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 "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_sse4_1_double.h"
  50 #include "kernelutil_x86_sse4_1_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_VF_sse4_1_double
  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_sse4_1_double
  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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwioffset3;
  89     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B;
  91     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 103     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 104     __m128d           c6grid_00;
 105     __m128d           c6grid_10;
 106     __m128d           c6grid_20;
 107     __m128d           c6grid_30;
 108     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 109     real             *vdwgridparam;
 110     __m128d           one_half = _mm_set1_pd(0.5);
 111     __m128d           minus_one = _mm_set1_pd(-1.0);
 112     __m128i          ewitab;
 113     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 114     real             *ewtab;
 115     __m128d          dummy_mask,cutoff_mask;
 116     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 117     __m128d          one     = _mm_set1_pd(1.0);
 118     __m128d          two     = _mm_set1_pd(2.0);
 119     x                = xx[0];
 120     f                = ff[0];
 121
 122     nri              = nlist->nri;
 123     iinr             = nlist->iinr;
 124     jindex           = nlist->jindex;
 125     jjnr             = nlist->jjnr;
 126     shiftidx         = nlist->shift;
 127     gid              = nlist->gid;
 128     shiftvec         = fr->shift_vec[0];
 129     fshift           = fr->fshift[0];
 130     facel            = _mm_set1_pd(fr->epsfac);
 131     charge           = mdatoms->chargeA;
 132     nvdwtype         = fr->ntype;
 133     vdwparam         = fr->nbfp;
 134     vdwtype          = mdatoms->typeA;
 135     vdwgridparam     = fr->ljpme_c6grid;
 136     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 137     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 138     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 139
 140     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 141     ewtab            = fr->ic->tabq_coul_FDV0;
 142     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 143     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 144
 145     /* Setup water-specific parameters */
 146     inr              = nlist->iinr[0];
 147     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 148     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 149     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 150     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 151
 152     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 153     rcutoff_scalar   = fr->rcoulomb;
 154     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 155     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 156
 157     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 158     rvdw             = _mm_set1_pd(fr->rvdw);
 159
 160     /* Avoid stupid compiler warnings */
 161     jnrA = jnrB = 0;
 162     j_coord_offsetA = 0;
 163     j_coord_offsetB = 0;
 164
 165     outeriter        = 0;
 166     inneriter        = 0;
 167
 168     /* Start outer loop over neighborlists */
 169     for(iidx=0; iidx<nri; iidx++)
 170     {
 171         /* Load shift vector for this list */
 172         i_shift_offset   = DIM*shiftidx[iidx];
 173
 174         /* Load limits for loop over neighbors */
 175         j_index_start    = jindex[iidx];
 176         j_index_end      = jindex[iidx+1];
 177
 178         /* Get outer coordinate index */
 179         inr              = iinr[iidx];
 180         i_coord_offset   = DIM*inr;
 181
 182         /* Load i particle coords and add shift vector */
 183         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 184                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 185
 186         fix0             = _mm_setzero_pd();
 187         fiy0             = _mm_setzero_pd();
 188         fiz0             = _mm_setzero_pd();
 189         fix1             = _mm_setzero_pd();
 190         fiy1             = _mm_setzero_pd();
 191         fiz1             = _mm_setzero_pd();
 192         fix2             = _mm_setzero_pd();
 193         fiy2             = _mm_setzero_pd();
 194         fiz2             = _mm_setzero_pd();
 195         fix3             = _mm_setzero_pd();
 196         fiy3             = _mm_setzero_pd();
 197         fiz3             = _mm_setzero_pd();
 198
 199         /* Reset potential sums */
 200         velecsum         = _mm_setzero_pd();
 201         vvdwsum          = _mm_setzero_pd();
 202
 203         /* Start inner kernel loop */
 204         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 205         {
 206
 207             /* Get j neighbor index, and coordinate index */
 208             jnrA             = jjnr[jidx];
 209             jnrB             = jjnr[jidx+1];
 210             j_coord_offsetA  = DIM*jnrA;
 211             j_coord_offsetB  = DIM*jnrB;
 212
 213             /* load j atom coordinates */
 214             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 215                                               &jx0,&jy0,&jz0);
 216
 217             /* Calculate displacement vector */
 218             dx00             = _mm_sub_pd(ix0,jx0);
 219             dy00             = _mm_sub_pd(iy0,jy0);
 220             dz00             = _mm_sub_pd(iz0,jz0);
 221             dx10             = _mm_sub_pd(ix1,jx0);
 222             dy10             = _mm_sub_pd(iy1,jy0);
 223             dz10             = _mm_sub_pd(iz1,jz0);
 224             dx20             = _mm_sub_pd(ix2,jx0);
 225             dy20             = _mm_sub_pd(iy2,jy0);
 226             dz20             = _mm_sub_pd(iz2,jz0);
 227             dx30             = _mm_sub_pd(ix3,jx0);
 228             dy30             = _mm_sub_pd(iy3,jy0);
 229             dz30             = _mm_sub_pd(iz3,jz0);
 230
 231             /* Calculate squared distance and things based on it */
 232             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 233             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 234             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 235             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 236
 237             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 238             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 239             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 240             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 241
 242             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 243             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 244             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 245             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 246
 247             /* Load parameters for j particles */
 248             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 249             vdwjidx0A        = 2*vdwtype[jnrA+0];
 250             vdwjidx0B        = 2*vdwtype[jnrB+0];
 251
 252             fjx0             = _mm_setzero_pd();
 253             fjy0             = _mm_setzero_pd();
 254             fjz0             = _mm_setzero_pd();
 255
 256             /**************************
 257              * CALCULATE INTERACTIONS *
 258              **************************/
 259
 260             if (gmx_mm_any_lt(rsq00,rcutoff2))
 261             {
 262
 263             r00              = _mm_mul_pd(rsq00,rinv00);
 264
 265             /* Compute parameters for interactions between i and j atoms */
 266             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 267                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 268             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 269                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 270
 271             /* Analytical LJ-PME */
 272             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 273             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 274             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 275             exponent         = gmx_simd_exp_d(ewcljrsq);
 276             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 277             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 278             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 279             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 280             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 281             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),
 282                                _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));
 283             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 284             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);
 285
 286             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 287
 288             /* Update potential sum for this i atom from the interaction with this j atom. */
 289             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 290             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 291
 292             fscal            = fvdw;
 293
 294             fscal            = _mm_and_pd(fscal,cutoff_mask);
 295
 296             /* Calculate temporary vectorial force */
 297             tx               = _mm_mul_pd(fscal,dx00);
 298             ty               = _mm_mul_pd(fscal,dy00);
 299             tz               = _mm_mul_pd(fscal,dz00);
 300
 301             /* Update vectorial force */
 302             fix0             = _mm_add_pd(fix0,tx);
 303             fiy0             = _mm_add_pd(fiy0,ty);
 304             fiz0             = _mm_add_pd(fiz0,tz);
 305
 306             fjx0             = _mm_add_pd(fjx0,tx);
 307             fjy0             = _mm_add_pd(fjy0,ty);
 308             fjz0             = _mm_add_pd(fjz0,tz);
 309
 310             }
 311
 312             /**************************
 313              * CALCULATE INTERACTIONS *
 314              **************************/
 315
 316             if (gmx_mm_any_lt(rsq10,rcutoff2))
 317             {
 318
 319             r10              = _mm_mul_pd(rsq10,rinv10);
 320
 321             /* Compute parameters for interactions between i and j atoms */
 322             qq10             = _mm_mul_pd(iq1,jq0);
 323
 324             /* EWALD ELECTROSTATICS */
 325
 326             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 327             ewrt             = _mm_mul_pd(r10,ewtabscale);
 328             ewitab           = _mm_cvttpd_epi32(ewrt);
 329             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 330             ewitab           = _mm_slli_epi32(ewitab,2);
 331             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 332             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 333             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 334             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 335             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 336             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 337             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 338             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 339             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 340             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 341
 342             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 343
 344             /* Update potential sum for this i atom from the interaction with this j atom. */
 345             velec            = _mm_and_pd(velec,cutoff_mask);
 346             velecsum         = _mm_add_pd(velecsum,velec);
 347
 348             fscal            = felec;
 349
 350             fscal            = _mm_and_pd(fscal,cutoff_mask);
 351
 352             /* Calculate temporary vectorial force */
 353             tx               = _mm_mul_pd(fscal,dx10);
 354             ty               = _mm_mul_pd(fscal,dy10);
 355             tz               = _mm_mul_pd(fscal,dz10);
 356
 357             /* Update vectorial force */
 358             fix1             = _mm_add_pd(fix1,tx);
 359             fiy1             = _mm_add_pd(fiy1,ty);
 360             fiz1             = _mm_add_pd(fiz1,tz);
 361
 362             fjx0             = _mm_add_pd(fjx0,tx);
 363             fjy0             = _mm_add_pd(fjy0,ty);
 364             fjz0             = _mm_add_pd(fjz0,tz);
 365
 366             }
 367
 368             /**************************
 369              * CALCULATE INTERACTIONS *
 370              **************************/
 371
 372             if (gmx_mm_any_lt(rsq20,rcutoff2))
 373             {
 374
 375             r20              = _mm_mul_pd(rsq20,rinv20);
 376
 377             /* Compute parameters for interactions between i and j atoms */
 378             qq20             = _mm_mul_pd(iq2,jq0);
 379
 380             /* EWALD ELECTROSTATICS */
 381
 382             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 383             ewrt             = _mm_mul_pd(r20,ewtabscale);
 384             ewitab           = _mm_cvttpd_epi32(ewrt);
 385             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 386             ewitab           = _mm_slli_epi32(ewitab,2);
 387             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 388             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 389             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 390             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 391             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 392             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 393             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 394             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 395             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 396             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 397
 398             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 399
 400             /* Update potential sum for this i atom from the interaction with this j atom. */
 401             velec            = _mm_and_pd(velec,cutoff_mask);
 402             velecsum         = _mm_add_pd(velecsum,velec);
 403
 404             fscal            = felec;
 405
 406             fscal            = _mm_and_pd(fscal,cutoff_mask);
 407
 408             /* Calculate temporary vectorial force */
 409             tx               = _mm_mul_pd(fscal,dx20);
 410             ty               = _mm_mul_pd(fscal,dy20);
 411             tz               = _mm_mul_pd(fscal,dz20);
 412
 413             /* Update vectorial force */
 414             fix2             = _mm_add_pd(fix2,tx);
 415             fiy2             = _mm_add_pd(fiy2,ty);
 416             fiz2             = _mm_add_pd(fiz2,tz);
 417
 418             fjx0             = _mm_add_pd(fjx0,tx);
 419             fjy0             = _mm_add_pd(fjy0,ty);
 420             fjz0             = _mm_add_pd(fjz0,tz);
 421
 422             }
 423
 424             /**************************
 425              * CALCULATE INTERACTIONS *
 426              **************************/
 427
 428             if (gmx_mm_any_lt(rsq30,rcutoff2))
 429             {
 430
 431             r30              = _mm_mul_pd(rsq30,rinv30);
 432
 433             /* Compute parameters for interactions between i and j atoms */
 434             qq30             = _mm_mul_pd(iq3,jq0);
 435
 436             /* EWALD ELECTROSTATICS */
 437
 438             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 439             ewrt             = _mm_mul_pd(r30,ewtabscale);
 440             ewitab           = _mm_cvttpd_epi32(ewrt);
 441             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 442             ewitab           = _mm_slli_epi32(ewitab,2);
 443             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 444             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 445             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 446             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 447             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 448             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 449             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 450             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 451             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_sub_pd(rinv30,sh_ewald),velec));
 452             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 453
 454             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 455
 456             /* Update potential sum for this i atom from the interaction with this j atom. */
 457             velec            = _mm_and_pd(velec,cutoff_mask);
 458             velecsum         = _mm_add_pd(velecsum,velec);
 459
 460             fscal            = felec;
 461
 462             fscal            = _mm_and_pd(fscal,cutoff_mask);
 463
 464             /* Calculate temporary vectorial force */
 465             tx               = _mm_mul_pd(fscal,dx30);
 466             ty               = _mm_mul_pd(fscal,dy30);
 467             tz               = _mm_mul_pd(fscal,dz30);
 468
 469             /* Update vectorial force */
 470             fix3             = _mm_add_pd(fix3,tx);
 471             fiy3             = _mm_add_pd(fiy3,ty);
 472             fiz3             = _mm_add_pd(fiz3,tz);
 473
 474             fjx0             = _mm_add_pd(fjx0,tx);
 475             fjy0             = _mm_add_pd(fjy0,ty);
 476             fjz0             = _mm_add_pd(fjz0,tz);
 477
 478             }
 479
 480             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 481
 482             /* Inner loop uses 202 flops */
 483         }
 484
 485         if(jidx<j_index_end)
 486         {
 487
 488             jnrA             = jjnr[jidx];
 489             j_coord_offsetA  = DIM*jnrA;
 490
 491             /* load j atom coordinates */
 492             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 493                                               &jx0,&jy0,&jz0);
 494
 495             /* Calculate displacement vector */
 496             dx00             = _mm_sub_pd(ix0,jx0);
 497             dy00             = _mm_sub_pd(iy0,jy0);
 498             dz00             = _mm_sub_pd(iz0,jz0);
 499             dx10             = _mm_sub_pd(ix1,jx0);
 500             dy10             = _mm_sub_pd(iy1,jy0);
 501             dz10             = _mm_sub_pd(iz1,jz0);
 502             dx20             = _mm_sub_pd(ix2,jx0);
 503             dy20             = _mm_sub_pd(iy2,jy0);
 504             dz20             = _mm_sub_pd(iz2,jz0);
 505             dx30             = _mm_sub_pd(ix3,jx0);
 506             dy30             = _mm_sub_pd(iy3,jy0);
 507             dz30             = _mm_sub_pd(iz3,jz0);
 508
 509             /* Calculate squared distance and things based on it */
 510             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 511             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 512             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 513             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 514
 515             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 516             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 517             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 518             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 519
 520             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 521             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 522             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 523             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 524
 525             /* Load parameters for j particles */
 526             jq0              = _mm_load_sd(charge+jnrA+0);
 527             vdwjidx0A        = 2*vdwtype[jnrA+0];
 528
 529             fjx0             = _mm_setzero_pd();
 530             fjy0             = _mm_setzero_pd();
 531             fjz0             = _mm_setzero_pd();
 532
 533             /**************************
 534              * CALCULATE INTERACTIONS *
 535              **************************/
 536
 537             if (gmx_mm_any_lt(rsq00,rcutoff2))
 538             {
 539
 540             r00              = _mm_mul_pd(rsq00,rinv00);
 541
 542             /* Compute parameters for interactions between i and j atoms */
 543             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 544
 545             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 546
 547             /* Analytical LJ-PME */
 548             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 549             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 550             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 551             exponent         = gmx_simd_exp_d(ewcljrsq);
 552             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 553             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 554             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 555             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 556             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 557             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),
 558                                _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));
 559             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 560             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);
 561
 562             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 563
 564             /* Update potential sum for this i atom from the interaction with this j atom. */
 565             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 566             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 567             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 568
 569             fscal            = fvdw;
 570
 571             fscal            = _mm_and_pd(fscal,cutoff_mask);
 572
 573             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 574
 575             /* Calculate temporary vectorial force */
 576             tx               = _mm_mul_pd(fscal,dx00);
 577             ty               = _mm_mul_pd(fscal,dy00);
 578             tz               = _mm_mul_pd(fscal,dz00);
 579
 580             /* Update vectorial force */
 581             fix0             = _mm_add_pd(fix0,tx);
 582             fiy0             = _mm_add_pd(fiy0,ty);
 583             fiz0             = _mm_add_pd(fiz0,tz);
 584
 585             fjx0             = _mm_add_pd(fjx0,tx);
 586             fjy0             = _mm_add_pd(fjy0,ty);
 587             fjz0             = _mm_add_pd(fjz0,tz);
 588
 589             }
 590
 591             /**************************
 592              * CALCULATE INTERACTIONS *
 593              **************************/
 594
 595             if (gmx_mm_any_lt(rsq10,rcutoff2))
 596             {
 597
 598             r10              = _mm_mul_pd(rsq10,rinv10);
 599
 600             /* Compute parameters for interactions between i and j atoms */
 601             qq10             = _mm_mul_pd(iq1,jq0);
 602
 603             /* EWALD ELECTROSTATICS */
 604
 605             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 606             ewrt             = _mm_mul_pd(r10,ewtabscale);
 607             ewitab           = _mm_cvttpd_epi32(ewrt);
 608             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 609             ewitab           = _mm_slli_epi32(ewitab,2);
 610             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 611             ewtabD           = _mm_setzero_pd();
 612             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 613             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 614             ewtabFn          = _mm_setzero_pd();
 615             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 616             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 617             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 618             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 619             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 620
 621             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 622
 623             /* Update potential sum for this i atom from the interaction with this j atom. */
 624             velec            = _mm_and_pd(velec,cutoff_mask);
 625             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 626             velecsum         = _mm_add_pd(velecsum,velec);
 627
 628             fscal            = felec;
 629
 630             fscal            = _mm_and_pd(fscal,cutoff_mask);
 631
 632             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 633
 634             /* Calculate temporary vectorial force */
 635             tx               = _mm_mul_pd(fscal,dx10);
 636             ty               = _mm_mul_pd(fscal,dy10);
 637             tz               = _mm_mul_pd(fscal,dz10);
 638
 639             /* Update vectorial force */
 640             fix1             = _mm_add_pd(fix1,tx);
 641             fiy1             = _mm_add_pd(fiy1,ty);
 642             fiz1             = _mm_add_pd(fiz1,tz);
 643
 644             fjx0             = _mm_add_pd(fjx0,tx);
 645             fjy0             = _mm_add_pd(fjy0,ty);
 646             fjz0             = _mm_add_pd(fjz0,tz);
 647
 648             }
 649
 650             /**************************
 651              * CALCULATE INTERACTIONS *
 652              **************************/
 653
 654             if (gmx_mm_any_lt(rsq20,rcutoff2))
 655             {
 656
 657             r20              = _mm_mul_pd(rsq20,rinv20);
 658
 659             /* Compute parameters for interactions between i and j atoms */
 660             qq20             = _mm_mul_pd(iq2,jq0);
 661
 662             /* EWALD ELECTROSTATICS */
 663
 664             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 665             ewrt             = _mm_mul_pd(r20,ewtabscale);
 666             ewitab           = _mm_cvttpd_epi32(ewrt);
 667             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 668             ewitab           = _mm_slli_epi32(ewitab,2);
 669             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 670             ewtabD           = _mm_setzero_pd();
 671             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 672             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 673             ewtabFn          = _mm_setzero_pd();
 674             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 675             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 676             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 677             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 678             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 679
 680             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 681
 682             /* Update potential sum for this i atom from the interaction with this j atom. */
 683             velec            = _mm_and_pd(velec,cutoff_mask);
 684             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 685             velecsum         = _mm_add_pd(velecsum,velec);
 686
 687             fscal            = felec;
 688
 689             fscal            = _mm_and_pd(fscal,cutoff_mask);
 690
 691             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 692
 693             /* Calculate temporary vectorial force */
 694             tx               = _mm_mul_pd(fscal,dx20);
 695             ty               = _mm_mul_pd(fscal,dy20);
 696             tz               = _mm_mul_pd(fscal,dz20);
 697
 698             /* Update vectorial force */
 699             fix2             = _mm_add_pd(fix2,tx);
 700             fiy2             = _mm_add_pd(fiy2,ty);
 701             fiz2             = _mm_add_pd(fiz2,tz);
 702
 703             fjx0             = _mm_add_pd(fjx0,tx);
 704             fjy0             = _mm_add_pd(fjy0,ty);
 705             fjz0             = _mm_add_pd(fjz0,tz);
 706
 707             }
 708
 709             /**************************
 710              * CALCULATE INTERACTIONS *
 711              **************************/
 712
 713             if (gmx_mm_any_lt(rsq30,rcutoff2))
 714             {
 715
 716             r30              = _mm_mul_pd(rsq30,rinv30);
 717
 718             /* Compute parameters for interactions between i and j atoms */
 719             qq30             = _mm_mul_pd(iq3,jq0);
 720
 721             /* EWALD ELECTROSTATICS */
 722
 723             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 724             ewrt             = _mm_mul_pd(r30,ewtabscale);
 725             ewitab           = _mm_cvttpd_epi32(ewrt);
 726             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 727             ewitab           = _mm_slli_epi32(ewitab,2);
 728             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 729             ewtabD           = _mm_setzero_pd();
 730             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 731             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 732             ewtabFn          = _mm_setzero_pd();
 733             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 734             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 735             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 736             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_sub_pd(rinv30,sh_ewald),velec));
 737             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 738
 739             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 740
 741             /* Update potential sum for this i atom from the interaction with this j atom. */
 742             velec            = _mm_and_pd(velec,cutoff_mask);
 743             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 744             velecsum         = _mm_add_pd(velecsum,velec);
 745
 746             fscal            = felec;
 747
 748             fscal            = _mm_and_pd(fscal,cutoff_mask);
 749
 750             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 751
 752             /* Calculate temporary vectorial force */
 753             tx               = _mm_mul_pd(fscal,dx30);
 754             ty               = _mm_mul_pd(fscal,dy30);
 755             tz               = _mm_mul_pd(fscal,dz30);
 756
 757             /* Update vectorial force */
 758             fix3             = _mm_add_pd(fix3,tx);
 759             fiy3             = _mm_add_pd(fiy3,ty);
 760             fiz3             = _mm_add_pd(fiz3,tz);
 761
 762             fjx0             = _mm_add_pd(fjx0,tx);
 763             fjy0             = _mm_add_pd(fjy0,ty);
 764             fjz0             = _mm_add_pd(fjz0,tz);
 765
 766             }
 767
 768             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 769
 770             /* Inner loop uses 202 flops */
 771         }
 772
 773         /* End of innermost loop */
 774
 775         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 776                                               f+i_coord_offset,fshift+i_shift_offset);
 777
 778         ggid                        = gid[iidx];
 779         /* Update potential energies */
 780         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 781         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 782
 783         /* Increment number of inner iterations */
 784         inneriter                  += j_index_end - j_index_start;
 785
 786         /* Outer loop uses 26 flops */
 787     }
 788
 789     /* Increment number of outer iterations */
 790     outeriter        += nri;
 791
 792     /* Update outer/inner flops */
 793
 794     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*202);
 795 }
 796 /*
 797  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_sse4_1_double
 798  * Electrostatics interaction: Ewald
 799  * VdW interaction:            LJEwald
 800  * Geometry:                   Water4-Particle
 801  * Calculate force/pot:        Force
 802  */
 803 void
 804 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_sse4_1_double
 805                     (t_nblist                    * gmx_restrict       nlist,
 806                      rvec                        * gmx_restrict          xx,
 807                      rvec                        * gmx_restrict          ff,
 808                      t_forcerec                  * gmx_restrict          fr,
 809                      t_mdatoms                   * gmx_restrict     mdatoms,
 810                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 811                      t_nrnb                      * gmx_restrict        nrnb)
 812 {
 813     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 814      * just 0 for non-waters.
 815      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 816      * jnr indices corresponding to data put in the four positions in the SIMD register.
 817      */
 818     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 819     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 820     int              jnrA,jnrB;
 821     int              j_coord_offsetA,j_coord_offsetB;
 822     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 823     real             rcutoff_scalar;
 824     real             *shiftvec,*fshift,*x,*f;
 825     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 826     int              vdwioffset0;
 827     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 828     int              vdwioffset1;
 829     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 830     int              vdwioffset2;
 831     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 832     int              vdwioffset3;
 833     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 834     int              vdwjidx0A,vdwjidx0B;
 835     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 836     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 837     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 838     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 839     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 840     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 841     real             *charge;
 842     int              nvdwtype;
 843     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 844     int              *vdwtype;
 845     real             *vdwparam;
 846     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 847     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 848     __m128d           c6grid_00;
 849     __m128d           c6grid_10;
 850     __m128d           c6grid_20;
 851     __m128d           c6grid_30;
 852     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 853     real             *vdwgridparam;
 854     __m128d           one_half = _mm_set1_pd(0.5);
 855     __m128d           minus_one = _mm_set1_pd(-1.0);
 856     __m128i          ewitab;
 857     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 858     real             *ewtab;
 859     __m128d          dummy_mask,cutoff_mask;
 860     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 861     __m128d          one     = _mm_set1_pd(1.0);
 862     __m128d          two     = _mm_set1_pd(2.0);
 863     x                = xx[0];
 864     f                = ff[0];
 865
 866     nri              = nlist->nri;
 867     iinr             = nlist->iinr;
 868     jindex           = nlist->jindex;
 869     jjnr             = nlist->jjnr;
 870     shiftidx         = nlist->shift;
 871     gid              = nlist->gid;
 872     shiftvec         = fr->shift_vec[0];
 873     fshift           = fr->fshift[0];
 874     facel            = _mm_set1_pd(fr->epsfac);
 875     charge           = mdatoms->chargeA;
 876     nvdwtype         = fr->ntype;
 877     vdwparam         = fr->nbfp;
 878     vdwtype          = mdatoms->typeA;
 879     vdwgridparam     = fr->ljpme_c6grid;
 880     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 881     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 882     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 883
 884     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 885     ewtab            = fr->ic->tabq_coul_F;
 886     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 887     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 888
 889     /* Setup water-specific parameters */
 890     inr              = nlist->iinr[0];
 891     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 892     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 893     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 894     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 895
 896     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 897     rcutoff_scalar   = fr->rcoulomb;
 898     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 899     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 900
 901     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 902     rvdw             = _mm_set1_pd(fr->rvdw);
 903
 904     /* Avoid stupid compiler warnings */
 905     jnrA = jnrB = 0;
 906     j_coord_offsetA = 0;
 907     j_coord_offsetB = 0;
 908
 909     outeriter        = 0;
 910     inneriter        = 0;
 911
 912     /* Start outer loop over neighborlists */
 913     for(iidx=0; iidx<nri; iidx++)
 914     {
 915         /* Load shift vector for this list */
 916         i_shift_offset   = DIM*shiftidx[iidx];
 917
 918         /* Load limits for loop over neighbors */
 919         j_index_start    = jindex[iidx];
 920         j_index_end      = jindex[iidx+1];
 921
 922         /* Get outer coordinate index */
 923         inr              = iinr[iidx];
 924         i_coord_offset   = DIM*inr;
 925
 926         /* Load i particle coords and add shift vector */
 927         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 928                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 929
 930         fix0             = _mm_setzero_pd();
 931         fiy0             = _mm_setzero_pd();
 932         fiz0             = _mm_setzero_pd();
 933         fix1             = _mm_setzero_pd();
 934         fiy1             = _mm_setzero_pd();
 935         fiz1             = _mm_setzero_pd();
 936         fix2             = _mm_setzero_pd();
 937         fiy2             = _mm_setzero_pd();
 938         fiz2             = _mm_setzero_pd();
 939         fix3             = _mm_setzero_pd();
 940         fiy3             = _mm_setzero_pd();
 941         fiz3             = _mm_setzero_pd();
 942
 943         /* Start inner kernel loop */
 944         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 945         {
 946
 947             /* Get j neighbor index, and coordinate index */
 948             jnrA             = jjnr[jidx];
 949             jnrB             = jjnr[jidx+1];
 950             j_coord_offsetA  = DIM*jnrA;
 951             j_coord_offsetB  = DIM*jnrB;
 952
 953             /* load j atom coordinates */
 954             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 955                                               &jx0,&jy0,&jz0);
 956
 957             /* Calculate displacement vector */
 958             dx00             = _mm_sub_pd(ix0,jx0);
 959             dy00             = _mm_sub_pd(iy0,jy0);
 960             dz00             = _mm_sub_pd(iz0,jz0);
 961             dx10             = _mm_sub_pd(ix1,jx0);
 962             dy10             = _mm_sub_pd(iy1,jy0);
 963             dz10             = _mm_sub_pd(iz1,jz0);
 964             dx20             = _mm_sub_pd(ix2,jx0);
 965             dy20             = _mm_sub_pd(iy2,jy0);
 966             dz20             = _mm_sub_pd(iz2,jz0);
 967             dx30             = _mm_sub_pd(ix3,jx0);
 968             dy30             = _mm_sub_pd(iy3,jy0);
 969             dz30             = _mm_sub_pd(iz3,jz0);
 970
 971             /* Calculate squared distance and things based on it */
 972             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 973             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 974             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 975             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 976
 977             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 978             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 979             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 980             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 981
 982             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 983             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 984             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 985             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 986
 987             /* Load parameters for j particles */
 988             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 989             vdwjidx0A        = 2*vdwtype[jnrA+0];
 990             vdwjidx0B        = 2*vdwtype[jnrB+0];
 991
 992             fjx0             = _mm_setzero_pd();
 993             fjy0             = _mm_setzero_pd();
 994             fjz0             = _mm_setzero_pd();
 995
 996             /**************************
 997              * CALCULATE INTERACTIONS *
 998              **************************/
 999
1000             if (gmx_mm_any_lt(rsq00,rcutoff2))
1001             {
1002
1003             r00              = _mm_mul_pd(rsq00,rinv00);
1004
1005             /* Compute parameters for interactions between i and j atoms */
1006             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
1007                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
1008             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
1009                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
1010
1011             /* Analytical LJ-PME */
1012             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1013             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1014             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1015             exponent         = gmx_simd_exp_d(ewcljrsq);
1016             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1017             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1018             /* f6A = 6 * C6grid * (1 - poly) */
1019             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1020             /* f6B = C6grid * exponent * beta^6 */
1021             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1022             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1023             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);
1024
1025             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1026
1027             fscal            = fvdw;
1028
1029             fscal            = _mm_and_pd(fscal,cutoff_mask);
1030
1031             /* Calculate temporary vectorial force */
1032             tx               = _mm_mul_pd(fscal,dx00);
1033             ty               = _mm_mul_pd(fscal,dy00);
1034             tz               = _mm_mul_pd(fscal,dz00);
1035
1036             /* Update vectorial force */
1037             fix0             = _mm_add_pd(fix0,tx);
1038             fiy0             = _mm_add_pd(fiy0,ty);
1039             fiz0             = _mm_add_pd(fiz0,tz);
1040
1041             fjx0             = _mm_add_pd(fjx0,tx);
1042             fjy0             = _mm_add_pd(fjy0,ty);
1043             fjz0             = _mm_add_pd(fjz0,tz);
1044
1045             }
1046
1047             /**************************
1048              * CALCULATE INTERACTIONS *
1049              **************************/
1050
1051             if (gmx_mm_any_lt(rsq10,rcutoff2))
1052             {
1053
1054             r10              = _mm_mul_pd(rsq10,rinv10);
1055
1056             /* Compute parameters for interactions between i and j atoms */
1057             qq10             = _mm_mul_pd(iq1,jq0);
1058
1059             /* EWALD ELECTROSTATICS */
1060
1061             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1062             ewrt             = _mm_mul_pd(r10,ewtabscale);
1063             ewitab           = _mm_cvttpd_epi32(ewrt);
1064             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1065             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1066                                          &ewtabF,&ewtabFn);
1067             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1068             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1069
1070             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1071
1072             fscal            = felec;
1073
1074             fscal            = _mm_and_pd(fscal,cutoff_mask);
1075
1076             /* Calculate temporary vectorial force */
1077             tx               = _mm_mul_pd(fscal,dx10);
1078             ty               = _mm_mul_pd(fscal,dy10);
1079             tz               = _mm_mul_pd(fscal,dz10);
1080
1081             /* Update vectorial force */
1082             fix1             = _mm_add_pd(fix1,tx);
1083             fiy1             = _mm_add_pd(fiy1,ty);
1084             fiz1             = _mm_add_pd(fiz1,tz);
1085
1086             fjx0             = _mm_add_pd(fjx0,tx);
1087             fjy0             = _mm_add_pd(fjy0,ty);
1088             fjz0             = _mm_add_pd(fjz0,tz);
1089
1090             }
1091
1092             /**************************
1093              * CALCULATE INTERACTIONS *
1094              **************************/
1095
1096             if (gmx_mm_any_lt(rsq20,rcutoff2))
1097             {
1098
1099             r20              = _mm_mul_pd(rsq20,rinv20);
1100
1101             /* Compute parameters for interactions between i and j atoms */
1102             qq20             = _mm_mul_pd(iq2,jq0);
1103
1104             /* EWALD ELECTROSTATICS */
1105
1106             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1107             ewrt             = _mm_mul_pd(r20,ewtabscale);
1108             ewitab           = _mm_cvttpd_epi32(ewrt);
1109             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1110             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1111                                          &ewtabF,&ewtabFn);
1112             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1113             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1114
1115             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1116
1117             fscal            = felec;
1118
1119             fscal            = _mm_and_pd(fscal,cutoff_mask);
1120
1121             /* Calculate temporary vectorial force */
1122             tx               = _mm_mul_pd(fscal,dx20);
1123             ty               = _mm_mul_pd(fscal,dy20);
1124             tz               = _mm_mul_pd(fscal,dz20);
1125
1126             /* Update vectorial force */
1127             fix2             = _mm_add_pd(fix2,tx);
1128             fiy2             = _mm_add_pd(fiy2,ty);
1129             fiz2             = _mm_add_pd(fiz2,tz);
1130
1131             fjx0             = _mm_add_pd(fjx0,tx);
1132             fjy0             = _mm_add_pd(fjy0,ty);
1133             fjz0             = _mm_add_pd(fjz0,tz);
1134
1135             }
1136
1137             /**************************
1138              * CALCULATE INTERACTIONS *
1139              **************************/
1140
1141             if (gmx_mm_any_lt(rsq30,rcutoff2))
1142             {
1143
1144             r30              = _mm_mul_pd(rsq30,rinv30);
1145
1146             /* Compute parameters for interactions between i and j atoms */
1147             qq30             = _mm_mul_pd(iq3,jq0);
1148
1149             /* EWALD ELECTROSTATICS */
1150
1151             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1152             ewrt             = _mm_mul_pd(r30,ewtabscale);
1153             ewitab           = _mm_cvttpd_epi32(ewrt);
1154             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1155             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1156                                          &ewtabF,&ewtabFn);
1157             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1158             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1159
1160             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1161
1162             fscal            = felec;
1163
1164             fscal            = _mm_and_pd(fscal,cutoff_mask);
1165
1166             /* Calculate temporary vectorial force */
1167             tx               = _mm_mul_pd(fscal,dx30);
1168             ty               = _mm_mul_pd(fscal,dy30);
1169             tz               = _mm_mul_pd(fscal,dz30);
1170
1171             /* Update vectorial force */
1172             fix3             = _mm_add_pd(fix3,tx);
1173             fiy3             = _mm_add_pd(fiy3,ty);
1174             fiz3             = _mm_add_pd(fiz3,tz);
1175
1176             fjx0             = _mm_add_pd(fjx0,tx);
1177             fjy0             = _mm_add_pd(fjy0,ty);
1178             fjz0             = _mm_add_pd(fjz0,tz);
1179
1180             }
1181
1182             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1183
1184             /* Inner loop uses 169 flops */
1185         }
1186
1187         if(jidx<j_index_end)
1188         {
1189
1190             jnrA             = jjnr[jidx];
1191             j_coord_offsetA  = DIM*jnrA;
1192
1193             /* load j atom coordinates */
1194             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1195                                               &jx0,&jy0,&jz0);
1196
1197             /* Calculate displacement vector */
1198             dx00             = _mm_sub_pd(ix0,jx0);
1199             dy00             = _mm_sub_pd(iy0,jy0);
1200             dz00             = _mm_sub_pd(iz0,jz0);
1201             dx10             = _mm_sub_pd(ix1,jx0);
1202             dy10             = _mm_sub_pd(iy1,jy0);
1203             dz10             = _mm_sub_pd(iz1,jz0);
1204             dx20             = _mm_sub_pd(ix2,jx0);
1205             dy20             = _mm_sub_pd(iy2,jy0);
1206             dz20             = _mm_sub_pd(iz2,jz0);
1207             dx30             = _mm_sub_pd(ix3,jx0);
1208             dy30             = _mm_sub_pd(iy3,jy0);
1209             dz30             = _mm_sub_pd(iz3,jz0);
1210
1211             /* Calculate squared distance and things based on it */
1212             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1213             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1214             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1215             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1216
1217             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1218             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1219             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1220             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1221
1222             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1223             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1224             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1225             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1226
1227             /* Load parameters for j particles */
1228             jq0              = _mm_load_sd(charge+jnrA+0);
1229             vdwjidx0A        = 2*vdwtype[jnrA+0];
1230
1231             fjx0             = _mm_setzero_pd();
1232             fjy0             = _mm_setzero_pd();
1233             fjz0             = _mm_setzero_pd();
1234
1235             /**************************
1236              * CALCULATE INTERACTIONS *
1237              **************************/
1238
1239             if (gmx_mm_any_lt(rsq00,rcutoff2))
1240             {
1241
1242             r00              = _mm_mul_pd(rsq00,rinv00);
1243
1244             /* Compute parameters for interactions between i and j atoms */
1245             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1246
1247             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
1248
1249             /* Analytical LJ-PME */
1250             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1251             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1252             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1253             exponent         = gmx_simd_exp_d(ewcljrsq);
1254             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1255             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1256             /* f6A = 6 * C6grid * (1 - poly) */
1257             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1258             /* f6B = C6grid * exponent * beta^6 */
1259             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1260             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1261             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);
1262
1263             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1264
1265             fscal            = fvdw;
1266
1267             fscal            = _mm_and_pd(fscal,cutoff_mask);
1268
1269             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1270
1271             /* Calculate temporary vectorial force */
1272             tx               = _mm_mul_pd(fscal,dx00);
1273             ty               = _mm_mul_pd(fscal,dy00);
1274             tz               = _mm_mul_pd(fscal,dz00);
1275
1276             /* Update vectorial force */
1277             fix0             = _mm_add_pd(fix0,tx);
1278             fiy0             = _mm_add_pd(fiy0,ty);
1279             fiz0             = _mm_add_pd(fiz0,tz);
1280
1281             fjx0             = _mm_add_pd(fjx0,tx);
1282             fjy0             = _mm_add_pd(fjy0,ty);
1283             fjz0             = _mm_add_pd(fjz0,tz);
1284
1285             }
1286
1287             /**************************
1288              * CALCULATE INTERACTIONS *
1289              **************************/
1290
1291             if (gmx_mm_any_lt(rsq10,rcutoff2))
1292             {
1293
1294             r10              = _mm_mul_pd(rsq10,rinv10);
1295
1296             /* Compute parameters for interactions between i and j atoms */
1297             qq10             = _mm_mul_pd(iq1,jq0);
1298
1299             /* EWALD ELECTROSTATICS */
1300
1301             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1302             ewrt             = _mm_mul_pd(r10,ewtabscale);
1303             ewitab           = _mm_cvttpd_epi32(ewrt);
1304             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1305             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1306             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1307             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1308
1309             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1310
1311             fscal            = felec;
1312
1313             fscal            = _mm_and_pd(fscal,cutoff_mask);
1314
1315             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1316
1317             /* Calculate temporary vectorial force */
1318             tx               = _mm_mul_pd(fscal,dx10);
1319             ty               = _mm_mul_pd(fscal,dy10);
1320             tz               = _mm_mul_pd(fscal,dz10);
1321
1322             /* Update vectorial force */
1323             fix1             = _mm_add_pd(fix1,tx);
1324             fiy1             = _mm_add_pd(fiy1,ty);
1325             fiz1             = _mm_add_pd(fiz1,tz);
1326
1327             fjx0             = _mm_add_pd(fjx0,tx);
1328             fjy0             = _mm_add_pd(fjy0,ty);
1329             fjz0             = _mm_add_pd(fjz0,tz);
1330
1331             }
1332
1333             /**************************
1334              * CALCULATE INTERACTIONS *
1335              **************************/
1336
1337             if (gmx_mm_any_lt(rsq20,rcutoff2))
1338             {
1339
1340             r20              = _mm_mul_pd(rsq20,rinv20);
1341
1342             /* Compute parameters for interactions between i and j atoms */
1343             qq20             = _mm_mul_pd(iq2,jq0);
1344
1345             /* EWALD ELECTROSTATICS */
1346
1347             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1348             ewrt             = _mm_mul_pd(r20,ewtabscale);
1349             ewitab           = _mm_cvttpd_epi32(ewrt);
1350             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1351             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1352             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1353             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1354
1355             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1356
1357             fscal            = felec;
1358
1359             fscal            = _mm_and_pd(fscal,cutoff_mask);
1360
1361             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1362
1363             /* Calculate temporary vectorial force */
1364             tx               = _mm_mul_pd(fscal,dx20);
1365             ty               = _mm_mul_pd(fscal,dy20);
1366             tz               = _mm_mul_pd(fscal,dz20);
1367
1368             /* Update vectorial force */
1369             fix2             = _mm_add_pd(fix2,tx);
1370             fiy2             = _mm_add_pd(fiy2,ty);
1371             fiz2             = _mm_add_pd(fiz2,tz);
1372
1373             fjx0             = _mm_add_pd(fjx0,tx);
1374             fjy0             = _mm_add_pd(fjy0,ty);
1375             fjz0             = _mm_add_pd(fjz0,tz);
1376
1377             }
1378
1379             /**************************
1380              * CALCULATE INTERACTIONS *
1381              **************************/
1382
1383             if (gmx_mm_any_lt(rsq30,rcutoff2))
1384             {
1385
1386             r30              = _mm_mul_pd(rsq30,rinv30);
1387
1388             /* Compute parameters for interactions between i and j atoms */
1389             qq30             = _mm_mul_pd(iq3,jq0);
1390
1391             /* EWALD ELECTROSTATICS */
1392
1393             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1394             ewrt             = _mm_mul_pd(r30,ewtabscale);
1395             ewitab           = _mm_cvttpd_epi32(ewrt);
1396             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1397             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1398             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1399             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1400
1401             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1402
1403             fscal            = felec;
1404
1405             fscal            = _mm_and_pd(fscal,cutoff_mask);
1406
1407             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1408
1409             /* Calculate temporary vectorial force */
1410             tx               = _mm_mul_pd(fscal,dx30);
1411             ty               = _mm_mul_pd(fscal,dy30);
1412             tz               = _mm_mul_pd(fscal,dz30);
1413
1414             /* Update vectorial force */
1415             fix3             = _mm_add_pd(fix3,tx);
1416             fiy3             = _mm_add_pd(fiy3,ty);
1417             fiz3             = _mm_add_pd(fiz3,tz);
1418
1419             fjx0             = _mm_add_pd(fjx0,tx);
1420             fjy0             = _mm_add_pd(fjy0,ty);
1421             fjz0             = _mm_add_pd(fjz0,tz);
1422
1423             }
1424
1425             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1426
1427             /* Inner loop uses 169 flops */
1428         }
1429
1430         /* End of innermost loop */
1431
1432         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1433                                               f+i_coord_offset,fshift+i_shift_offset);
1434
1435         /* Increment number of inner iterations */
1436         inneriter                  += j_index_end - j_index_start;
1437
1438         /* Outer loop uses 24 flops */
1439     }
1440
1441     /* Increment number of outer iterations */
1442     outeriter        += nri;
1443
1444     /* Update outer/inner flops */
1445
1446     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*169);
1447 }