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 "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_sse2_double.h"
  50 #include "kernelutil_x86_sse2_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_VF_sse2_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_sse2_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
 269             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 270                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 271
 272             /* Analytical LJ-PME */
 273             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 274             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 275             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 276             exponent         = gmx_simd_exp_d(ewcljrsq);
 277             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 278             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 279             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 280             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 281             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 282             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),
 283                                _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));
 284             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 285             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);
 286
 287             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 288
 289             /* Update potential sum for this i atom from the interaction with this j atom. */
 290             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 291             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 292
 293             fscal            = fvdw;
 294
 295             fscal            = _mm_and_pd(fscal,cutoff_mask);
 296
 297             /* Calculate temporary vectorial force */
 298             tx               = _mm_mul_pd(fscal,dx00);
 299             ty               = _mm_mul_pd(fscal,dy00);
 300             tz               = _mm_mul_pd(fscal,dz00);
 301
 302             /* Update vectorial force */
 303             fix0             = _mm_add_pd(fix0,tx);
 304             fiy0             = _mm_add_pd(fiy0,ty);
 305             fiz0             = _mm_add_pd(fiz0,tz);
 306
 307             fjx0             = _mm_add_pd(fjx0,tx);
 308             fjy0             = _mm_add_pd(fjy0,ty);
 309             fjz0             = _mm_add_pd(fjz0,tz);
 310
 311             }
 312
 313             /**************************
 314              * CALCULATE INTERACTIONS *
 315              **************************/
 316
 317             if (gmx_mm_any_lt(rsq10,rcutoff2))
 318             {
 319
 320             r10              = _mm_mul_pd(rsq10,rinv10);
 321
 322             /* Compute parameters for interactions between i and j atoms */
 323             qq10             = _mm_mul_pd(iq1,jq0);
 324
 325             /* EWALD ELECTROSTATICS */
 326
 327             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 328             ewrt             = _mm_mul_pd(r10,ewtabscale);
 329             ewitab           = _mm_cvttpd_epi32(ewrt);
 330             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 331             ewitab           = _mm_slli_epi32(ewitab,2);
 332             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 333             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 334             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 335             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 336             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 337             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 338             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 339             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 340             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 341             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 342
 343             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 344
 345             /* Update potential sum for this i atom from the interaction with this j atom. */
 346             velec            = _mm_and_pd(velec,cutoff_mask);
 347             velecsum         = _mm_add_pd(velecsum,velec);
 348
 349             fscal            = felec;
 350
 351             fscal            = _mm_and_pd(fscal,cutoff_mask);
 352
 353             /* Calculate temporary vectorial force */
 354             tx               = _mm_mul_pd(fscal,dx10);
 355             ty               = _mm_mul_pd(fscal,dy10);
 356             tz               = _mm_mul_pd(fscal,dz10);
 357
 358             /* Update vectorial force */
 359             fix1             = _mm_add_pd(fix1,tx);
 360             fiy1             = _mm_add_pd(fiy1,ty);
 361             fiz1             = _mm_add_pd(fiz1,tz);
 362
 363             fjx0             = _mm_add_pd(fjx0,tx);
 364             fjy0             = _mm_add_pd(fjy0,ty);
 365             fjz0             = _mm_add_pd(fjz0,tz);
 366
 367             }
 368
 369             /**************************
 370              * CALCULATE INTERACTIONS *
 371              **************************/
 372
 373             if (gmx_mm_any_lt(rsq20,rcutoff2))
 374             {
 375
 376             r20              = _mm_mul_pd(rsq20,rinv20);
 377
 378             /* Compute parameters for interactions between i and j atoms */
 379             qq20             = _mm_mul_pd(iq2,jq0);
 380
 381             /* EWALD ELECTROSTATICS */
 382
 383             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 384             ewrt             = _mm_mul_pd(r20,ewtabscale);
 385             ewitab           = _mm_cvttpd_epi32(ewrt);
 386             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 387             ewitab           = _mm_slli_epi32(ewitab,2);
 388             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 389             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 390             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 391             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 392             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 393             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 394             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 395             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 396             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 397             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 398
 399             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 400
 401             /* Update potential sum for this i atom from the interaction with this j atom. */
 402             velec            = _mm_and_pd(velec,cutoff_mask);
 403             velecsum         = _mm_add_pd(velecsum,velec);
 404
 405             fscal            = felec;
 406
 407             fscal            = _mm_and_pd(fscal,cutoff_mask);
 408
 409             /* Calculate temporary vectorial force */
 410             tx               = _mm_mul_pd(fscal,dx20);
 411             ty               = _mm_mul_pd(fscal,dy20);
 412             tz               = _mm_mul_pd(fscal,dz20);
 413
 414             /* Update vectorial force */
 415             fix2             = _mm_add_pd(fix2,tx);
 416             fiy2             = _mm_add_pd(fiy2,ty);
 417             fiz2             = _mm_add_pd(fiz2,tz);
 418
 419             fjx0             = _mm_add_pd(fjx0,tx);
 420             fjy0             = _mm_add_pd(fjy0,ty);
 421             fjz0             = _mm_add_pd(fjz0,tz);
 422
 423             }
 424
 425             /**************************
 426              * CALCULATE INTERACTIONS *
 427              **************************/
 428
 429             if (gmx_mm_any_lt(rsq30,rcutoff2))
 430             {
 431
 432             r30              = _mm_mul_pd(rsq30,rinv30);
 433
 434             /* Compute parameters for interactions between i and j atoms */
 435             qq30             = _mm_mul_pd(iq3,jq0);
 436
 437             /* EWALD ELECTROSTATICS */
 438
 439             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 440             ewrt             = _mm_mul_pd(r30,ewtabscale);
 441             ewitab           = _mm_cvttpd_epi32(ewrt);
 442             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 443             ewitab           = _mm_slli_epi32(ewitab,2);
 444             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 445             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 446             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 447             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 448             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 449             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 450             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 451             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 452             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_sub_pd(rinv30,sh_ewald),velec));
 453             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 454
 455             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 456
 457             /* Update potential sum for this i atom from the interaction with this j atom. */
 458             velec            = _mm_and_pd(velec,cutoff_mask);
 459             velecsum         = _mm_add_pd(velecsum,velec);
 460
 461             fscal            = felec;
 462
 463             fscal            = _mm_and_pd(fscal,cutoff_mask);
 464
 465             /* Calculate temporary vectorial force */
 466             tx               = _mm_mul_pd(fscal,dx30);
 467             ty               = _mm_mul_pd(fscal,dy30);
 468             tz               = _mm_mul_pd(fscal,dz30);
 469
 470             /* Update vectorial force */
 471             fix3             = _mm_add_pd(fix3,tx);
 472             fiy3             = _mm_add_pd(fiy3,ty);
 473             fiz3             = _mm_add_pd(fiz3,tz);
 474
 475             fjx0             = _mm_add_pd(fjx0,tx);
 476             fjy0             = _mm_add_pd(fjy0,ty);
 477             fjz0             = _mm_add_pd(fjz0,tz);
 478
 479             }
 480
 481             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 482
 483             /* Inner loop uses 203 flops */
 484         }
 485
 486         if(jidx<j_index_end)
 487         {
 488
 489             jnrA             = jjnr[jidx];
 490             j_coord_offsetA  = DIM*jnrA;
 491
 492             /* load j atom coordinates */
 493             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 494                                               &jx0,&jy0,&jz0);
 495
 496             /* Calculate displacement vector */
 497             dx00             = _mm_sub_pd(ix0,jx0);
 498             dy00             = _mm_sub_pd(iy0,jy0);
 499             dz00             = _mm_sub_pd(iz0,jz0);
 500             dx10             = _mm_sub_pd(ix1,jx0);
 501             dy10             = _mm_sub_pd(iy1,jy0);
 502             dz10             = _mm_sub_pd(iz1,jz0);
 503             dx20             = _mm_sub_pd(ix2,jx0);
 504             dy20             = _mm_sub_pd(iy2,jy0);
 505             dz20             = _mm_sub_pd(iz2,jz0);
 506             dx30             = _mm_sub_pd(ix3,jx0);
 507             dy30             = _mm_sub_pd(iy3,jy0);
 508             dz30             = _mm_sub_pd(iz3,jz0);
 509
 510             /* Calculate squared distance and things based on it */
 511             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 512             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 513             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 514             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 515
 516             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 517             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 518             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 519             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 520
 521             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 522             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 523             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 524             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 525
 526             /* Load parameters for j particles */
 527             jq0              = _mm_load_sd(charge+jnrA+0);
 528             vdwjidx0A        = 2*vdwtype[jnrA+0];
 529
 530             fjx0             = _mm_setzero_pd();
 531             fjy0             = _mm_setzero_pd();
 532             fjz0             = _mm_setzero_pd();
 533
 534             /**************************
 535              * CALCULATE INTERACTIONS *
 536              **************************/
 537
 538             if (gmx_mm_any_lt(rsq00,rcutoff2))
 539             {
 540
 541             r00              = _mm_mul_pd(rsq00,rinv00);
 542
 543             /* Compute parameters for interactions between i and j atoms */
 544             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 545
 546             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 547
 548             /* Analytical LJ-PME */
 549             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 550             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 551             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 552             exponent         = gmx_simd_exp_d(ewcljrsq);
 553             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 554             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 555             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 556             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 557             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 558             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),
 559                                _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));
 560             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 561             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);
 562
 563             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 564
 565             /* Update potential sum for this i atom from the interaction with this j atom. */
 566             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 567             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 568             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 569
 570             fscal            = fvdw;
 571
 572             fscal            = _mm_and_pd(fscal,cutoff_mask);
 573
 574             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 575
 576             /* Calculate temporary vectorial force */
 577             tx               = _mm_mul_pd(fscal,dx00);
 578             ty               = _mm_mul_pd(fscal,dy00);
 579             tz               = _mm_mul_pd(fscal,dz00);
 580
 581             /* Update vectorial force */
 582             fix0             = _mm_add_pd(fix0,tx);
 583             fiy0             = _mm_add_pd(fiy0,ty);
 584             fiz0             = _mm_add_pd(fiz0,tz);
 585
 586             fjx0             = _mm_add_pd(fjx0,tx);
 587             fjy0             = _mm_add_pd(fjy0,ty);
 588             fjz0             = _mm_add_pd(fjz0,tz);
 589
 590             }
 591
 592             /**************************
 593              * CALCULATE INTERACTIONS *
 594              **************************/
 595
 596             if (gmx_mm_any_lt(rsq10,rcutoff2))
 597             {
 598
 599             r10              = _mm_mul_pd(rsq10,rinv10);
 600
 601             /* Compute parameters for interactions between i and j atoms */
 602             qq10             = _mm_mul_pd(iq1,jq0);
 603
 604             /* EWALD ELECTROSTATICS */
 605
 606             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 607             ewrt             = _mm_mul_pd(r10,ewtabscale);
 608             ewitab           = _mm_cvttpd_epi32(ewrt);
 609             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 610             ewitab           = _mm_slli_epi32(ewitab,2);
 611             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 612             ewtabD           = _mm_setzero_pd();
 613             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 614             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 615             ewtabFn          = _mm_setzero_pd();
 616             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 617             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 618             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 619             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 620             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 621
 622             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 623
 624             /* Update potential sum for this i atom from the interaction with this j atom. */
 625             velec            = _mm_and_pd(velec,cutoff_mask);
 626             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 627             velecsum         = _mm_add_pd(velecsum,velec);
 628
 629             fscal            = felec;
 630
 631             fscal            = _mm_and_pd(fscal,cutoff_mask);
 632
 633             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 634
 635             /* Calculate temporary vectorial force */
 636             tx               = _mm_mul_pd(fscal,dx10);
 637             ty               = _mm_mul_pd(fscal,dy10);
 638             tz               = _mm_mul_pd(fscal,dz10);
 639
 640             /* Update vectorial force */
 641             fix1             = _mm_add_pd(fix1,tx);
 642             fiy1             = _mm_add_pd(fiy1,ty);
 643             fiz1             = _mm_add_pd(fiz1,tz);
 644
 645             fjx0             = _mm_add_pd(fjx0,tx);
 646             fjy0             = _mm_add_pd(fjy0,ty);
 647             fjz0             = _mm_add_pd(fjz0,tz);
 648
 649             }
 650
 651             /**************************
 652              * CALCULATE INTERACTIONS *
 653              **************************/
 654
 655             if (gmx_mm_any_lt(rsq20,rcutoff2))
 656             {
 657
 658             r20              = _mm_mul_pd(rsq20,rinv20);
 659
 660             /* Compute parameters for interactions between i and j atoms */
 661             qq20             = _mm_mul_pd(iq2,jq0);
 662
 663             /* EWALD ELECTROSTATICS */
 664
 665             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 666             ewrt             = _mm_mul_pd(r20,ewtabscale);
 667             ewitab           = _mm_cvttpd_epi32(ewrt);
 668             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 669             ewitab           = _mm_slli_epi32(ewitab,2);
 670             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 671             ewtabD           = _mm_setzero_pd();
 672             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 673             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 674             ewtabFn          = _mm_setzero_pd();
 675             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 676             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 677             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 678             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 679             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 680
 681             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 682
 683             /* Update potential sum for this i atom from the interaction with this j atom. */
 684             velec            = _mm_and_pd(velec,cutoff_mask);
 685             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 686             velecsum         = _mm_add_pd(velecsum,velec);
 687
 688             fscal            = felec;
 689
 690             fscal            = _mm_and_pd(fscal,cutoff_mask);
 691
 692             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 693
 694             /* Calculate temporary vectorial force */
 695             tx               = _mm_mul_pd(fscal,dx20);
 696             ty               = _mm_mul_pd(fscal,dy20);
 697             tz               = _mm_mul_pd(fscal,dz20);
 698
 699             /* Update vectorial force */
 700             fix2             = _mm_add_pd(fix2,tx);
 701             fiy2             = _mm_add_pd(fiy2,ty);
 702             fiz2             = _mm_add_pd(fiz2,tz);
 703
 704             fjx0             = _mm_add_pd(fjx0,tx);
 705             fjy0             = _mm_add_pd(fjy0,ty);
 706             fjz0             = _mm_add_pd(fjz0,tz);
 707
 708             }
 709
 710             /**************************
 711              * CALCULATE INTERACTIONS *
 712              **************************/
 713
 714             if (gmx_mm_any_lt(rsq30,rcutoff2))
 715             {
 716
 717             r30              = _mm_mul_pd(rsq30,rinv30);
 718
 719             /* Compute parameters for interactions between i and j atoms */
 720             qq30             = _mm_mul_pd(iq3,jq0);
 721
 722             /* EWALD ELECTROSTATICS */
 723
 724             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 725             ewrt             = _mm_mul_pd(r30,ewtabscale);
 726             ewitab           = _mm_cvttpd_epi32(ewrt);
 727             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 728             ewitab           = _mm_slli_epi32(ewitab,2);
 729             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 730             ewtabD           = _mm_setzero_pd();
 731             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 732             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 733             ewtabFn          = _mm_setzero_pd();
 734             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 735             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 736             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 737             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_sub_pd(rinv30,sh_ewald),velec));
 738             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 739
 740             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 741
 742             /* Update potential sum for this i atom from the interaction with this j atom. */
 743             velec            = _mm_and_pd(velec,cutoff_mask);
 744             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 745             velecsum         = _mm_add_pd(velecsum,velec);
 746
 747             fscal            = felec;
 748
 749             fscal            = _mm_and_pd(fscal,cutoff_mask);
 750
 751             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 752
 753             /* Calculate temporary vectorial force */
 754             tx               = _mm_mul_pd(fscal,dx30);
 755             ty               = _mm_mul_pd(fscal,dy30);
 756             tz               = _mm_mul_pd(fscal,dz30);
 757
 758             /* Update vectorial force */
 759             fix3             = _mm_add_pd(fix3,tx);
 760             fiy3             = _mm_add_pd(fiy3,ty);
 761             fiz3             = _mm_add_pd(fiz3,tz);
 762
 763             fjx0             = _mm_add_pd(fjx0,tx);
 764             fjy0             = _mm_add_pd(fjy0,ty);
 765             fjz0             = _mm_add_pd(fjz0,tz);
 766
 767             }
 768
 769             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 770
 771             /* Inner loop uses 203 flops */
 772         }
 773
 774         /* End of innermost loop */
 775
 776         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 777                                               f+i_coord_offset,fshift+i_shift_offset);
 778
 779         ggid                        = gid[iidx];
 780         /* Update potential energies */
 781         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 782         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 783
 784         /* Increment number of inner iterations */
 785         inneriter                  += j_index_end - j_index_start;
 786
 787         /* Outer loop uses 26 flops */
 788     }
 789
 790     /* Increment number of outer iterations */
 791     outeriter        += nri;
 792
 793     /* Update outer/inner flops */
 794
 795     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*203);
 796 }
 797 /*
 798  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_sse2_double
 799  * Electrostatics interaction: Ewald
 800  * VdW interaction:            LJEwald
 801  * Geometry:                   Water4-Particle
 802  * Calculate force/pot:        Force
 803  */
 804 void
 805 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_sse2_double
 806                     (t_nblist                    * gmx_restrict       nlist,
 807                      rvec                        * gmx_restrict          xx,
 808                      rvec                        * gmx_restrict          ff,
 809                      t_forcerec                  * gmx_restrict          fr,
 810                      t_mdatoms                   * gmx_restrict     mdatoms,
 811                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 812                      t_nrnb                      * gmx_restrict        nrnb)
 813 {
 814     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 815      * just 0 for non-waters.
 816      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 817      * jnr indices corresponding to data put in the four positions in the SIMD register.
 818      */
 819     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 820     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 821     int              jnrA,jnrB;
 822     int              j_coord_offsetA,j_coord_offsetB;
 823     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 824     real             rcutoff_scalar;
 825     real             *shiftvec,*fshift,*x,*f;
 826     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 827     int              vdwioffset0;
 828     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 829     int              vdwioffset1;
 830     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 831     int              vdwioffset2;
 832     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 833     int              vdwioffset3;
 834     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 835     int              vdwjidx0A,vdwjidx0B;
 836     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 837     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 838     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 839     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 840     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 841     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 842     real             *charge;
 843     int              nvdwtype;
 844     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 845     int              *vdwtype;
 846     real             *vdwparam;
 847     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 848     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 849     __m128d           c6grid_00;
 850     __m128d           c6grid_10;
 851     __m128d           c6grid_20;
 852     __m128d           c6grid_30;
 853     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 854     real             *vdwgridparam;
 855     __m128d           one_half = _mm_set1_pd(0.5);
 856     __m128d           minus_one = _mm_set1_pd(-1.0);
 857     __m128i          ewitab;
 858     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 859     real             *ewtab;
 860     __m128d          dummy_mask,cutoff_mask;
 861     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 862     __m128d          one     = _mm_set1_pd(1.0);
 863     __m128d          two     = _mm_set1_pd(2.0);
 864     x                = xx[0];
 865     f                = ff[0];
 866
 867     nri              = nlist->nri;
 868     iinr             = nlist->iinr;
 869     jindex           = nlist->jindex;
 870     jjnr             = nlist->jjnr;
 871     shiftidx         = nlist->shift;
 872     gid              = nlist->gid;
 873     shiftvec         = fr->shift_vec[0];
 874     fshift           = fr->fshift[0];
 875     facel            = _mm_set1_pd(fr->epsfac);
 876     charge           = mdatoms->chargeA;
 877     nvdwtype         = fr->ntype;
 878     vdwparam         = fr->nbfp;
 879     vdwtype          = mdatoms->typeA;
 880     vdwgridparam     = fr->ljpme_c6grid;
 881     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 882     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 883     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 884
 885     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 886     ewtab            = fr->ic->tabq_coul_F;
 887     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 888     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 889
 890     /* Setup water-specific parameters */
 891     inr              = nlist->iinr[0];
 892     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 893     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 894     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 895     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 896
 897     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 898     rcutoff_scalar   = fr->rcoulomb;
 899     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 900     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 901
 902     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 903     rvdw             = _mm_set1_pd(fr->rvdw);
 904
 905     /* Avoid stupid compiler warnings */
 906     jnrA = jnrB = 0;
 907     j_coord_offsetA = 0;
 908     j_coord_offsetB = 0;
 909
 910     outeriter        = 0;
 911     inneriter        = 0;
 912
 913     /* Start outer loop over neighborlists */
 914     for(iidx=0; iidx<nri; iidx++)
 915     {
 916         /* Load shift vector for this list */
 917         i_shift_offset   = DIM*shiftidx[iidx];
 918
 919         /* Load limits for loop over neighbors */
 920         j_index_start    = jindex[iidx];
 921         j_index_end      = jindex[iidx+1];
 922
 923         /* Get outer coordinate index */
 924         inr              = iinr[iidx];
 925         i_coord_offset   = DIM*inr;
 926
 927         /* Load i particle coords and add shift vector */
 928         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 929                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 930
 931         fix0             = _mm_setzero_pd();
 932         fiy0             = _mm_setzero_pd();
 933         fiz0             = _mm_setzero_pd();
 934         fix1             = _mm_setzero_pd();
 935         fiy1             = _mm_setzero_pd();
 936         fiz1             = _mm_setzero_pd();
 937         fix2             = _mm_setzero_pd();
 938         fiy2             = _mm_setzero_pd();
 939         fiz2             = _mm_setzero_pd();
 940         fix3             = _mm_setzero_pd();
 941         fiy3             = _mm_setzero_pd();
 942         fiz3             = _mm_setzero_pd();
 943
 944         /* Start inner kernel loop */
 945         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 946         {
 947
 948             /* Get j neighbor index, and coordinate index */
 949             jnrA             = jjnr[jidx];
 950             jnrB             = jjnr[jidx+1];
 951             j_coord_offsetA  = DIM*jnrA;
 952             j_coord_offsetB  = DIM*jnrB;
 953
 954             /* load j atom coordinates */
 955             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 956                                               &jx0,&jy0,&jz0);
 957
 958             /* Calculate displacement vector */
 959             dx00             = _mm_sub_pd(ix0,jx0);
 960             dy00             = _mm_sub_pd(iy0,jy0);
 961             dz00             = _mm_sub_pd(iz0,jz0);
 962             dx10             = _mm_sub_pd(ix1,jx0);
 963             dy10             = _mm_sub_pd(iy1,jy0);
 964             dz10             = _mm_sub_pd(iz1,jz0);
 965             dx20             = _mm_sub_pd(ix2,jx0);
 966             dy20             = _mm_sub_pd(iy2,jy0);
 967             dz20             = _mm_sub_pd(iz2,jz0);
 968             dx30             = _mm_sub_pd(ix3,jx0);
 969             dy30             = _mm_sub_pd(iy3,jy0);
 970             dz30             = _mm_sub_pd(iz3,jz0);
 971
 972             /* Calculate squared distance and things based on it */
 973             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 974             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 975             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 976             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 977
 978             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 979             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 980             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 981             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 982
 983             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 984             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 985             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 986             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 987
 988             /* Load parameters for j particles */
 989             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 990             vdwjidx0A        = 2*vdwtype[jnrA+0];
 991             vdwjidx0B        = 2*vdwtype[jnrB+0];
 992
 993             fjx0             = _mm_setzero_pd();
 994             fjy0             = _mm_setzero_pd();
 995             fjz0             = _mm_setzero_pd();
 996
 997             /**************************
 998              * CALCULATE INTERACTIONS *
 999              **************************/
1000
1001             if (gmx_mm_any_lt(rsq00,rcutoff2))
1002             {
1003
1004             r00              = _mm_mul_pd(rsq00,rinv00);
1005
1006             /* Compute parameters for interactions between i and j atoms */
1007             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
1008                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
1009
1010             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
1011                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
1012
1013             /* Analytical LJ-PME */
1014             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1015             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1016             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1017             exponent         = gmx_simd_exp_d(ewcljrsq);
1018             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1019             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1020             /* f6A = 6 * C6grid * (1 - poly) */
1021             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1022             /* f6B = C6grid * exponent * beta^6 */
1023             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1024             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1025             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);
1026
1027             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1028
1029             fscal            = fvdw;
1030
1031             fscal            = _mm_and_pd(fscal,cutoff_mask);
1032
1033             /* Calculate temporary vectorial force */
1034             tx               = _mm_mul_pd(fscal,dx00);
1035             ty               = _mm_mul_pd(fscal,dy00);
1036             tz               = _mm_mul_pd(fscal,dz00);
1037
1038             /* Update vectorial force */
1039             fix0             = _mm_add_pd(fix0,tx);
1040             fiy0             = _mm_add_pd(fiy0,ty);
1041             fiz0             = _mm_add_pd(fiz0,tz);
1042
1043             fjx0             = _mm_add_pd(fjx0,tx);
1044             fjy0             = _mm_add_pd(fjy0,ty);
1045             fjz0             = _mm_add_pd(fjz0,tz);
1046
1047             }
1048
1049             /**************************
1050              * CALCULATE INTERACTIONS *
1051              **************************/
1052
1053             if (gmx_mm_any_lt(rsq10,rcutoff2))
1054             {
1055
1056             r10              = _mm_mul_pd(rsq10,rinv10);
1057
1058             /* Compute parameters for interactions between i and j atoms */
1059             qq10             = _mm_mul_pd(iq1,jq0);
1060
1061             /* EWALD ELECTROSTATICS */
1062
1063             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1064             ewrt             = _mm_mul_pd(r10,ewtabscale);
1065             ewitab           = _mm_cvttpd_epi32(ewrt);
1066             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1067             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1068                                          &ewtabF,&ewtabFn);
1069             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1070             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1071
1072             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1073
1074             fscal            = felec;
1075
1076             fscal            = _mm_and_pd(fscal,cutoff_mask);
1077
1078             /* Calculate temporary vectorial force */
1079             tx               = _mm_mul_pd(fscal,dx10);
1080             ty               = _mm_mul_pd(fscal,dy10);
1081             tz               = _mm_mul_pd(fscal,dz10);
1082
1083             /* Update vectorial force */
1084             fix1             = _mm_add_pd(fix1,tx);
1085             fiy1             = _mm_add_pd(fiy1,ty);
1086             fiz1             = _mm_add_pd(fiz1,tz);
1087
1088             fjx0             = _mm_add_pd(fjx0,tx);
1089             fjy0             = _mm_add_pd(fjy0,ty);
1090             fjz0             = _mm_add_pd(fjz0,tz);
1091
1092             }
1093
1094             /**************************
1095              * CALCULATE INTERACTIONS *
1096              **************************/
1097
1098             if (gmx_mm_any_lt(rsq20,rcutoff2))
1099             {
1100
1101             r20              = _mm_mul_pd(rsq20,rinv20);
1102
1103             /* Compute parameters for interactions between i and j atoms */
1104             qq20             = _mm_mul_pd(iq2,jq0);
1105
1106             /* EWALD ELECTROSTATICS */
1107
1108             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1109             ewrt             = _mm_mul_pd(r20,ewtabscale);
1110             ewitab           = _mm_cvttpd_epi32(ewrt);
1111             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1112             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1113                                          &ewtabF,&ewtabFn);
1114             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1115             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1116
1117             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1118
1119             fscal            = felec;
1120
1121             fscal            = _mm_and_pd(fscal,cutoff_mask);
1122
1123             /* Calculate temporary vectorial force */
1124             tx               = _mm_mul_pd(fscal,dx20);
1125             ty               = _mm_mul_pd(fscal,dy20);
1126             tz               = _mm_mul_pd(fscal,dz20);
1127
1128             /* Update vectorial force */
1129             fix2             = _mm_add_pd(fix2,tx);
1130             fiy2             = _mm_add_pd(fiy2,ty);
1131             fiz2             = _mm_add_pd(fiz2,tz);
1132
1133             fjx0             = _mm_add_pd(fjx0,tx);
1134             fjy0             = _mm_add_pd(fjy0,ty);
1135             fjz0             = _mm_add_pd(fjz0,tz);
1136
1137             }
1138
1139             /**************************
1140              * CALCULATE INTERACTIONS *
1141              **************************/
1142
1143             if (gmx_mm_any_lt(rsq30,rcutoff2))
1144             {
1145
1146             r30              = _mm_mul_pd(rsq30,rinv30);
1147
1148             /* Compute parameters for interactions between i and j atoms */
1149             qq30             = _mm_mul_pd(iq3,jq0);
1150
1151             /* EWALD ELECTROSTATICS */
1152
1153             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1154             ewrt             = _mm_mul_pd(r30,ewtabscale);
1155             ewitab           = _mm_cvttpd_epi32(ewrt);
1156             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1157             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1158                                          &ewtabF,&ewtabFn);
1159             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1160             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1161
1162             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1163
1164             fscal            = felec;
1165
1166             fscal            = _mm_and_pd(fscal,cutoff_mask);
1167
1168             /* Calculate temporary vectorial force */
1169             tx               = _mm_mul_pd(fscal,dx30);
1170             ty               = _mm_mul_pd(fscal,dy30);
1171             tz               = _mm_mul_pd(fscal,dz30);
1172
1173             /* Update vectorial force */
1174             fix3             = _mm_add_pd(fix3,tx);
1175             fiy3             = _mm_add_pd(fiy3,ty);
1176             fiz3             = _mm_add_pd(fiz3,tz);
1177
1178             fjx0             = _mm_add_pd(fjx0,tx);
1179             fjy0             = _mm_add_pd(fjy0,ty);
1180             fjz0             = _mm_add_pd(fjz0,tz);
1181
1182             }
1183
1184             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1185
1186             /* Inner loop uses 169 flops */
1187         }
1188
1189         if(jidx<j_index_end)
1190         {
1191
1192             jnrA             = jjnr[jidx];
1193             j_coord_offsetA  = DIM*jnrA;
1194
1195             /* load j atom coordinates */
1196             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1197                                               &jx0,&jy0,&jz0);
1198
1199             /* Calculate displacement vector */
1200             dx00             = _mm_sub_pd(ix0,jx0);
1201             dy00             = _mm_sub_pd(iy0,jy0);
1202             dz00             = _mm_sub_pd(iz0,jz0);
1203             dx10             = _mm_sub_pd(ix1,jx0);
1204             dy10             = _mm_sub_pd(iy1,jy0);
1205             dz10             = _mm_sub_pd(iz1,jz0);
1206             dx20             = _mm_sub_pd(ix2,jx0);
1207             dy20             = _mm_sub_pd(iy2,jy0);
1208             dz20             = _mm_sub_pd(iz2,jz0);
1209             dx30             = _mm_sub_pd(ix3,jx0);
1210             dy30             = _mm_sub_pd(iy3,jy0);
1211             dz30             = _mm_sub_pd(iz3,jz0);
1212
1213             /* Calculate squared distance and things based on it */
1214             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1215             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1216             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1217             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1218
1219             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1220             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1221             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1222             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1223
1224             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1225             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1226             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1227             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1228
1229             /* Load parameters for j particles */
1230             jq0              = _mm_load_sd(charge+jnrA+0);
1231             vdwjidx0A        = 2*vdwtype[jnrA+0];
1232
1233             fjx0             = _mm_setzero_pd();
1234             fjy0             = _mm_setzero_pd();
1235             fjz0             = _mm_setzero_pd();
1236
1237             /**************************
1238              * CALCULATE INTERACTIONS *
1239              **************************/
1240
1241             if (gmx_mm_any_lt(rsq00,rcutoff2))
1242             {
1243
1244             r00              = _mm_mul_pd(rsq00,rinv00);
1245
1246             /* Compute parameters for interactions between i and j atoms */
1247             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1248
1249             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
1250
1251             /* Analytical LJ-PME */
1252             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1253             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1254             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1255             exponent         = gmx_simd_exp_d(ewcljrsq);
1256             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1257             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1258             /* f6A = 6 * C6grid * (1 - poly) */
1259             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1260             /* f6B = C6grid * exponent * beta^6 */
1261             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1262             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1263             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);
1264
1265             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1266
1267             fscal            = fvdw;
1268
1269             fscal            = _mm_and_pd(fscal,cutoff_mask);
1270
1271             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1272
1273             /* Calculate temporary vectorial force */
1274             tx               = _mm_mul_pd(fscal,dx00);
1275             ty               = _mm_mul_pd(fscal,dy00);
1276             tz               = _mm_mul_pd(fscal,dz00);
1277
1278             /* Update vectorial force */
1279             fix0             = _mm_add_pd(fix0,tx);
1280             fiy0             = _mm_add_pd(fiy0,ty);
1281             fiz0             = _mm_add_pd(fiz0,tz);
1282
1283             fjx0             = _mm_add_pd(fjx0,tx);
1284             fjy0             = _mm_add_pd(fjy0,ty);
1285             fjz0             = _mm_add_pd(fjz0,tz);
1286
1287             }
1288
1289             /**************************
1290              * CALCULATE INTERACTIONS *
1291              **************************/
1292
1293             if (gmx_mm_any_lt(rsq10,rcutoff2))
1294             {
1295
1296             r10              = _mm_mul_pd(rsq10,rinv10);
1297
1298             /* Compute parameters for interactions between i and j atoms */
1299             qq10             = _mm_mul_pd(iq1,jq0);
1300
1301             /* EWALD ELECTROSTATICS */
1302
1303             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1304             ewrt             = _mm_mul_pd(r10,ewtabscale);
1305             ewitab           = _mm_cvttpd_epi32(ewrt);
1306             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1307             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1308             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1309             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1310
1311             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1312
1313             fscal            = felec;
1314
1315             fscal            = _mm_and_pd(fscal,cutoff_mask);
1316
1317             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1318
1319             /* Calculate temporary vectorial force */
1320             tx               = _mm_mul_pd(fscal,dx10);
1321             ty               = _mm_mul_pd(fscal,dy10);
1322             tz               = _mm_mul_pd(fscal,dz10);
1323
1324             /* Update vectorial force */
1325             fix1             = _mm_add_pd(fix1,tx);
1326             fiy1             = _mm_add_pd(fiy1,ty);
1327             fiz1             = _mm_add_pd(fiz1,tz);
1328
1329             fjx0             = _mm_add_pd(fjx0,tx);
1330             fjy0             = _mm_add_pd(fjy0,ty);
1331             fjz0             = _mm_add_pd(fjz0,tz);
1332
1333             }
1334
1335             /**************************
1336              * CALCULATE INTERACTIONS *
1337              **************************/
1338
1339             if (gmx_mm_any_lt(rsq20,rcutoff2))
1340             {
1341
1342             r20              = _mm_mul_pd(rsq20,rinv20);
1343
1344             /* Compute parameters for interactions between i and j atoms */
1345             qq20             = _mm_mul_pd(iq2,jq0);
1346
1347             /* EWALD ELECTROSTATICS */
1348
1349             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1350             ewrt             = _mm_mul_pd(r20,ewtabscale);
1351             ewitab           = _mm_cvttpd_epi32(ewrt);
1352             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1353             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1354             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1355             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1356
1357             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1358
1359             fscal            = felec;
1360
1361             fscal            = _mm_and_pd(fscal,cutoff_mask);
1362
1363             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1364
1365             /* Calculate temporary vectorial force */
1366             tx               = _mm_mul_pd(fscal,dx20);
1367             ty               = _mm_mul_pd(fscal,dy20);
1368             tz               = _mm_mul_pd(fscal,dz20);
1369
1370             /* Update vectorial force */
1371             fix2             = _mm_add_pd(fix2,tx);
1372             fiy2             = _mm_add_pd(fiy2,ty);
1373             fiz2             = _mm_add_pd(fiz2,tz);
1374
1375             fjx0             = _mm_add_pd(fjx0,tx);
1376             fjy0             = _mm_add_pd(fjy0,ty);
1377             fjz0             = _mm_add_pd(fjz0,tz);
1378
1379             }
1380
1381             /**************************
1382              * CALCULATE INTERACTIONS *
1383              **************************/
1384
1385             if (gmx_mm_any_lt(rsq30,rcutoff2))
1386             {
1387
1388             r30              = _mm_mul_pd(rsq30,rinv30);
1389
1390             /* Compute parameters for interactions between i and j atoms */
1391             qq30             = _mm_mul_pd(iq3,jq0);
1392
1393             /* EWALD ELECTROSTATICS */
1394
1395             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1396             ewrt             = _mm_mul_pd(r30,ewtabscale);
1397             ewitab           = _mm_cvttpd_epi32(ewrt);
1398             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1399             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1400             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1401             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1402
1403             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1404
1405             fscal            = felec;
1406
1407             fscal            = _mm_and_pd(fscal,cutoff_mask);
1408
1409             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1410
1411             /* Calculate temporary vectorial force */
1412             tx               = _mm_mul_pd(fscal,dx30);
1413             ty               = _mm_mul_pd(fscal,dy30);
1414             tz               = _mm_mul_pd(fscal,dz30);
1415
1416             /* Update vectorial force */
1417             fix3             = _mm_add_pd(fix3,tx);
1418             fiy3             = _mm_add_pd(fiy3,ty);
1419             fiz3             = _mm_add_pd(fiz3,tz);
1420
1421             fjx0             = _mm_add_pd(fjx0,tx);
1422             fjy0             = _mm_add_pd(fjy0,ty);
1423             fjz0             = _mm_add_pd(fjz0,tz);
1424
1425             }
1426
1427             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1428
1429             /* Inner loop uses 169 flops */
1430         }
1431
1432         /* End of innermost loop */
1433
1434         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1435                                               f+i_coord_offset,fshift+i_shift_offset);
1436
1437         /* Increment number of inner iterations */
1438         inneriter                  += j_index_end - j_index_start;
1439
1440         /* Outer loop uses 24 flops */
1441     }
1442
1443     /* Increment number of outer iterations */
1444     outeriter        += nri;
1445
1446     /* Update outer/inner flops */
1447
1448     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*169);
1449 }