src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_sse2_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 sse2_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_double.h"
  48 #include "kernelutil_x86_sse2_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_VF_sse2_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_sse2_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
 267             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 268                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 269
 270             /* Analytical LJ-PME */
 271             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 272             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 273             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 274             exponent         = gmx_simd_exp_d(ewcljrsq);
 275             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 276             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 277             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 278             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 279             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 280             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),
 281                                _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));
 282             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 283             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);
 284
 285             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 286
 287             /* Update potential sum for this i atom from the interaction with this j atom. */
 288             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 289             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 290
 291             fscal            = fvdw;
 292
 293             fscal            = _mm_and_pd(fscal,cutoff_mask);
 294
 295             /* Calculate temporary vectorial force */
 296             tx               = _mm_mul_pd(fscal,dx00);
 297             ty               = _mm_mul_pd(fscal,dy00);
 298             tz               = _mm_mul_pd(fscal,dz00);
 299
 300             /* Update vectorial force */
 301             fix0             = _mm_add_pd(fix0,tx);
 302             fiy0             = _mm_add_pd(fiy0,ty);
 303             fiz0             = _mm_add_pd(fiz0,tz);
 304
 305             fjx0             = _mm_add_pd(fjx0,tx);
 306             fjy0             = _mm_add_pd(fjy0,ty);
 307             fjz0             = _mm_add_pd(fjz0,tz);
 308
 309             }
 310
 311             /**************************
 312              * CALCULATE INTERACTIONS *
 313              **************************/
 314
 315             if (gmx_mm_any_lt(rsq10,rcutoff2))
 316             {
 317
 318             r10              = _mm_mul_pd(rsq10,rinv10);
 319
 320             /* Compute parameters for interactions between i and j atoms */
 321             qq10             = _mm_mul_pd(iq1,jq0);
 322
 323             /* EWALD ELECTROSTATICS */
 324
 325             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 326             ewrt             = _mm_mul_pd(r10,ewtabscale);
 327             ewitab           = _mm_cvttpd_epi32(ewrt);
 328             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 329             ewitab           = _mm_slli_epi32(ewitab,2);
 330             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 331             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 332             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 333             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 334             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 335             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 336             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 337             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 338             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 339             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 340
 341             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 342
 343             /* Update potential sum for this i atom from the interaction with this j atom. */
 344             velec            = _mm_and_pd(velec,cutoff_mask);
 345             velecsum         = _mm_add_pd(velecsum,velec);
 346
 347             fscal            = felec;
 348
 349             fscal            = _mm_and_pd(fscal,cutoff_mask);
 350
 351             /* Calculate temporary vectorial force */
 352             tx               = _mm_mul_pd(fscal,dx10);
 353             ty               = _mm_mul_pd(fscal,dy10);
 354             tz               = _mm_mul_pd(fscal,dz10);
 355
 356             /* Update vectorial force */
 357             fix1             = _mm_add_pd(fix1,tx);
 358             fiy1             = _mm_add_pd(fiy1,ty);
 359             fiz1             = _mm_add_pd(fiz1,tz);
 360
 361             fjx0             = _mm_add_pd(fjx0,tx);
 362             fjy0             = _mm_add_pd(fjy0,ty);
 363             fjz0             = _mm_add_pd(fjz0,tz);
 364
 365             }
 366
 367             /**************************
 368              * CALCULATE INTERACTIONS *
 369              **************************/
 370
 371             if (gmx_mm_any_lt(rsq20,rcutoff2))
 372             {
 373
 374             r20              = _mm_mul_pd(rsq20,rinv20);
 375
 376             /* Compute parameters for interactions between i and j atoms */
 377             qq20             = _mm_mul_pd(iq2,jq0);
 378
 379             /* EWALD ELECTROSTATICS */
 380
 381             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 382             ewrt             = _mm_mul_pd(r20,ewtabscale);
 383             ewitab           = _mm_cvttpd_epi32(ewrt);
 384             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 385             ewitab           = _mm_slli_epi32(ewitab,2);
 386             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 387             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 388             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 389             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 390             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 391             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 392             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 393             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 394             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 395             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 396
 397             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 398
 399             /* Update potential sum for this i atom from the interaction with this j atom. */
 400             velec            = _mm_and_pd(velec,cutoff_mask);
 401             velecsum         = _mm_add_pd(velecsum,velec);
 402
 403             fscal            = felec;
 404
 405             fscal            = _mm_and_pd(fscal,cutoff_mask);
 406
 407             /* Calculate temporary vectorial force */
 408             tx               = _mm_mul_pd(fscal,dx20);
 409             ty               = _mm_mul_pd(fscal,dy20);
 410             tz               = _mm_mul_pd(fscal,dz20);
 411
 412             /* Update vectorial force */
 413             fix2             = _mm_add_pd(fix2,tx);
 414             fiy2             = _mm_add_pd(fiy2,ty);
 415             fiz2             = _mm_add_pd(fiz2,tz);
 416
 417             fjx0             = _mm_add_pd(fjx0,tx);
 418             fjy0             = _mm_add_pd(fjy0,ty);
 419             fjz0             = _mm_add_pd(fjz0,tz);
 420
 421             }
 422
 423             /**************************
 424              * CALCULATE INTERACTIONS *
 425              **************************/
 426
 427             if (gmx_mm_any_lt(rsq30,rcutoff2))
 428             {
 429
 430             r30              = _mm_mul_pd(rsq30,rinv30);
 431
 432             /* Compute parameters for interactions between i and j atoms */
 433             qq30             = _mm_mul_pd(iq3,jq0);
 434
 435             /* EWALD ELECTROSTATICS */
 436
 437             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 438             ewrt             = _mm_mul_pd(r30,ewtabscale);
 439             ewitab           = _mm_cvttpd_epi32(ewrt);
 440             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 441             ewitab           = _mm_slli_epi32(ewitab,2);
 442             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 443             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 444             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 445             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 446             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 447             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 448             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 449             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 450             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_sub_pd(rinv30,sh_ewald),velec));
 451             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 452
 453             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 454
 455             /* Update potential sum for this i atom from the interaction with this j atom. */
 456             velec            = _mm_and_pd(velec,cutoff_mask);
 457             velecsum         = _mm_add_pd(velecsum,velec);
 458
 459             fscal            = felec;
 460
 461             fscal            = _mm_and_pd(fscal,cutoff_mask);
 462
 463             /* Calculate temporary vectorial force */
 464             tx               = _mm_mul_pd(fscal,dx30);
 465             ty               = _mm_mul_pd(fscal,dy30);
 466             tz               = _mm_mul_pd(fscal,dz30);
 467
 468             /* Update vectorial force */
 469             fix3             = _mm_add_pd(fix3,tx);
 470             fiy3             = _mm_add_pd(fiy3,ty);
 471             fiz3             = _mm_add_pd(fiz3,tz);
 472
 473             fjx0             = _mm_add_pd(fjx0,tx);
 474             fjy0             = _mm_add_pd(fjy0,ty);
 475             fjz0             = _mm_add_pd(fjz0,tz);
 476
 477             }
 478
 479             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 480
 481             /* Inner loop uses 203 flops */
 482         }
 483
 484         if(jidx<j_index_end)
 485         {
 486
 487             jnrA             = jjnr[jidx];
 488             j_coord_offsetA  = DIM*jnrA;
 489
 490             /* load j atom coordinates */
 491             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 492                                               &jx0,&jy0,&jz0);
 493
 494             /* Calculate displacement vector */
 495             dx00             = _mm_sub_pd(ix0,jx0);
 496             dy00             = _mm_sub_pd(iy0,jy0);
 497             dz00             = _mm_sub_pd(iz0,jz0);
 498             dx10             = _mm_sub_pd(ix1,jx0);
 499             dy10             = _mm_sub_pd(iy1,jy0);
 500             dz10             = _mm_sub_pd(iz1,jz0);
 501             dx20             = _mm_sub_pd(ix2,jx0);
 502             dy20             = _mm_sub_pd(iy2,jy0);
 503             dz20             = _mm_sub_pd(iz2,jz0);
 504             dx30             = _mm_sub_pd(ix3,jx0);
 505             dy30             = _mm_sub_pd(iy3,jy0);
 506             dz30             = _mm_sub_pd(iz3,jz0);
 507
 508             /* Calculate squared distance and things based on it */
 509             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 510             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 511             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 512             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 513
 514             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 515             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 516             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 517             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 518
 519             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 520             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 521             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 522             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 523
 524             /* Load parameters for j particles */
 525             jq0              = _mm_load_sd(charge+jnrA+0);
 526             vdwjidx0A        = 2*vdwtype[jnrA+0];
 527
 528             fjx0             = _mm_setzero_pd();
 529             fjy0             = _mm_setzero_pd();
 530             fjz0             = _mm_setzero_pd();
 531
 532             /**************************
 533              * CALCULATE INTERACTIONS *
 534              **************************/
 535
 536             if (gmx_mm_any_lt(rsq00,rcutoff2))
 537             {
 538
 539             r00              = _mm_mul_pd(rsq00,rinv00);
 540
 541             /* Compute parameters for interactions between i and j atoms */
 542             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 543
 544             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 545
 546             /* Analytical LJ-PME */
 547             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 548             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 549             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 550             exponent         = gmx_simd_exp_d(ewcljrsq);
 551             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 552             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 553             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 554             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 555             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 556             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),
 557                                _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));
 558             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 559             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);
 560
 561             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 562
 563             /* Update potential sum for this i atom from the interaction with this j atom. */
 564             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 565             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 566             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 567
 568             fscal            = fvdw;
 569
 570             fscal            = _mm_and_pd(fscal,cutoff_mask);
 571
 572             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 573
 574             /* Calculate temporary vectorial force */
 575             tx               = _mm_mul_pd(fscal,dx00);
 576             ty               = _mm_mul_pd(fscal,dy00);
 577             tz               = _mm_mul_pd(fscal,dz00);
 578
 579             /* Update vectorial force */
 580             fix0             = _mm_add_pd(fix0,tx);
 581             fiy0             = _mm_add_pd(fiy0,ty);
 582             fiz0             = _mm_add_pd(fiz0,tz);
 583
 584             fjx0             = _mm_add_pd(fjx0,tx);
 585             fjy0             = _mm_add_pd(fjy0,ty);
 586             fjz0             = _mm_add_pd(fjz0,tz);
 587
 588             }
 589
 590             /**************************
 591              * CALCULATE INTERACTIONS *
 592              **************************/
 593
 594             if (gmx_mm_any_lt(rsq10,rcutoff2))
 595             {
 596
 597             r10              = _mm_mul_pd(rsq10,rinv10);
 598
 599             /* Compute parameters for interactions between i and j atoms */
 600             qq10             = _mm_mul_pd(iq1,jq0);
 601
 602             /* EWALD ELECTROSTATICS */
 603
 604             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 605             ewrt             = _mm_mul_pd(r10,ewtabscale);
 606             ewitab           = _mm_cvttpd_epi32(ewrt);
 607             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 608             ewitab           = _mm_slli_epi32(ewitab,2);
 609             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 610             ewtabD           = _mm_setzero_pd();
 611             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 612             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 613             ewtabFn          = _mm_setzero_pd();
 614             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 615             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 616             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 617             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 618             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 619
 620             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 621
 622             /* Update potential sum for this i atom from the interaction with this j atom. */
 623             velec            = _mm_and_pd(velec,cutoff_mask);
 624             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 625             velecsum         = _mm_add_pd(velecsum,velec);
 626
 627             fscal            = felec;
 628
 629             fscal            = _mm_and_pd(fscal,cutoff_mask);
 630
 631             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 632
 633             /* Calculate temporary vectorial force */
 634             tx               = _mm_mul_pd(fscal,dx10);
 635             ty               = _mm_mul_pd(fscal,dy10);
 636             tz               = _mm_mul_pd(fscal,dz10);
 637
 638             /* Update vectorial force */
 639             fix1             = _mm_add_pd(fix1,tx);
 640             fiy1             = _mm_add_pd(fiy1,ty);
 641             fiz1             = _mm_add_pd(fiz1,tz);
 642
 643             fjx0             = _mm_add_pd(fjx0,tx);
 644             fjy0             = _mm_add_pd(fjy0,ty);
 645             fjz0             = _mm_add_pd(fjz0,tz);
 646
 647             }
 648
 649             /**************************
 650              * CALCULATE INTERACTIONS *
 651              **************************/
 652
 653             if (gmx_mm_any_lt(rsq20,rcutoff2))
 654             {
 655
 656             r20              = _mm_mul_pd(rsq20,rinv20);
 657
 658             /* Compute parameters for interactions between i and j atoms */
 659             qq20             = _mm_mul_pd(iq2,jq0);
 660
 661             /* EWALD ELECTROSTATICS */
 662
 663             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 664             ewrt             = _mm_mul_pd(r20,ewtabscale);
 665             ewitab           = _mm_cvttpd_epi32(ewrt);
 666             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 667             ewitab           = _mm_slli_epi32(ewitab,2);
 668             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 669             ewtabD           = _mm_setzero_pd();
 670             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 671             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 672             ewtabFn          = _mm_setzero_pd();
 673             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 674             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 675             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 676             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 677             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 678
 679             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 680
 681             /* Update potential sum for this i atom from the interaction with this j atom. */
 682             velec            = _mm_and_pd(velec,cutoff_mask);
 683             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 684             velecsum         = _mm_add_pd(velecsum,velec);
 685
 686             fscal            = felec;
 687
 688             fscal            = _mm_and_pd(fscal,cutoff_mask);
 689
 690             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 691
 692             /* Calculate temporary vectorial force */
 693             tx               = _mm_mul_pd(fscal,dx20);
 694             ty               = _mm_mul_pd(fscal,dy20);
 695             tz               = _mm_mul_pd(fscal,dz20);
 696
 697             /* Update vectorial force */
 698             fix2             = _mm_add_pd(fix2,tx);
 699             fiy2             = _mm_add_pd(fiy2,ty);
 700             fiz2             = _mm_add_pd(fiz2,tz);
 701
 702             fjx0             = _mm_add_pd(fjx0,tx);
 703             fjy0             = _mm_add_pd(fjy0,ty);
 704             fjz0             = _mm_add_pd(fjz0,tz);
 705
 706             }
 707
 708             /**************************
 709              * CALCULATE INTERACTIONS *
 710              **************************/
 711
 712             if (gmx_mm_any_lt(rsq30,rcutoff2))
 713             {
 714
 715             r30              = _mm_mul_pd(rsq30,rinv30);
 716
 717             /* Compute parameters for interactions between i and j atoms */
 718             qq30             = _mm_mul_pd(iq3,jq0);
 719
 720             /* EWALD ELECTROSTATICS */
 721
 722             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 723             ewrt             = _mm_mul_pd(r30,ewtabscale);
 724             ewitab           = _mm_cvttpd_epi32(ewrt);
 725             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 726             ewitab           = _mm_slli_epi32(ewitab,2);
 727             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 728             ewtabD           = _mm_setzero_pd();
 729             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 730             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 731             ewtabFn          = _mm_setzero_pd();
 732             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 733             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 734             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 735             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_sub_pd(rinv30,sh_ewald),velec));
 736             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 737
 738             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 739
 740             /* Update potential sum for this i atom from the interaction with this j atom. */
 741             velec            = _mm_and_pd(velec,cutoff_mask);
 742             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 743             velecsum         = _mm_add_pd(velecsum,velec);
 744
 745             fscal            = felec;
 746
 747             fscal            = _mm_and_pd(fscal,cutoff_mask);
 748
 749             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 750
 751             /* Calculate temporary vectorial force */
 752             tx               = _mm_mul_pd(fscal,dx30);
 753             ty               = _mm_mul_pd(fscal,dy30);
 754             tz               = _mm_mul_pd(fscal,dz30);
 755
 756             /* Update vectorial force */
 757             fix3             = _mm_add_pd(fix3,tx);
 758             fiy3             = _mm_add_pd(fiy3,ty);
 759             fiz3             = _mm_add_pd(fiz3,tz);
 760
 761             fjx0             = _mm_add_pd(fjx0,tx);
 762             fjy0             = _mm_add_pd(fjy0,ty);
 763             fjz0             = _mm_add_pd(fjz0,tz);
 764
 765             }
 766
 767             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 768
 769             /* Inner loop uses 203 flops */
 770         }
 771
 772         /* End of innermost loop */
 773
 774         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 775                                               f+i_coord_offset,fshift+i_shift_offset);
 776
 777         ggid                        = gid[iidx];
 778         /* Update potential energies */
 779         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 780         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 781
 782         /* Increment number of inner iterations */
 783         inneriter                  += j_index_end - j_index_start;
 784
 785         /* Outer loop uses 26 flops */
 786     }
 787
 788     /* Increment number of outer iterations */
 789     outeriter        += nri;
 790
 791     /* Update outer/inner flops */
 792
 793     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*203);
 794 }
 795 /*
 796  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_sse2_double
 797  * Electrostatics interaction: Ewald
 798  * VdW interaction:            LJEwald
 799  * Geometry:                   Water4-Particle
 800  * Calculate force/pot:        Force
 801  */
 802 void
 803 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_sse2_double
 804                     (t_nblist                    * gmx_restrict       nlist,
 805                      rvec                        * gmx_restrict          xx,
 806                      rvec                        * gmx_restrict          ff,
 807                      t_forcerec                  * gmx_restrict          fr,
 808                      t_mdatoms                   * gmx_restrict     mdatoms,
 809                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 810                      t_nrnb                      * gmx_restrict        nrnb)
 811 {
 812     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 813      * just 0 for non-waters.
 814      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 815      * jnr indices corresponding to data put in the four positions in the SIMD register.
 816      */
 817     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 818     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 819     int              jnrA,jnrB;
 820     int              j_coord_offsetA,j_coord_offsetB;
 821     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 822     real             rcutoff_scalar;
 823     real             *shiftvec,*fshift,*x,*f;
 824     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 825     int              vdwioffset0;
 826     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 827     int              vdwioffset1;
 828     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 829     int              vdwioffset2;
 830     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 831     int              vdwioffset3;
 832     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 833     int              vdwjidx0A,vdwjidx0B;
 834     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 835     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 836     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 837     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 838     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 839     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 840     real             *charge;
 841     int              nvdwtype;
 842     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 843     int              *vdwtype;
 844     real             *vdwparam;
 845     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 846     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 847     __m128d           c6grid_00;
 848     __m128d           c6grid_10;
 849     __m128d           c6grid_20;
 850     __m128d           c6grid_30;
 851     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 852     real             *vdwgridparam;
 853     __m128d           one_half = _mm_set1_pd(0.5);
 854     __m128d           minus_one = _mm_set1_pd(-1.0);
 855     __m128i          ewitab;
 856     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 857     real             *ewtab;
 858     __m128d          dummy_mask,cutoff_mask;
 859     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 860     __m128d          one     = _mm_set1_pd(1.0);
 861     __m128d          two     = _mm_set1_pd(2.0);
 862     x                = xx[0];
 863     f                = ff[0];
 864
 865     nri              = nlist->nri;
 866     iinr             = nlist->iinr;
 867     jindex           = nlist->jindex;
 868     jjnr             = nlist->jjnr;
 869     shiftidx         = nlist->shift;
 870     gid              = nlist->gid;
 871     shiftvec         = fr->shift_vec[0];
 872     fshift           = fr->fshift[0];
 873     facel            = _mm_set1_pd(fr->epsfac);
 874     charge           = mdatoms->chargeA;
 875     nvdwtype         = fr->ntype;
 876     vdwparam         = fr->nbfp;
 877     vdwtype          = mdatoms->typeA;
 878     vdwgridparam     = fr->ljpme_c6grid;
 879     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 880     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 881     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 882
 883     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 884     ewtab            = fr->ic->tabq_coul_F;
 885     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 886     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 887
 888     /* Setup water-specific parameters */
 889     inr              = nlist->iinr[0];
 890     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 891     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 892     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 893     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 894
 895     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 896     rcutoff_scalar   = fr->rcoulomb;
 897     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 898     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 899
 900     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 901     rvdw             = _mm_set1_pd(fr->rvdw);
 902
 903     /* Avoid stupid compiler warnings */
 904     jnrA = jnrB = 0;
 905     j_coord_offsetA = 0;
 906     j_coord_offsetB = 0;
 907
 908     outeriter        = 0;
 909     inneriter        = 0;
 910
 911     /* Start outer loop over neighborlists */
 912     for(iidx=0; iidx<nri; iidx++)
 913     {
 914         /* Load shift vector for this list */
 915         i_shift_offset   = DIM*shiftidx[iidx];
 916
 917         /* Load limits for loop over neighbors */
 918         j_index_start    = jindex[iidx];
 919         j_index_end      = jindex[iidx+1];
 920
 921         /* Get outer coordinate index */
 922         inr              = iinr[iidx];
 923         i_coord_offset   = DIM*inr;
 924
 925         /* Load i particle coords and add shift vector */
 926         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 927                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 928
 929         fix0             = _mm_setzero_pd();
 930         fiy0             = _mm_setzero_pd();
 931         fiz0             = _mm_setzero_pd();
 932         fix1             = _mm_setzero_pd();
 933         fiy1             = _mm_setzero_pd();
 934         fiz1             = _mm_setzero_pd();
 935         fix2             = _mm_setzero_pd();
 936         fiy2             = _mm_setzero_pd();
 937         fiz2             = _mm_setzero_pd();
 938         fix3             = _mm_setzero_pd();
 939         fiy3             = _mm_setzero_pd();
 940         fiz3             = _mm_setzero_pd();
 941
 942         /* Start inner kernel loop */
 943         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 944         {
 945
 946             /* Get j neighbor index, and coordinate index */
 947             jnrA             = jjnr[jidx];
 948             jnrB             = jjnr[jidx+1];
 949             j_coord_offsetA  = DIM*jnrA;
 950             j_coord_offsetB  = DIM*jnrB;
 951
 952             /* load j atom coordinates */
 953             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 954                                               &jx0,&jy0,&jz0);
 955
 956             /* Calculate displacement vector */
 957             dx00             = _mm_sub_pd(ix0,jx0);
 958             dy00             = _mm_sub_pd(iy0,jy0);
 959             dz00             = _mm_sub_pd(iz0,jz0);
 960             dx10             = _mm_sub_pd(ix1,jx0);
 961             dy10             = _mm_sub_pd(iy1,jy0);
 962             dz10             = _mm_sub_pd(iz1,jz0);
 963             dx20             = _mm_sub_pd(ix2,jx0);
 964             dy20             = _mm_sub_pd(iy2,jy0);
 965             dz20             = _mm_sub_pd(iz2,jz0);
 966             dx30             = _mm_sub_pd(ix3,jx0);
 967             dy30             = _mm_sub_pd(iy3,jy0);
 968             dz30             = _mm_sub_pd(iz3,jz0);
 969
 970             /* Calculate squared distance and things based on it */
 971             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 972             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 973             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 974             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 975
 976             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 977             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 978             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 979             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 980
 981             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 982             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 983             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 984             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 985
 986             /* Load parameters for j particles */
 987             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 988             vdwjidx0A        = 2*vdwtype[jnrA+0];
 989             vdwjidx0B        = 2*vdwtype[jnrB+0];
 990
 991             fjx0             = _mm_setzero_pd();
 992             fjy0             = _mm_setzero_pd();
 993             fjz0             = _mm_setzero_pd();
 994
 995             /**************************
 996              * CALCULATE INTERACTIONS *
 997              **************************/
 998
 999             if (gmx_mm_any_lt(rsq00,rcutoff2))
1000             {
1001
1002             r00              = _mm_mul_pd(rsq00,rinv00);
1003
1004             /* Compute parameters for interactions between i and j atoms */
1005             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
1006                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
1007
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_cvtepi32_pd(ewitab));
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_cvtepi32_pd(ewitab));
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_cvtepi32_pd(ewitab));
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_cvtepi32_pd(ewitab));
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_cvtepi32_pd(ewitab));
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_cvtepi32_pd(ewitab));
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 }