_pd(ix3,jx0);
 226             dy30             = _mm_sub_pd(iy3,jy0);
 227             dz30             = _mm_sub_pd(iz3,jz0);
 228
 229             /* Calculate squared distance and things based on it */
 230             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 231             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 232             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 233             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 234
 235             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 236             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 237             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 238             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 239
 240             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 241             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 242             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 243             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 244
 245             /* Load parameters for j particles */
 246             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 247             vdwjidx0A        = 2*vdwtype[jnrA+0];
 248             vdwjidx0B        = 2*vdwtype[jnrB+0];
 249
 250             fjx0             = _mm_setzero_pd();
 251             fjy0             = _mm_setzero_pd();
 252             fjz0             = _mm_setzero_pd();
 253
 254             /**************************
 255              * CALCULATE INTERACTIONS *
 256              **************************/
 257
 258             if (gmx_mm_any_lt(rsq00,rcutoff2))
 259             {
 260
 261             r00              = _mm_mul_pd(rsq00,rinv00);
 262
 263             /* Compute parameters for interactions between i and j atoms */
 264             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 265                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 266             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 267                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 268
 269             /* Analytical LJ-PME */
 270             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 271             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 272             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 273             exponent         = gmx_simd_exp_d(ewcljrsq);
 274             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 275             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 276             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 277             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 278             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 279             vvdw             = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))),one_twelfth),
 280                                _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_add_pd(_mm_mul_pd(c6_00,sh_vdw_invrcut6),_mm_mul_pd(c6grid_00,sh_lj_ewald))),one_sixth));
 281             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 282             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,_mm_sub_pd(vvdw6,_mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6)))),rinvsq00);
 283
 284             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 285
 286             /* Update potential sum for this i atom from the interaction with this j atom. */
 287             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 288             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 289
 290             fscal            = fvdw;
 291
 292             fscal            = _mm_and_pd(fscal,cutoff_mask);
 293
 294             /* Calculate temporary vectorial force */
 295             tx               = _mm_mul_pd(fscal,dx00);
 296             ty               = _mm_mul_pd(fscal,dy00);
 297             tz               = _mm_mul_pd(fscal,dz00);
 298
 299             /* Update vectorial force */
 300             fix0             = _mm_add_pd(fix0,tx);
 301             fiy0             = _mm_add_pd(fiy0,ty);
 302             fiz0             = _mm_add_pd(fiz0,tz);
 303
 304             fjx0             = _mm_add_pd(fjx0,tx);
 305             fjy0             = _mm_add_pd(fjy0,ty);
 306             fjz0             = _mm_add_pd(fjz0,tz);
 307
 308             }
 309
 310             /**************************
 311              * CALCULATE INTERACTIONS *
 312              **************************/
 313
 314             if (gmx_mm_any_lt(rsq10,rcutoff2))
 315             {
 316
 317             r10              = _mm_mul_pd(rsq10,rinv10);
 318
 319             /* Compute parameters for interactions between i and j atoms */
 320             qq10             = _mm_mul_pd(iq1,jq0);
 321
 322             /* EWALD ELECTROSTATICS */
 323
 324             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 325             ewrt             = _mm_mul_pd(r10,ewtabscale);
 326             ewitab           = _mm_cvttpd_epi32(ewrt);
 327             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 328             ewitab           = _mm_slli_epi32(ewitab,2);
 329             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 330             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 331             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 332             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 333             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 334             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 335             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 336             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 337             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 338             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 339
 340             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 341
 342             /* Update potential sum for this i atom from the interaction with this j atom. */
 343             velec            = _mm_and_pd(velec,cutoff_mask);
 344             velecsum         = _mm_add_pd(velecsum,velec);
 345
 346             fscal            = felec;
 347
 348             fscal            = _mm_and_pd(fscal,cutoff_mask);
 349
 350             /* Calculate temporary vectorial force */
 351             tx               = _mm_mul_pd(fscal,dx10);
 352             ty               = _mm_mul_pd(fscal,dy10);
 353             tz               = _mm_mul_pd(fscal,dz10);
 354
 355             /* Update vectorial force */
 356             fix1             = _mm_add_pd(fix1,tx);
 357             fiy1             = _mm_add_pd(fiy1,ty);
 358             fiz1             = _mm_add_pd(fiz1,tz);
 359
 360             fjx0             = _mm_add_pd(fjx0,tx);
 361             fjy0             = _mm_add_pd(fjy0,ty);
 362             fjz0             = _mm_add_pd(fjz0,tz);
 363
 364             }
 365
 366             /**************************
 367              * CALCULATE INTERACTIONS *
 368              **************************/
 369
 370             if (gmx_mm_any_lt(rsq20,rcutoff2))
 371             {
 372
 373             r20              = _mm_mul_pd(rsq20,rinv20);
 374
 375             /* Compute parameters for interactions between i and j atoms */
 376             qq20             = _mm_mul_pd(iq2,jq0);
 377
 378             /* EWALD ELECTROSTATICS */
 379
 380             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 381             ewrt             = _mm_mul_pd(r20,ewtabscale);
 382             ewitab           = _mm_cvttpd_epi32(ewrt);
 383             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 384             ewitab           = _mm_slli_epi32(ewitab,2);
 385             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 386             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 387             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 388             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 389             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 390             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 391             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 392             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 393             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 394             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 395
 396             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 397
 398             /* Update potential sum for this i atom from the interaction with this j atom. */
 399             velec            = _mm_and_pd(velec,cutoff_mask);
 400             velecsum         = _mm_add_pd(velecsum,velec);
 401
 402             fscal            = felec;
 403
 404             fscal            = _mm_and_pd(fscal,cutoff_mask);
 405
 406             /* Calculate temporary vectorial force */
 407             tx               = _mm_mul_pd(fscal,dx20);
 408             ty               = _mm_mul_pd(fscal,dy20);
 409             tz               = _mm_mul_pd(fscal,dz20);
 410
 411             /* Update vectorial force */
 412             fix2             = _mm_add_pd(fix2,tx);
 413             fiy2             = _mm_add_pd(fiy2,ty);
 414             fiz2             = _mm_add_pd(fiz2,tz);
 415
 416             fjx0             = _mm_add_pd(fjx0,tx);
 417             fjy0             = _mm_add_pd(fjy0,ty);
 418             fjz0             = _mm_add_pd(fjz0,tz);
 419
 420             }
 421
 422             /**************************
 423              * CALCULATE INTERACTIONS *
 424              **************************/
 425
 426             if (gmx_mm_any_lt(rsq30,rcutoff2))
 427             {
 428
 429             r30              = _mm_mul_pd(rsq30,rinv30);
 430
 431             /* Compute parameters for interactions between i and j atoms */
 432             qq30             = _mm_mul_pd(iq3,jq0);
 433
 434             /* EWALD ELECTROSTATICS */
 435
 436             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 437             ewrt             = _mm_mul_pd(r30,ewtabscale);
 438             ewitab           = _mm_cvttpd_epi32(ewrt);
 439             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 440             ewitab           = _mm_slli_epi32(ewitab,2);
 441             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 442             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 443             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 444             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 445             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 446             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 447             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 448             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 449             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_sub_pd(rinv30,sh_ewald),velec));
 450             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 451
 452             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 453
 454             /* Update potential sum for this i atom from the interaction with this j atom. */
 455             velec            = _mm_and_pd(velec,cutoff_mask);
 456             velecsum         = _mm_add_pd(velecsum,velec);
 457
 458             fscal            = felec;
 459
 460             fscal            = _mm_and_pd(fscal,cutoff_mask);
 461
 462             /* Calculate temporary vectorial force */
 463             tx               = _mm_mul_pd(fscal,dx30);
 464             ty               = _mm_mul_pd(fscal,dy30);
 465             tz               = _mm_mul_pd(fscal,dz30);
 466
 467             /* Update vectorial force */
 468             fix3             = _mm_add_pd(fix3,tx);
 469             fiy3             = _mm_add_pd(fiy3,ty);
 470             fiz3             = _mm_add_pd(fiz3,tz);
 471
 472             fjx0             = _mm_add_pd(fjx0,tx);
 473             fjy0             = _mm_add_pd(fjy0,ty);
 474             fjz0             = _mm_add_pd(fjz0,tz);
 475
 476             }
 477
 478             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 479
 480             /* Inner loop uses 202 flops */
 481         }
 482
 483         if(jidx<j_index_end)
 484         {
 485
 486             jnrA             = jjnr[jidx];
 487             j_coord_offsetA  = DIM*jnrA;
 488
 489             /* load j atom coordinates */
 490             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 491                                               &jx0,&jy0,&jz0);
 492
 493             /* Calculate displacement vector */
 494             dx00             = _mm_sub_pd(ix0,jx0);
 495             dy00             = _mm_sub_pd(iy0,jy0);
 496             dz00             = _mm_sub_pd(iz0,jz0);
 497             dx10             = _mm_sub_pd(ix1,jx0);
 498             dy10             = _mm_sub_pd(iy1,jy0);
 499             dz10             = _mm_sub_pd(iz1,jz0);
 500             dx20             = _mm_sub_pd(ix2,jx0);
 501             dy20             = _mm_sub_pd(iy2,jy0);
 502             dz20             = _mm_sub_pd(iz2,jz0);
 503             dx30             = _mm_sub_pd(ix3,jx0);
 504             dy30             = _mm_sub_pd(iy3,jy0);
 505             dz30             = _mm_sub_pd(iz3,jz0);
 506
 507             /* Calculate squared distance and things based on it */
 508             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 509             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 510             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 511             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 512
 513             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 514             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 515             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 516             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 517
 518             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 519             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 520             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 521             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 522
 523             /* Load parameters for j particles */
 524             jq0              = _mm_load_sd(charge+jnrA+0);
 525             vdwjidx0A        = 2*vdwtype[jnrA+0];
 526
 527             fjx0             = _mm_setzero_pd();
 528             fjy0             = _mm_setzero_pd();
 529             fjz0             = _mm_setzero_pd();
 530
 531             /**************************
 532              * CALCULATE INTERACTIONS *
 533              **************************/
 534
 535             if (gmx_mm_any_lt(rsq00,rcutoff2))
 536             {
 537
 538             r00              = _mm_mul_pd(rsq00,rinv00);
 539
 540             /* Compute parameters for interactions between i and j atoms */
 541             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 542
 543             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 544
 545             /* Analytical LJ-PME */
 546             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 547             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 548             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 549             exponent         = gmx_simd_exp_d(ewcljrsq);
 550             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 551             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 552             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 553             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 554             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 555             vvdw             = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))),one_twelfth),
 556                                _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_add_pd(_mm_mul_pd(c6_00,sh_vdw_invrcut6),_mm_mul_pd(c6grid_00,sh_lj_ewald))),one_sixth));
 557             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 558             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,_mm_sub_pd(vvdw6,_mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6)))),rinvsq00);
 559
 560             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 561
 562             /* Update potential sum for this i atom from the interaction with this j atom. */
 563             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 564             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 565             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 566
 567             fscal            = fvdw;
 568
 569             fscal            = _mm_and_pd(fscal,cutoff_mask);
 570
 571             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 572
 573             /* Calculate temporary vectorial force */
 574             tx               = _mm_mul_pd(fscal,dx00);
 575             ty               = _mm_mul_pd(fscal,dy00);
 576             tz               = _mm_mul_pd(fscal,dz00);
 577
 578             /* Update vectorial force */
 579             fix0             = _mm_add_pd(fix0,tx);
 580             fiy0             = _mm_add_pd(fiy0,ty);
 581             fiz0             = _mm_add_pd(fiz0,tz);
 582
 583             fjx0             = _mm_add_pd(fjx0,tx);
 584             fjy0             = _mm_add_pd(fjy0,ty);
 585             fjz0             = _mm_add_pd(fjz0,tz);
 586
 587             }
 588
 589             /**************************
 590              * CALCULATE INTERACTIONS *
 591              **************************/
 592
 593             if (gmx_mm_any_lt(rsq10,rcutoff2))
 594             {
 595
 596             r10              = _mm_mul_pd(rsq10,rinv10);
 597
 598             /* Compute parameters for interactions between i and j atoms */
 599             qq10             = _mm_mul_pd(iq1,jq0);
 600
 601             /* EWALD ELECTROSTATICS */
 602
 603             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 604             ewrt             = _mm_mul_pd(r10,ewtabscale);
 605             ewitab           = _mm_cvttpd_epi32(ewrt);
 606             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 607             ewitab           = _mm_slli_epi32(ewitab,2);
 608             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 609             ewtabD           = _mm_setzero_pd();
 610             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 611             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 612             ewtabFn          = _mm_setzero_pd();
 613             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 614             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 615             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 616             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 617             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 618
 619             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 620
 621             /* Update potential sum for this i atom from the interaction with this j atom. */
 622             velec            = _mm_and_pd(velec,cutoff_mask);
 623             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 624             velecsum         = _mm_add_pd(velecsum,velec);
 625
 626             fscal            = felec;
 627
 628             fscal            = _mm_and_pd(fscal,cutoff_mask);
 629
 630             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 631
 632             /* Calculate temporary vectorial force */
 633             tx               = _mm_mul_pd(fscal,dx10);
 634             ty               = _mm_mul_pd(fscal,dy10);
 635             tz               = _mm_mul_pd(fscal,dz10);
 636
 637             /* Update vectorial force */
 638             fix1             = _mm_add_pd(fix1,tx);
 639             fiy1             = _mm_add_pd(fiy1,ty);
 640             fiz1             = _mm_add_pd(fiz1,tz);
 641
 642             fjx0             = _mm_add_pd(fjx0,tx);
 643             fjy0             = _mm_add_pd(fjy0,ty);
 644             fjz0             = _mm_add_pd(fjz0,tz);
 645
 646             }
 647
 648             /**************************
 649              * CALCULATE INTERACTIONS *
 650              **************************/
 651
 652             if (gmx_mm_any_lt(rsq20,rcutoff2))
 653             {
 654
 655             r20              = _mm_mul_pd(rsq20,rinv20);
 656
 657             /* Compute parameters for interactions between i and j atoms */
 658             qq20             = _mm_mul_pd(iq2,jq0);
 659
 660             /* EWALD ELECTROSTATICS */
 661
 662             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 663             ewrt             = _mm_mul_pd(r20,ewtabscale);
 664             ewitab           = _mm_cvttpd_epi32(ewrt);
 665             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 666             ewitab           = _mm_slli_epi32(ewitab,2);
 667             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 668             ewtabD           = _mm_setzero_pd();
 669             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 670             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 671             ewtabFn          = _mm_setzero_pd();
 672             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 673             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 674             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 675             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 676             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 677
 678             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 679
 680             /* Update potential sum for this i atom from the interaction with this j atom. */
 681             velec            = _mm_and_pd(velec,cutoff_mask);
 682             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 683             velecsum         = _mm_add_pd(velecsum,velec);
 684
 685             fscal            = felec;
 686
 687             fscal            = _mm_and_pd(fscal,cutoff_mask);
 688
 689             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 690
 691             /* Calculate temporary vectorial force */
 692             tx               = _mm_mul_pd(fscal,dx20);
 693             ty               = _mm_mul_pd(fscal,dy20);
 694             tz               = _mm_mul_pd(fscal,dz20);
 695
 696             /* Update vectorial force */
 697             fix2             = _mm_add_pd(fix2,tx);
 698             fiy2             = _mm_add_pd(fiy2,ty);
 699             fiz2             = _mm_add_pd(fiz2,tz);
 700
 701             fjx0             = _mm_add_pd(fjx0,tx);
 702             fjy0             = _mm_add_pd(fjy0,ty);
 703             fjz0             = _mm_add_pd(fjz0,tz);
 704
 705             }
 706
 707             /**************************
 708              * CALCULATE INTERACTIONS *
 709              **************************/
 710
 711             if (gmx_mm_any_lt(rsq30,rcutoff2))
 712             {
 713
 714             r30              = _mm_mul_pd(rsq30,rinv30);
 715
 716             /* Compute parameters for interactions between i and j atoms */
 717             qq30             = _mm_mul_pd(iq3,jq0);
 718
 719             /* EWALD ELECTROSTATICS */
 720
 721             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 722             ewrt             = _mm_mul_pd(r30,ewtabscale);
 723             ewitab           = _mm_cvttpd_epi32(ewrt);
 724             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 725             ewitab           = _mm_slli_epi32(ewitab,2);
 726             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 727             ewtabD           = _mm_setzero_pd();
 728             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 729             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 730             ewtabFn          = _mm_setzero_pd();
 731             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 732             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 733             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 734             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_sub_pd(rinv30,sh_ewald),velec));
 735             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 736
 737             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 738
 739             /* Update potential sum for this i atom from the interaction with this j atom. */
 740             velec            = _mm_and_pd(velec,cutoff_mask);
 741             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 742             velecsum         = _mm_add_pd(velecsum,velec);
 743
 744             fscal            = felec;
 745
 746             fscal            = _mm_and_pd(fscal,cutoff_mask);
 747
 748             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 749
 750             /* Calculate temporary vectorial force */
 751             tx               = _mm_mul_pd(fscal,dx30);
 752             ty               = _mm_mul_pd(fscal,dy30);
 753             tz               = _mm_mul_pd(fscal,dz30);
 754
 755             /* Update vectorial force */
 756             fix3             = _mm_add_pd(fix3,tx);
 757             fiy3             = _mm_add_pd(fiy3,ty);
 758             fiz3             = _mm_add_pd(fiz3,tz);
 759
 760             fjx0             = _mm_add_pd(fjx0,tx);
 761             fjy0             = _mm_add_pd(fjy0,ty);
 762             fjz0             = _mm_add_pd(fjz0,tz);
 763
 764             }
 765
 766             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 767
 768             /* Inner loop uses 202 flops */
 769         }
 770
 771         /* End of innermost loop */
 772
 773         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 774                                               f+i_coord_offset,fshift+i_shift_offset);
 775
 776         ggid                        = gid[iidx];
 777         /* Update potential energies */
 778         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 779         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 780
 781         /* Increment number of inner iterations */
 782         inneriter                  += j_index_end - j_index_start;
 783
 784         /* Outer loop uses 26 flops */
 785     }
 786
 787     /* Increment number of outer iterations */
 788     outeriter        += nri;
 789
 790     /* Update outer/inner flops */
 791
 792     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*202);
 793 }
 794 /*
 795  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_sse4_1_double
 796  * Electrostatics interaction: Ewald
 797  * VdW interaction:            LJEwald
 798  * Geometry:                   Water4-Particle
 799  * Calculate force/pot:        Force
 800  */
 801 void
 802 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_sse4_1_double
 803                     (t_nblist                    * gmx_restrict       nlist,
 804                      rvec                        * gmx_restrict          xx,
 805                      rvec                        * gmx_restrict          ff,
 806                      t_forcerec                  * gmx_restrict          fr,
 807                      t_mdatoms                   * gmx_restrict     mdatoms,
 808                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 809                      t_nrnb                      * gmx_restrict        nrnb)
 810 {
 811     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 812      * just 0 for non-waters.
 813      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 814      * jnr indices corresponding to data put in the four positions in the SIMD register.
 815      */
 816     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 817     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 818     int              jnrA,jnrB;
 819     int              j_coord_offsetA,j_coord_offsetB;
 820     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 821     real             rcutoff_scalar;
 822     real             *shiftvec,*fshift,*x,*f;
 823     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 824     int              vdwioffset0;
 825     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 826     int              vdwioffset1;
 827     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 828     int              vdwioffset2;
 829     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 830     int              vdwioffset3;
 831     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 832     int              vdwjidx0A,vdwjidx0B;
 833     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 834     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 835     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 836     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 837     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 838     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 839     real             *charge;
 840     int              nvdwtype;
 841     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 842     int              *vdwtype;
 843     real             *vdwparam;
 844     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 845     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 846     __m128d           c6grid_00;
 847     __m128d           c6grid_10;
 848     __m128d           c6grid_20;
 849     __m128d           c6grid_30;
 850     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 851     real             *vdwgridparam;
 852     __m128d           one_half = _mm_set1_pd(0.5);
 853     __m128d           minus_one = _mm_set1_pd(-1.0);
 854     __m128i          ewitab;
 855     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 856     real             *ewtab;
 857     __m128d          dummy_mask,cutoff_mask;
 858     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 859     __m128d          one     = _mm_set1_pd(1.0);
 860     __m128d          two     = _mm_set1_pd(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_pd(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_pd(fr->ic->sh_lj_ewald);
 879     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 880     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 881
 882     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 883     ewtab            = fr->ic->tabq_coul_F;
 884     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 885     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 886
 887     /* Setup water-specific parameters */
 888     inr              = nlist->iinr[0];
 889     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 890     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 891     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 892     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 893
 894     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 895     rcutoff_scalar   = fr->rcoulomb;
 896     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 897     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 898
 899     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 900     rvdw             = _mm_set1_pd(fr->rvdw);
 901
 902     /* Avoid stupid compiler warnings */
 903     jnrA = jnrB = 0;
 904     j_coord_offsetA = 0;
 905     j_coord_offsetB = 0;
 906
 907     outeriter        = 0;
 908     inneriter        = 0;
 909
 910     /* Start outer loop over neighborlists */
 911     for(iidx=0; iidx<nri; iidx++)
 912     {
 913         /* Load shift vector for this list */
 914         i_shift_offset   = DIM*shiftidx[iidx];
 915
 916         /* Load limits for loop over neighbors */
 917         j_index_start    = jindex[iidx];
 918         j_index_end      = jindex[iidx+1];
 919
 920         /* Get outer coordinate index */
 921         inr              = iinr[iidx];
 922         i_coord_offset   = DIM*inr;
 923
 924         /* Load i particle coords and add shift vector */
 925         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 926                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 927
 928         fix0             = _mm_setzero_pd();
 929         fiy0             = _mm_setzero_pd();
 930         fiz0             = _mm_setzero_pd();
 931         fix1             = _mm_setzero_pd();
 932         fiy1             = _mm_setzero_pd();
 933         fiz1             = _mm_setzero_pd();
 934         fix2             = _mm_setzero_pd();
 935         fiy2             = _mm_setzero_pd();
 936         fiz2             = _mm_setzero_pd();
 937         fix3             = _mm_setzero_pd();
 938         fiy3             = _mm_setzero_pd();
 939         fiz3             = _mm_setzero_pd();
 940
 941         /* Start inner kernel loop */
 942         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 943         {
 944
 945             /* Get j neighbor index, and coordinate index */
 946             jnrA             = jjnr[jidx];
 947             jnrB             = jjnr[jidx+1];
 948             j_coord_offsetA  = DIM*jnrA;
 949             j_coord_offsetB  = DIM*jnrB;
 950
 951             /* load j atom coordinates */
 952             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 953                                               &jx0,&jy0,&jz0);
 954
 955             /* Calculate displacement vector */
 956             dx00             = _mm_sub_pd(ix0,jx0);
 957             dy00             = _mm_sub_pd(iy0,jy0);
 958             dz00             = _mm_sub_pd(iz0,jz0);
 959             dx10             = _mm_sub_pd(ix1,jx0);
 960             dy10             = _mm_sub_pd(iy1,jy0);
 961             dz10             = _mm_sub_pd(iz1,jz0);
 962             dx20             = _mm_sub_pd(ix2,jx0);
 963             dy20             = _mm_sub_pd(iy2,jy0);
 964             dz20             = _mm_sub_pd(iz2,jz0);
 965             dx30             = _mm_sub_pd(ix3,jx0);
 966             dy30             = _mm_sub_pd(iy3,jy0);
 967             dz30             = _mm_sub_pd(iz3,jz0);
 968
 969             /* Calculate squared distance and things based on it */
 970             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 971             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 972             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 973             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 974
 975             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 976             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 977             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 978             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 979
 980             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 981             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 982             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 983             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 984
 985             /* Load parameters for j particles */
 986             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 987             vdwjidx0A        = 2*vdwtype[jnrA+0];
 988             vdwjidx0B        = 2*vdwtype[jnrB+0];
 989
 990             fjx0             = _mm_setzero_pd();
 991             fjy0             = _mm_setzero_pd();
 992             fjz0             = _mm_setzero_pd();
 993
 994             /**************************
 995              * CALCULATE INTERACTIONS *
 996              **************************/
 997
 998             if (gmx_mm_any_lt(rsq00,rcutoff2))
 999             {
1000
1001             r00              = _mm_mul_pd(rsq00,rinv00);
1002
1003             /* Compute parameters for interactions between i and j atoms */
1004             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
1005                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
1006             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
1007                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
1008
1009             /* Analytical LJ-PME */
1010             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1011             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1012             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1013             exponent         = gmx_simd_exp_d(ewcljrsq);
1014             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1015             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1016             /* f6A = 6 * C6grid * (1 - poly) */
1017             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1018             /* f6B = C6grid * exponent * beta^6 */
1019             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1020             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1021             fvdw              = _mm_mul_pd(_mm_add_pd(_mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),_mm_sub_pd(c6_00,f6A)),rinvsix),f6B),rinvsq00);
1022
1023             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1024
1025             fscal            = fvdw;
1026
1027             fscal            = _mm_and_pd(fscal,cutoff_mask);
1028
1029             /* Calculate temporary vectorial force */
1030             tx               = _mm_mul_pd(fscal,dx00);
1031             ty               = _mm_mul_pd(fscal,dy00);
1032             tz               = _mm_mul_pd(fscal,dz00);
1033
1034             /* Update vectorial force */
1035             fix0             = _mm_add_pd(fix0,tx);
1036             fiy0             = _mm_add_pd(fiy0,ty);
1037             fiz0             = _mm_add_pd(fiz0,tz);
1038
1039             fjx0             = _mm_add_pd(fjx0,tx);
1040             fjy0             = _mm_add_pd(fjy0,ty);
1041             fjz0             = _mm_add_pd(fjz0,tz);
1042
1043             }
1044
1045             /**************************
1046              * CALCULATE INTERACTIONS *
1047              **************************/
1048
1049             if (gmx_mm_any_lt(rsq10,rcutoff2))
1050             {
1051
1052             r10              = _mm_mul_pd(rsq10,rinv10);
1053
1054             /* Compute parameters for interactions between i and j atoms */
1055             qq10             = _mm_mul_pd(iq1,jq0);
1056
1057             /* EWALD ELECTROSTATICS */
1058
1059             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1060             ewrt             = _mm_mul_pd(r10,ewtabscale);
1061             ewitab           = _mm_cvttpd_epi32(ewrt);
1062             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1063             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1064                                          &ewtabF,&ewtabFn);
1065             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1066             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1067
1068             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1069
1070             fscal            = felec;
1071
1072             fscal            = _mm_and_pd(fscal,cutoff_mask);
1073
1074             /* Calculate temporary vectorial force */
1075             tx               = _mm_mul_pd(fscal,dx10);
1076             ty               = _mm_mul_pd(fscal,dy10);
1077             tz               = _mm_mul_pd(fscal,dz10);
1078
1079             /* Update vectorial force */
1080             fix1             = _mm_add_pd(fix1,tx);
1081             fiy1             = _mm_add_pd(fiy1,ty);
1082             fiz1             = _mm_add_pd(fiz1,tz);
1083
1084             fjx0             = _mm_add_pd(fjx0,tx);
1085             fjy0             = _mm_add_pd(fjy0,ty);
1086             fjz0             = _mm_add_pd(fjz0,tz);
1087
1088             }
1089
1090             /**************************
1091              * CALCULATE INTERACTIONS *
1092              **************************/
1093
1094             if (gmx_mm_any_lt(rsq20,rcutoff2))
1095             {
1096
1097             r20              = _mm_mul_pd(rsq20,rinv20);
1098
1099             /* Compute parameters for interactions between i and j atoms */
1100             qq20             = _mm_mul_pd(iq2,jq0);
1101
1102             /* EWALD ELECTROSTATICS */
1103
1104             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1105             ewrt             = _mm_mul_pd(r20,ewtabscale);
1106             ewitab           = _mm_cvttpd_epi32(ewrt);
1107             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1108             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1109                                          &ewtabF,&ewtabFn);
1110             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1111             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1112
1113             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1114
1115             fscal            = felec;
1116
1117             fscal            = _mm_and_pd(fscal,cutoff_mask);
1118
1119             /* Calculate temporary vectorial force */
1120             tx               = _mm_mul_pd(fscal,dx20);
1121             ty               = _mm_mul_pd(fscal,dy20);
1122             tz               = _mm_mul_pd(fscal,dz20);
1123
1124             /* Update vectorial force */
1125             fix2             = _mm_add_pd(fix2,tx);
1126             fiy2             = _mm_add_pd(fiy2,ty);
1127             fiz2             = _mm_add_pd(fiz2,tz);
1128
1129             fjx0             = _mm_add_pd(fjx0,tx);
1130             fjy0             = _mm_add_pd(fjy0,ty);
1131             fjz0             = _mm_add_pd(fjz0,tz);
1132
1133             }
1134
1135             /**************************
1136              * CALCULATE INTERACTIONS *
1137              **************************/
1138
1139             if (gmx_mm_any_lt(rsq30,rcutoff2))
1140             {
1141
1142             r30              = _mm_mul_pd(rsq30,rinv30);
1143
1144             /* Compute parameters for interactions between i and j atoms */
1145             qq30             = _mm_mul_pd(iq3,jq0);
1146
1147             /* EWALD ELECTROSTATICS */
1148
1149             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1150             ewrt             = _mm_mul_pd(r30,ewtabscale);
1151             ewitab           = _mm_cvttpd_epi32(ewrt);
1152             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1153             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1154                                          &ewtabF,&ewtabFn);
1155             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1156             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1157
1158             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1159
1160             fscal            = felec;
1161
1162             fscal            = _mm_and_pd(fscal,cutoff_mask);
1163
1164             /* Calculate temporary vectorial force */
1165             tx               = _mm_mul_pd(fscal,dx30);
1166             ty               = _mm_mul_pd(fscal,dy30);
1167             tz               = _mm_mul_pd(fscal,dz30);
1168
1169             /* Update vectorial force */
1170             fix3             = _mm_add_pd(fix3,tx);
1171             fiy3             = _mm_add_pd(fiy3,ty);
1172             fiz3             = _mm_add_pd(fiz3,tz);
1173
1174             fjx0             = _mm_add_pd(fjx0,tx);
1175             fjy0             = _mm_add_pd(fjy0,ty);
1176             fjz0             = _mm_add_pd(fjz0,tz);
1177
1178             }
1179
1180             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1181
1182             /* Inner loop uses 169 flops */
1183         }
1184
1185         if(jidx<j_index_end)
1186         {
1187
1188             jnrA             = jjnr[jidx];
1189             j_coord_offsetA  = DIM*jnrA;
1190
1191             /* load j atom coordinates */
1192             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1193                                               &jx0,&jy0,&jz0);
1194
1195             /* Calculate displacement vector */
1196             dx00             = _mm_sub_pd(ix0,jx0);
1197             dy00             = _mm_sub_pd(iy0,jy0);
1198             dz00             = _mm_sub_pd(iz0,jz0);
1199             dx10             = _mm_sub_pd(ix1,jx0);
1200             dy10             = _mm_sub_pd(iy1,jy0);
1201             dz10             = _mm_sub_pd(iz1,jz0);
1202             dx20             = _mm_sub_pd(ix2,jx0);
1203             dy20             = _mm_sub_pd(iy2,jy0);
1204             dz20             = _mm_sub_pd(iz2,jz0);
1205             dx30             = _mm_sub_pd(ix3,jx0);
1206             dy30             = _mm_sub_pd(iy3,jy0);
1207             dz30             = _mm_sub_pd(iz3,jz0);
1208
1209             /* Calculate squared distance and things based on it */
1210             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1211             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1212             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1213             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1214
1215             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1216             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1217             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1218             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1219
1220             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1221             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1222             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1223             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1224
1225             /* Load parameters for j particles */
1226             jq0              = _mm_load_sd(charge+jnrA+0);
1227             vdwjidx0A        = 2*vdwtype[jnrA+0];
1228
1229             fjx0             = _mm_setzero_pd();
1230             fjy0             = _mm_setzero_pd();
1231             fjz0             = _mm_setzero_pd();
1232
1233             /**************************
1234              * CALCULATE INTERACTIONS *
1235              **************************/
1236
1237             if (gmx_mm_any_lt(rsq00,rcutoff2))
1238             {
1239
1240             r00              = _mm_mul_pd(rsq00,rinv00);
1241
1242             /* Compute parameters for interactions between i and j atoms */
1243             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1244
1245             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
1246
1247             /* Analytical LJ-PME */
1248             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1249             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1250             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1251             exponent         = gmx_simd_exp_d(ewcljrsq);
1252             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1253             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1254             /* f6A = 6 * C6grid * (1 - poly) */
1255             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1256             /* f6B = C6grid * exponent * beta^6 */
1257             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1258             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1259             fvdw              = _mm_mul_pd(_mm_add_pd(_mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),_mm_sub_pd(c6_00,f6A)),rinvsix),f6B),rinvsq00);
1260
1261             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1262
1263             fscal            = fvdw;
1264
1265             fscal            = _mm_and_pd(fscal,cutoff_mask);
1266
1267             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1268
1269             /* Calculate temporary vectorial force */
1270             tx               = _mm_mul_pd(fscal,dx00);
1271             ty               = _mm_mul_pd(fscal,dy00);
1272             tz               = _mm_mul_pd(fscal,dz00);
1273
1274             /* Update vectorial force */
1275             fix0             = _mm_add_pd(fix0,tx);
1276             fiy0             = _mm_add_pd(fiy0,ty);
1277             fiz0             = _mm_add_pd(fiz0,tz);
1278
1279             fjx0             = _mm_add_pd(fjx0,tx);
1280             fjy0             = _mm_add_pd(fjy0,ty);
1281             fjz0             = _mm_add_pd(fjz0,tz);
1282
1283             }
1284
1285             /**************************
1286              * CALCULATE INTERACTIONS *
1287              **************************/
1288
1289             if (gmx_mm_any_lt(rsq10,rcutoff2))
1290             {
1291
1292             r10              = _mm_mul_pd(rsq10,rinv10);
1293
1294             /* Compute parameters for interactions between i and j atoms */
1295             qq10             = _mm_mul_pd(iq1,jq0);
1296
1297             /* EWALD ELECTROSTATICS */
1298
1299             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1300             ewrt             = _mm_mul_pd(r10,ewtabscale);
1301             ewitab           = _mm_cvttpd_epi32(ewrt);
1302             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1303             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1304             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1305             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1306
1307             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1308
1309             fscal            = felec;
1310
1311             fscal            = _mm_and_pd(fscal,cutoff_mask);
1312
1313             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1314
1315             /* Calculate temporary vectorial force */
1316             tx               = _mm_mul_pd(fscal,dx10);
1317             ty               = _mm_mul_pd(fscal,dy10);
1318             tz               = _mm_mul_pd(fscal,dz10);
1319
1320             /* Update vectorial force */
1321             fix1             = _mm_add_pd(fix1,tx);
1322             fiy1             = _mm_add_pd(fiy1,ty);
1323             fiz1             = _mm_add_pd(fiz1,tz);
1324
1325             fjx0             = _mm_add_pd(fjx0,tx);
1326             fjy0             = _mm_add_pd(fjy0,ty);
1327             fjz0             = _mm_add_pd(fjz0,tz);
1328
1329             }
1330
1331             /**************************
1332              * CALCULATE INTERACTIONS *
1333              **************************/
1334
1335             if (gmx_mm_any_lt(rsq20,rcutoff2))
1336             {
1337
1338             r20              = _mm_mul_pd(rsq20,rinv20);
1339
1340             /* Compute parameters for interactions between i and j atoms */
1341             qq20             = _mm_mul_pd(iq2,jq0);
1342
1343             /* EWALD ELECTROSTATICS */
1344
1345             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1346             ewrt             = _mm_mul_pd(r20,ewtabscale);
1347             ewitab           = _mm_cvttpd_epi32(ewrt);
1348             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1349             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1350             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1351             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1352
1353             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1354
1355             fscal            = felec;
1356
1357             fscal            = _mm_and_pd(fscal,cutoff_mask);
1358
1359             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1360
1361             /* Calculate temporary vectorial force */
1362             tx               = _mm_mul_pd(fscal,dx20);
1363             ty               = _mm_mul_pd(fscal,dy20);
1364             tz               = _mm_mul_pd(fscal,dz20);
1365
1366             /* Update vectorial force */
1367             fix2             = _mm_add_pd(fix2,tx);
1368             fiy2             = _mm_add_pd(fiy2,ty);
1369             fiz2             = _mm_add_pd(fiz2,tz);
1370
1371             fjx0             = _mm_add_pd(fjx0,tx);
1372             fjy0             = _mm_add_pd(fjy0,ty);
1373             fjz0             = _mm_add_pd(fjz0,tz);
1374
1375             }
1376
1377             /**************************
1378              * CALCULATE INTERACTIONS *
1379              **************************/
1380
1381             if (gmx_mm_any_lt(rsq30,rcutoff2))
1382             {
1383
1384             r30              = _mm_mul_pd(rsq30,rinv30);
1385
1386             /* Compute parameters for interactions between i and j atoms */
1387             qq30             = _mm_mul_pd(iq3,jq0);
1388
1389             /* EWALD ELECTROSTATICS */
1390
1391             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1392             ewrt             = _mm_mul_pd(r30,ewtabscale);
1393             ewitab           = _mm_cvttpd_epi32(ewrt);
1394             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1395             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1396             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1397             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1398
1399             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1400
1401             fscal            = felec;
1402
1403             fscal            = _mm_and_pd(fscal,cutoff_mask);
1404
1405             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1406
1407             /* Calculate temporary vectorial force */
1408             tx               = _mm_mul_pd(fscal,dx30);
1409             ty               = _mm_mul_pd(fscal,dy30);
1410             tz               = _mm_mul_pd(fscal,dz30);
1411
1412             /* Update vectorial force */
1413             fix3             = _mm_add_pd(fix3,tx);
1414             fiy3             = _mm_add_pd(fiy3,ty);
1415             fiz3             = _mm_add_pd(fiz3,tz);
1416
1417             fjx0             = _mm_add_pd(fjx0,tx);
1418             fjy0             = _mm_add_pd(fjy0,ty);
1419             fjz0             = _mm_add_pd(fjz0,tz);
1420
1421             }
1422
1423             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1424
1425             /* Inner loop uses 169 flops */
1426         }
1427
1428         /* End of innermost loop */
1429
1430         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1431                                               f+i_coord_offset,fshift+i_shift_offset);
1432
1433         /* Increment number of inner iterations */
1434         inneriter                  += j_index_end - j_index_start;
1435
1436         /* Outer loop uses 24 flops */
1437     }
1438
1439     /* Increment number of outer iterations */
1440     outeriter        += nri;
1441
1442     /* Update outer/inner flops */
1443
1444     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*169);
1445 }