src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_sse2_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "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_VdwLJSh_GeomW4P1_VF_sse2_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwLJSh_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     __m128i          ewitab;
 105     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 106     real             *ewtab;
 107     __m128d          dummy_mask,cutoff_mask;
 108     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 109     __m128d          one     = _mm_set1_pd(1.0);
 110     __m128d          two     = _mm_set1_pd(2.0);
 111     x                = xx[0];
 112     f                = ff[0];
 113
 114     nri              = nlist->nri;
 115     iinr             = nlist->iinr;
 116     jindex           = nlist->jindex;
 117     jjnr             = nlist->jjnr;
 118     shiftidx         = nlist->shift;
 119     gid              = nlist->gid;
 120     shiftvec         = fr->shift_vec[0];
 121     fshift           = fr->fshift[0];
 122     facel            = _mm_set1_pd(fr->epsfac);
 123     charge           = mdatoms->chargeA;
 124     nvdwtype         = fr->ntype;
 125     vdwparam         = fr->nbfp;
 126     vdwtype          = mdatoms->typeA;
 127
 128     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 129     ewtab            = fr->ic->tabq_coul_FDV0;
 130     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 131     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[0];
 135     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 136     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 137     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 138     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 139
 140     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 141     rcutoff_scalar   = fr->rcoulomb;
 142     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 143     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 144
 145     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 146     rvdw             = _mm_set1_pd(fr->rvdw);
 147
 148     /* Avoid stupid compiler warnings */
 149     jnrA = jnrB = 0;
 150     j_coord_offsetA = 0;
 151     j_coord_offsetB = 0;
 152
 153     outeriter        = 0;
 154     inneriter        = 0;
 155
 156     /* Start outer loop over neighborlists */
 157     for(iidx=0; iidx<nri; iidx++)
 158     {
 159         /* Load shift vector for this list */
 160         i_shift_offset   = DIM*shiftidx[iidx];
 161
 162         /* Load limits for loop over neighbors */
 163         j_index_start    = jindex[iidx];
 164         j_index_end      = jindex[iidx+1];
 165
 166         /* Get outer coordinate index */
 167         inr              = iinr[iidx];
 168         i_coord_offset   = DIM*inr;
 169
 170         /* Load i particle coords and add shift vector */
 171         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 172                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 173
 174         fix0             = _mm_setzero_pd();
 175         fiy0             = _mm_setzero_pd();
 176         fiz0             = _mm_setzero_pd();
 177         fix1             = _mm_setzero_pd();
 178         fiy1             = _mm_setzero_pd();
 179         fiz1             = _mm_setzero_pd();
 180         fix2             = _mm_setzero_pd();
 181         fiy2             = _mm_setzero_pd();
 182         fiz2             = _mm_setzero_pd();
 183         fix3             = _mm_setzero_pd();
 184         fiy3             = _mm_setzero_pd();
 185         fiz3             = _mm_setzero_pd();
 186
 187         /* Reset potential sums */
 188         velecsum         = _mm_setzero_pd();
 189         vvdwsum          = _mm_setzero_pd();
 190
 191         /* Start inner kernel loop */
 192         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 193         {
 194
 195             /* Get j neighbor index, and coordinate index */
 196             jnrA             = jjnr[jidx];
 197             jnrB             = jjnr[jidx+1];
 198             j_coord_offsetA  = DIM*jnrA;
 199             j_coord_offsetB  = DIM*jnrB;
 200
 201             /* load j atom coordinates */
 202             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 203                                               &jx0,&jy0,&jz0);
 204
 205             /* Calculate displacement vector */
 206             dx00             = _mm_sub_pd(ix0,jx0);
 207             dy00             = _mm_sub_pd(iy0,jy0);
 208             dz00             = _mm_sub_pd(iz0,jz0);
 209             dx10             = _mm_sub_pd(ix1,jx0);
 210             dy10             = _mm_sub_pd(iy1,jy0);
 211             dz10             = _mm_sub_pd(iz1,jz0);
 212             dx20             = _mm_sub_pd(ix2,jx0);
 213             dy20             = _mm_sub_pd(iy2,jy0);
 214             dz20             = _mm_sub_pd(iz2,jz0);
 215             dx30             = _mm_sub_pd(ix3,jx0);
 216             dy30             = _mm_sub_pd(iy3,jy0);
 217             dz30             = _mm_sub_pd(iz3,jz0);
 218
 219             /* Calculate squared distance and things based on it */
 220             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 221             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 222             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 223             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 224
 225             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 226             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 227             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 228
 229             rinvsq00         = gmx_mm_inv_pd(rsq00);
 230             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 231             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 232             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 233
 234             /* Load parameters for j particles */
 235             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 236             vdwjidx0A        = 2*vdwtype[jnrA+0];
 237             vdwjidx0B        = 2*vdwtype[jnrB+0];
 238
 239             fjx0             = _mm_setzero_pd();
 240             fjy0             = _mm_setzero_pd();
 241             fjz0             = _mm_setzero_pd();
 242
 243             /**************************
 244              * CALCULATE INTERACTIONS *
 245              **************************/
 246
 247             if (gmx_mm_any_lt(rsq00,rcutoff2))
 248             {
 249
 250             /* Compute parameters for interactions between i and j atoms */
 251             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 252                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 253
 254             /* LENNARD-JONES DISPERSION/REPULSION */
 255
 256             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 257             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 258             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 259             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) ,
 260                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 261             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 262
 263             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 264
 265             /* Update potential sum for this i atom from the interaction with this j atom. */
 266             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 267             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 268
 269             fscal            = fvdw;
 270
 271             fscal            = _mm_and_pd(fscal,cutoff_mask);
 272
 273             /* Calculate temporary vectorial force */
 274             tx               = _mm_mul_pd(fscal,dx00);
 275             ty               = _mm_mul_pd(fscal,dy00);
 276             tz               = _mm_mul_pd(fscal,dz00);
 277
 278             /* Update vectorial force */
 279             fix0             = _mm_add_pd(fix0,tx);
 280             fiy0             = _mm_add_pd(fiy0,ty);
 281             fiz0             = _mm_add_pd(fiz0,tz);
 282
 283             fjx0             = _mm_add_pd(fjx0,tx);
 284             fjy0             = _mm_add_pd(fjy0,ty);
 285             fjz0             = _mm_add_pd(fjz0,tz);
 286
 287             }
 288
 289             /**************************
 290              * CALCULATE INTERACTIONS *
 291              **************************/
 292
 293             if (gmx_mm_any_lt(rsq10,rcutoff2))
 294             {
 295
 296             r10              = _mm_mul_pd(rsq10,rinv10);
 297
 298             /* Compute parameters for interactions between i and j atoms */
 299             qq10             = _mm_mul_pd(iq1,jq0);
 300
 301             /* EWALD ELECTROSTATICS */
 302
 303             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 304             ewrt             = _mm_mul_pd(r10,ewtabscale);
 305             ewitab           = _mm_cvttpd_epi32(ewrt);
 306             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 307             ewitab           = _mm_slli_epi32(ewitab,2);
 308             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 309             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 310             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 311             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 312             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 313             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 314             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 315             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 316             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 317             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 318
 319             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 320
 321             /* Update potential sum for this i atom from the interaction with this j atom. */
 322             velec            = _mm_and_pd(velec,cutoff_mask);
 323             velecsum         = _mm_add_pd(velecsum,velec);
 324
 325             fscal            = felec;
 326
 327             fscal            = _mm_and_pd(fscal,cutoff_mask);
 328
 329             /* Calculate temporary vectorial force */
 330             tx               = _mm_mul_pd(fscal,dx10);
 331             ty               = _mm_mul_pd(fscal,dy10);
 332             tz               = _mm_mul_pd(fscal,dz10);
 333
 334             /* Update vectorial force */
 335             fix1             = _mm_add_pd(fix1,tx);
 336             fiy1             = _mm_add_pd(fiy1,ty);
 337             fiz1             = _mm_add_pd(fiz1,tz);
 338
 339             fjx0             = _mm_add_pd(fjx0,tx);
 340             fjy0             = _mm_add_pd(fjy0,ty);
 341             fjz0             = _mm_add_pd(fjz0,tz);
 342
 343             }
 344
 345             /**************************
 346              * CALCULATE INTERACTIONS *
 347              **************************/
 348
 349             if (gmx_mm_any_lt(rsq20,rcutoff2))
 350             {
 351
 352             r20              = _mm_mul_pd(rsq20,rinv20);
 353
 354             /* Compute parameters for interactions between i and j atoms */
 355             qq20             = _mm_mul_pd(iq2,jq0);
 356
 357             /* EWALD ELECTROSTATICS */
 358
 359             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 360             ewrt             = _mm_mul_pd(r20,ewtabscale);
 361             ewitab           = _mm_cvttpd_epi32(ewrt);
 362             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 363             ewitab           = _mm_slli_epi32(ewitab,2);
 364             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 365             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 366             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 367             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 368             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 369             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 370             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 371             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 372             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 373             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 374
 375             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 376
 377             /* Update potential sum for this i atom from the interaction with this j atom. */
 378             velec            = _mm_and_pd(velec,cutoff_mask);
 379             velecsum         = _mm_add_pd(velecsum,velec);
 380
 381             fscal            = felec;
 382
 383             fscal            = _mm_and_pd(fscal,cutoff_mask);
 384
 385             /* Calculate temporary vectorial force */
 386             tx               = _mm_mul_pd(fscal,dx20);
 387             ty               = _mm_mul_pd(fscal,dy20);
 388             tz               = _mm_mul_pd(fscal,dz20);
 389
 390             /* Update vectorial force */
 391             fix2             = _mm_add_pd(fix2,tx);
 392             fiy2             = _mm_add_pd(fiy2,ty);
 393             fiz2             = _mm_add_pd(fiz2,tz);
 394
 395             fjx0             = _mm_add_pd(fjx0,tx);
 396             fjy0             = _mm_add_pd(fjy0,ty);
 397             fjz0             = _mm_add_pd(fjz0,tz);
 398
 399             }
 400
 401             /**************************
 402              * CALCULATE INTERACTIONS *
 403              **************************/
 404
 405             if (gmx_mm_any_lt(rsq30,rcutoff2))
 406             {
 407
 408             r30              = _mm_mul_pd(rsq30,rinv30);
 409
 410             /* Compute parameters for interactions between i and j atoms */
 411             qq30             = _mm_mul_pd(iq3,jq0);
 412
 413             /* EWALD ELECTROSTATICS */
 414
 415             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 416             ewrt             = _mm_mul_pd(r30,ewtabscale);
 417             ewitab           = _mm_cvttpd_epi32(ewrt);
 418             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 419             ewitab           = _mm_slli_epi32(ewitab,2);
 420             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 421             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 422             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 423             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 424             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 425             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 426             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 427             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 428             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_sub_pd(rinv30,sh_ewald),velec));
 429             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 430
 431             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 432
 433             /* Update potential sum for this i atom from the interaction with this j atom. */
 434             velec            = _mm_and_pd(velec,cutoff_mask);
 435             velecsum         = _mm_add_pd(velecsum,velec);
 436
 437             fscal            = felec;
 438
 439             fscal            = _mm_and_pd(fscal,cutoff_mask);
 440
 441             /* Calculate temporary vectorial force */
 442             tx               = _mm_mul_pd(fscal,dx30);
 443             ty               = _mm_mul_pd(fscal,dy30);
 444             tz               = _mm_mul_pd(fscal,dz30);
 445
 446             /* Update vectorial force */
 447             fix3             = _mm_add_pd(fix3,tx);
 448             fiy3             = _mm_add_pd(fiy3,ty);
 449             fiz3             = _mm_add_pd(fiz3,tz);
 450
 451             fjx0             = _mm_add_pd(fjx0,tx);
 452             fjy0             = _mm_add_pd(fjy0,ty);
 453             fjz0             = _mm_add_pd(fjz0,tz);
 454
 455             }
 456
 457             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 458
 459             /* Inner loop uses 182 flops */
 460         }
 461
 462         if(jidx<j_index_end)
 463         {
 464
 465             jnrA             = jjnr[jidx];
 466             j_coord_offsetA  = DIM*jnrA;
 467
 468             /* load j atom coordinates */
 469             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 470                                               &jx0,&jy0,&jz0);
 471
 472             /* Calculate displacement vector */
 473             dx00             = _mm_sub_pd(ix0,jx0);
 474             dy00             = _mm_sub_pd(iy0,jy0);
 475             dz00             = _mm_sub_pd(iz0,jz0);
 476             dx10             = _mm_sub_pd(ix1,jx0);
 477             dy10             = _mm_sub_pd(iy1,jy0);
 478             dz10             = _mm_sub_pd(iz1,jz0);
 479             dx20             = _mm_sub_pd(ix2,jx0);
 480             dy20             = _mm_sub_pd(iy2,jy0);
 481             dz20             = _mm_sub_pd(iz2,jz0);
 482             dx30             = _mm_sub_pd(ix3,jx0);
 483             dy30             = _mm_sub_pd(iy3,jy0);
 484             dz30             = _mm_sub_pd(iz3,jz0);
 485
 486             /* Calculate squared distance and things based on it */
 487             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 488             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 489             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 490             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 491
 492             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 493             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 494             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 495
 496             rinvsq00         = gmx_mm_inv_pd(rsq00);
 497             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 498             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 499             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 500
 501             /* Load parameters for j particles */
 502             jq0              = _mm_load_sd(charge+jnrA+0);
 503             vdwjidx0A        = 2*vdwtype[jnrA+0];
 504
 505             fjx0             = _mm_setzero_pd();
 506             fjy0             = _mm_setzero_pd();
 507             fjz0             = _mm_setzero_pd();
 508
 509             /**************************
 510              * CALCULATE INTERACTIONS *
 511              **************************/
 512
 513             if (gmx_mm_any_lt(rsq00,rcutoff2))
 514             {
 515
 516             /* Compute parameters for interactions between i and j atoms */
 517             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 518
 519             /* LENNARD-JONES DISPERSION/REPULSION */
 520
 521             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 522             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 523             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 524             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) ,
 525                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 526             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 527
 528             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 529
 530             /* Update potential sum for this i atom from the interaction with this j atom. */
 531             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 532             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 533             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 534
 535             fscal            = fvdw;
 536
 537             fscal            = _mm_and_pd(fscal,cutoff_mask);
 538
 539             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 540
 541             /* Calculate temporary vectorial force */
 542             tx               = _mm_mul_pd(fscal,dx00);
 543             ty               = _mm_mul_pd(fscal,dy00);
 544             tz               = _mm_mul_pd(fscal,dz00);
 545
 546             /* Update vectorial force */
 547             fix0             = _mm_add_pd(fix0,tx);
 548             fiy0             = _mm_add_pd(fiy0,ty);
 549             fiz0             = _mm_add_pd(fiz0,tz);
 550
 551             fjx0             = _mm_add_pd(fjx0,tx);
 552             fjy0             = _mm_add_pd(fjy0,ty);
 553             fjz0             = _mm_add_pd(fjz0,tz);
 554
 555             }
 556
 557             /**************************
 558              * CALCULATE INTERACTIONS *
 559              **************************/
 560
 561             if (gmx_mm_any_lt(rsq10,rcutoff2))
 562             {
 563
 564             r10              = _mm_mul_pd(rsq10,rinv10);
 565
 566             /* Compute parameters for interactions between i and j atoms */
 567             qq10             = _mm_mul_pd(iq1,jq0);
 568
 569             /* EWALD ELECTROSTATICS */
 570
 571             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 572             ewrt             = _mm_mul_pd(r10,ewtabscale);
 573             ewitab           = _mm_cvttpd_epi32(ewrt);
 574             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 575             ewitab           = _mm_slli_epi32(ewitab,2);
 576             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 577             ewtabD           = _mm_setzero_pd();
 578             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 579             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 580             ewtabFn          = _mm_setzero_pd();
 581             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 582             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 583             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 584             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 585             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 586
 587             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 588
 589             /* Update potential sum for this i atom from the interaction with this j atom. */
 590             velec            = _mm_and_pd(velec,cutoff_mask);
 591             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 592             velecsum         = _mm_add_pd(velecsum,velec);
 593
 594             fscal            = felec;
 595
 596             fscal            = _mm_and_pd(fscal,cutoff_mask);
 597
 598             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 599
 600             /* Calculate temporary vectorial force */
 601             tx               = _mm_mul_pd(fscal,dx10);
 602             ty               = _mm_mul_pd(fscal,dy10);
 603             tz               = _mm_mul_pd(fscal,dz10);
 604
 605             /* Update vectorial force */
 606             fix1             = _mm_add_pd(fix1,tx);
 607             fiy1             = _mm_add_pd(fiy1,ty);
 608             fiz1             = _mm_add_pd(fiz1,tz);
 609
 610             fjx0             = _mm_add_pd(fjx0,tx);
 611             fjy0             = _mm_add_pd(fjy0,ty);
 612             fjz0             = _mm_add_pd(fjz0,tz);
 613
 614             }
 615
 616             /**************************
 617              * CALCULATE INTERACTIONS *
 618              **************************/
 619
 620             if (gmx_mm_any_lt(rsq20,rcutoff2))
 621             {
 622
 623             r20              = _mm_mul_pd(rsq20,rinv20);
 624
 625             /* Compute parameters for interactions between i and j atoms */
 626             qq20             = _mm_mul_pd(iq2,jq0);
 627
 628             /* EWALD ELECTROSTATICS */
 629
 630             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 631             ewrt             = _mm_mul_pd(r20,ewtabscale);
 632             ewitab           = _mm_cvttpd_epi32(ewrt);
 633             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 634             ewitab           = _mm_slli_epi32(ewitab,2);
 635             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 636             ewtabD           = _mm_setzero_pd();
 637             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 638             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 639             ewtabFn          = _mm_setzero_pd();
 640             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 641             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 642             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 643             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 644             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 645
 646             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 647
 648             /* Update potential sum for this i atom from the interaction with this j atom. */
 649             velec            = _mm_and_pd(velec,cutoff_mask);
 650             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 651             velecsum         = _mm_add_pd(velecsum,velec);
 652
 653             fscal            = felec;
 654
 655             fscal            = _mm_and_pd(fscal,cutoff_mask);
 656
 657             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 658
 659             /* Calculate temporary vectorial force */
 660             tx               = _mm_mul_pd(fscal,dx20);
 661             ty               = _mm_mul_pd(fscal,dy20);
 662             tz               = _mm_mul_pd(fscal,dz20);
 663
 664             /* Update vectorial force */
 665             fix2             = _mm_add_pd(fix2,tx);
 666             fiy2             = _mm_add_pd(fiy2,ty);
 667             fiz2             = _mm_add_pd(fiz2,tz);
 668
 669             fjx0             = _mm_add_pd(fjx0,tx);
 670             fjy0             = _mm_add_pd(fjy0,ty);
 671             fjz0             = _mm_add_pd(fjz0,tz);
 672
 673             }
 674
 675             /**************************
 676              * CALCULATE INTERACTIONS *
 677              **************************/
 678
 679             if (gmx_mm_any_lt(rsq30,rcutoff2))
 680             {
 681
 682             r30              = _mm_mul_pd(rsq30,rinv30);
 683
 684             /* Compute parameters for interactions between i and j atoms */
 685             qq30             = _mm_mul_pd(iq3,jq0);
 686
 687             /* EWALD ELECTROSTATICS */
 688
 689             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 690             ewrt             = _mm_mul_pd(r30,ewtabscale);
 691             ewitab           = _mm_cvttpd_epi32(ewrt);
 692             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 693             ewitab           = _mm_slli_epi32(ewitab,2);
 694             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 695             ewtabD           = _mm_setzero_pd();
 696             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 697             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 698             ewtabFn          = _mm_setzero_pd();
 699             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 700             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 701             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 702             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_sub_pd(rinv30,sh_ewald),velec));
 703             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 704
 705             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 706
 707             /* Update potential sum for this i atom from the interaction with this j atom. */
 708             velec            = _mm_and_pd(velec,cutoff_mask);
 709             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 710             velecsum         = _mm_add_pd(velecsum,velec);
 711
 712             fscal            = felec;
 713
 714             fscal            = _mm_and_pd(fscal,cutoff_mask);
 715
 716             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 717
 718             /* Calculate temporary vectorial force */
 719             tx               = _mm_mul_pd(fscal,dx30);
 720             ty               = _mm_mul_pd(fscal,dy30);
 721             tz               = _mm_mul_pd(fscal,dz30);
 722
 723             /* Update vectorial force */
 724             fix3             = _mm_add_pd(fix3,tx);
 725             fiy3             = _mm_add_pd(fiy3,ty);
 726             fiz3             = _mm_add_pd(fiz3,tz);
 727
 728             fjx0             = _mm_add_pd(fjx0,tx);
 729             fjy0             = _mm_add_pd(fjy0,ty);
 730             fjz0             = _mm_add_pd(fjz0,tz);
 731
 732             }
 733
 734             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 735
 736             /* Inner loop uses 182 flops */
 737         }
 738
 739         /* End of innermost loop */
 740
 741         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 742                                               f+i_coord_offset,fshift+i_shift_offset);
 743
 744         ggid                        = gid[iidx];
 745         /* Update potential energies */
 746         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 747         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 748
 749         /* Increment number of inner iterations */
 750         inneriter                  += j_index_end - j_index_start;
 751
 752         /* Outer loop uses 26 flops */
 753     }
 754
 755     /* Increment number of outer iterations */
 756     outeriter        += nri;
 757
 758     /* Update outer/inner flops */
 759
 760     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*182);
 761 }
 762 /*
 763  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_sse2_double
 764  * Electrostatics interaction: Ewald
 765  * VdW interaction:            LennardJones
 766  * Geometry:                   Water4-Particle
 767  * Calculate force/pot:        Force
 768  */
 769 void
 770 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_sse2_double
 771                     (t_nblist                    * gmx_restrict       nlist,
 772                      rvec                        * gmx_restrict          xx,
 773                      rvec                        * gmx_restrict          ff,
 774                      t_forcerec                  * gmx_restrict          fr,
 775                      t_mdatoms                   * gmx_restrict     mdatoms,
 776                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 777                      t_nrnb                      * gmx_restrict        nrnb)
 778 {
 779     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 780      * just 0 for non-waters.
 781      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 782      * jnr indices corresponding to data put in the four positions in the SIMD register.
 783      */
 784     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 785     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 786     int              jnrA,jnrB;
 787     int              j_coord_offsetA,j_coord_offsetB;
 788     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 789     real             rcutoff_scalar;
 790     real             *shiftvec,*fshift,*x,*f;
 791     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 792     int              vdwioffset0;
 793     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 794     int              vdwioffset1;
 795     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 796     int              vdwioffset2;
 797     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 798     int              vdwioffset3;
 799     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 800     int              vdwjidx0A,vdwjidx0B;
 801     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 802     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 803     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 804     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 805     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 806     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 807     real             *charge;
 808     int              nvdwtype;
 809     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 810     int              *vdwtype;
 811     real             *vdwparam;
 812     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 813     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 814     __m128i          ewitab;
 815     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 816     real             *ewtab;
 817     __m128d          dummy_mask,cutoff_mask;
 818     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 819     __m128d          one     = _mm_set1_pd(1.0);
 820     __m128d          two     = _mm_set1_pd(2.0);
 821     x                = xx[0];
 822     f                = ff[0];
 823
 824     nri              = nlist->nri;
 825     iinr             = nlist->iinr;
 826     jindex           = nlist->jindex;
 827     jjnr             = nlist->jjnr;
 828     shiftidx         = nlist->shift;
 829     gid              = nlist->gid;
 830     shiftvec         = fr->shift_vec[0];
 831     fshift           = fr->fshift[0];
 832     facel            = _mm_set1_pd(fr->epsfac);
 833     charge           = mdatoms->chargeA;
 834     nvdwtype         = fr->ntype;
 835     vdwparam         = fr->nbfp;
 836     vdwtype          = mdatoms->typeA;
 837
 838     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 839     ewtab            = fr->ic->tabq_coul_F;
 840     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 841     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 842
 843     /* Setup water-specific parameters */
 844     inr              = nlist->iinr[0];
 845     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 846     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 847     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 848     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 849
 850     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 851     rcutoff_scalar   = fr->rcoulomb;
 852     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 853     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 854
 855     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 856     rvdw             = _mm_set1_pd(fr->rvdw);
 857
 858     /* Avoid stupid compiler warnings */
 859     jnrA = jnrB = 0;
 860     j_coord_offsetA = 0;
 861     j_coord_offsetB = 0;
 862
 863     outeriter        = 0;
 864     inneriter        = 0;
 865
 866     /* Start outer loop over neighborlists */
 867     for(iidx=0; iidx<nri; iidx++)
 868     {
 869         /* Load shift vector for this list */
 870         i_shift_offset   = DIM*shiftidx[iidx];
 871
 872         /* Load limits for loop over neighbors */
 873         j_index_start    = jindex[iidx];
 874         j_index_end      = jindex[iidx+1];
 875
 876         /* Get outer coordinate index */
 877         inr              = iinr[iidx];
 878         i_coord_offset   = DIM*inr;
 879
 880         /* Load i particle coords and add shift vector */
 881         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 882                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 883
 884         fix0             = _mm_setzero_pd();
 885         fiy0             = _mm_setzero_pd();
 886         fiz0             = _mm_setzero_pd();
 887         fix1             = _mm_setzero_pd();
 888         fiy1             = _mm_setzero_pd();
 889         fiz1             = _mm_setzero_pd();
 890         fix2             = _mm_setzero_pd();
 891         fiy2             = _mm_setzero_pd();
 892         fiz2             = _mm_setzero_pd();
 893         fix3             = _mm_setzero_pd();
 894         fiy3             = _mm_setzero_pd();
 895         fiz3             = _mm_setzero_pd();
 896
 897         /* Start inner kernel loop */
 898         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 899         {
 900
 901             /* Get j neighbor index, and coordinate index */
 902             jnrA             = jjnr[jidx];
 903             jnrB             = jjnr[jidx+1];
 904             j_coord_offsetA  = DIM*jnrA;
 905             j_coord_offsetB  = DIM*jnrB;
 906
 907             /* load j atom coordinates */
 908             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 909                                               &jx0,&jy0,&jz0);
 910
 911             /* Calculate displacement vector */
 912             dx00             = _mm_sub_pd(ix0,jx0);
 913             dy00             = _mm_sub_pd(iy0,jy0);
 914             dz00             = _mm_sub_pd(iz0,jz0);
 915             dx10             = _mm_sub_pd(ix1,jx0);
 916             dy10             = _mm_sub_pd(iy1,jy0);
 917             dz10             = _mm_sub_pd(iz1,jz0);
 918             dx20             = _mm_sub_pd(ix2,jx0);
 919             dy20             = _mm_sub_pd(iy2,jy0);
 920             dz20             = _mm_sub_pd(iz2,jz0);
 921             dx30             = _mm_sub_pd(ix3,jx0);
 922             dy30             = _mm_sub_pd(iy3,jy0);
 923             dz30             = _mm_sub_pd(iz3,jz0);
 924
 925             /* Calculate squared distance and things based on it */
 926             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 927             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 928             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 929             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 930
 931             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 932             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 933             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 934
 935             rinvsq00         = gmx_mm_inv_pd(rsq00);
 936             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 937             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 938             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 939
 940             /* Load parameters for j particles */
 941             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 942             vdwjidx0A        = 2*vdwtype[jnrA+0];
 943             vdwjidx0B        = 2*vdwtype[jnrB+0];
 944
 945             fjx0             = _mm_setzero_pd();
 946             fjy0             = _mm_setzero_pd();
 947             fjz0             = _mm_setzero_pd();
 948
 949             /**************************
 950              * CALCULATE INTERACTIONS *
 951              **************************/
 952
 953             if (gmx_mm_any_lt(rsq00,rcutoff2))
 954             {
 955
 956             /* Compute parameters for interactions between i and j atoms */
 957             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 958                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 959
 960             /* LENNARD-JONES DISPERSION/REPULSION */
 961
 962             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 963             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 964
 965             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 966
 967             fscal            = fvdw;
 968
 969             fscal            = _mm_and_pd(fscal,cutoff_mask);
 970
 971             /* Calculate temporary vectorial force */
 972             tx               = _mm_mul_pd(fscal,dx00);
 973             ty               = _mm_mul_pd(fscal,dy00);
 974             tz               = _mm_mul_pd(fscal,dz00);
 975
 976             /* Update vectorial force */
 977             fix0             = _mm_add_pd(fix0,tx);
 978             fiy0             = _mm_add_pd(fiy0,ty);
 979             fiz0             = _mm_add_pd(fiz0,tz);
 980
 981             fjx0             = _mm_add_pd(fjx0,tx);
 982             fjy0             = _mm_add_pd(fjy0,ty);
 983             fjz0             = _mm_add_pd(fjz0,tz);
 984
 985             }
 986
 987             /**************************
 988              * CALCULATE INTERACTIONS *
 989              **************************/
 990
 991             if (gmx_mm_any_lt(rsq10,rcutoff2))
 992             {
 993
 994             r10              = _mm_mul_pd(rsq10,rinv10);
 995
 996             /* Compute parameters for interactions between i and j atoms */
 997             qq10             = _mm_mul_pd(iq1,jq0);
 998
 999             /* EWALD ELECTROSTATICS */
1000
1001             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1002             ewrt             = _mm_mul_pd(r10,ewtabscale);
1003             ewitab           = _mm_cvttpd_epi32(ewrt);
1004             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1005             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1006                                          &ewtabF,&ewtabFn);
1007             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1008             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1009
1010             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1011
1012             fscal            = felec;
1013
1014             fscal            = _mm_and_pd(fscal,cutoff_mask);
1015
1016             /* Calculate temporary vectorial force */
1017             tx               = _mm_mul_pd(fscal,dx10);
1018             ty               = _mm_mul_pd(fscal,dy10);
1019             tz               = _mm_mul_pd(fscal,dz10);
1020
1021             /* Update vectorial force */
1022             fix1             = _mm_add_pd(fix1,tx);
1023             fiy1             = _mm_add_pd(fiy1,ty);
1024             fiz1             = _mm_add_pd(fiz1,tz);
1025
1026             fjx0             = _mm_add_pd(fjx0,tx);
1027             fjy0             = _mm_add_pd(fjy0,ty);
1028             fjz0             = _mm_add_pd(fjz0,tz);
1029
1030             }
1031
1032             /**************************
1033              * CALCULATE INTERACTIONS *
1034              **************************/
1035
1036             if (gmx_mm_any_lt(rsq20,rcutoff2))
1037             {
1038
1039             r20              = _mm_mul_pd(rsq20,rinv20);
1040
1041             /* Compute parameters for interactions between i and j atoms */
1042             qq20             = _mm_mul_pd(iq2,jq0);
1043
1044             /* EWALD ELECTROSTATICS */
1045
1046             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1047             ewrt             = _mm_mul_pd(r20,ewtabscale);
1048             ewitab           = _mm_cvttpd_epi32(ewrt);
1049             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1050             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1051                                          &ewtabF,&ewtabFn);
1052             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1053             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1054
1055             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1056
1057             fscal            = felec;
1058
1059             fscal            = _mm_and_pd(fscal,cutoff_mask);
1060
1061             /* Calculate temporary vectorial force */
1062             tx               = _mm_mul_pd(fscal,dx20);
1063             ty               = _mm_mul_pd(fscal,dy20);
1064             tz               = _mm_mul_pd(fscal,dz20);
1065
1066             /* Update vectorial force */
1067             fix2             = _mm_add_pd(fix2,tx);
1068             fiy2             = _mm_add_pd(fiy2,ty);
1069             fiz2             = _mm_add_pd(fiz2,tz);
1070
1071             fjx0             = _mm_add_pd(fjx0,tx);
1072             fjy0             = _mm_add_pd(fjy0,ty);
1073             fjz0             = _mm_add_pd(fjz0,tz);
1074
1075             }
1076
1077             /**************************
1078              * CALCULATE INTERACTIONS *
1079              **************************/
1080
1081             if (gmx_mm_any_lt(rsq30,rcutoff2))
1082             {
1083
1084             r30              = _mm_mul_pd(rsq30,rinv30);
1085
1086             /* Compute parameters for interactions between i and j atoms */
1087             qq30             = _mm_mul_pd(iq3,jq0);
1088
1089             /* EWALD ELECTROSTATICS */
1090
1091             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1092             ewrt             = _mm_mul_pd(r30,ewtabscale);
1093             ewitab           = _mm_cvttpd_epi32(ewrt);
1094             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1095             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1096                                          &ewtabF,&ewtabFn);
1097             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1098             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1099
1100             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1101
1102             fscal            = felec;
1103
1104             fscal            = _mm_and_pd(fscal,cutoff_mask);
1105
1106             /* Calculate temporary vectorial force */
1107             tx               = _mm_mul_pd(fscal,dx30);
1108             ty               = _mm_mul_pd(fscal,dy30);
1109             tz               = _mm_mul_pd(fscal,dz30);
1110
1111             /* Update vectorial force */
1112             fix3             = _mm_add_pd(fix3,tx);
1113             fiy3             = _mm_add_pd(fiy3,ty);
1114             fiz3             = _mm_add_pd(fiz3,tz);
1115
1116             fjx0             = _mm_add_pd(fjx0,tx);
1117             fjy0             = _mm_add_pd(fjy0,ty);
1118             fjz0             = _mm_add_pd(fjz0,tz);
1119
1120             }
1121
1122             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1123
1124             /* Inner loop uses 150 flops */
1125         }
1126
1127         if(jidx<j_index_end)
1128         {
1129
1130             jnrA             = jjnr[jidx];
1131             j_coord_offsetA  = DIM*jnrA;
1132
1133             /* load j atom coordinates */
1134             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1135                                               &jx0,&jy0,&jz0);
1136
1137             /* Calculate displacement vector */
1138             dx00             = _mm_sub_pd(ix0,jx0);
1139             dy00             = _mm_sub_pd(iy0,jy0);
1140             dz00             = _mm_sub_pd(iz0,jz0);
1141             dx10             = _mm_sub_pd(ix1,jx0);
1142             dy10             = _mm_sub_pd(iy1,jy0);
1143             dz10             = _mm_sub_pd(iz1,jz0);
1144             dx20             = _mm_sub_pd(ix2,jx0);
1145             dy20             = _mm_sub_pd(iy2,jy0);
1146             dz20             = _mm_sub_pd(iz2,jz0);
1147             dx30             = _mm_sub_pd(ix3,jx0);
1148             dy30             = _mm_sub_pd(iy3,jy0);
1149             dz30             = _mm_sub_pd(iz3,jz0);
1150
1151             /* Calculate squared distance and things based on it */
1152             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1153             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1154             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1155             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1156
1157             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1158             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1159             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1160
1161             rinvsq00         = gmx_mm_inv_pd(rsq00);
1162             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1163             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1164             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1165
1166             /* Load parameters for j particles */
1167             jq0              = _mm_load_sd(charge+jnrA+0);
1168             vdwjidx0A        = 2*vdwtype[jnrA+0];
1169
1170             fjx0             = _mm_setzero_pd();
1171             fjy0             = _mm_setzero_pd();
1172             fjz0             = _mm_setzero_pd();
1173
1174             /**************************
1175              * CALCULATE INTERACTIONS *
1176              **************************/
1177
1178             if (gmx_mm_any_lt(rsq00,rcutoff2))
1179             {
1180
1181             /* Compute parameters for interactions between i and j atoms */
1182             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1183
1184             /* LENNARD-JONES DISPERSION/REPULSION */
1185
1186             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1187             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1188
1189             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1190
1191             fscal            = fvdw;
1192
1193             fscal            = _mm_and_pd(fscal,cutoff_mask);
1194
1195             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1196
1197             /* Calculate temporary vectorial force */
1198             tx               = _mm_mul_pd(fscal,dx00);
1199             ty               = _mm_mul_pd(fscal,dy00);
1200             tz               = _mm_mul_pd(fscal,dz00);
1201
1202             /* Update vectorial force */
1203             fix0             = _mm_add_pd(fix0,tx);
1204             fiy0             = _mm_add_pd(fiy0,ty);
1205             fiz0             = _mm_add_pd(fiz0,tz);
1206
1207             fjx0             = _mm_add_pd(fjx0,tx);
1208             fjy0             = _mm_add_pd(fjy0,ty);
1209             fjz0             = _mm_add_pd(fjz0,tz);
1210
1211             }
1212
1213             /**************************
1214              * CALCULATE INTERACTIONS *
1215              **************************/
1216
1217             if (gmx_mm_any_lt(rsq10,rcutoff2))
1218             {
1219
1220             r10              = _mm_mul_pd(rsq10,rinv10);
1221
1222             /* Compute parameters for interactions between i and j atoms */
1223             qq10             = _mm_mul_pd(iq1,jq0);
1224
1225             /* EWALD ELECTROSTATICS */
1226
1227             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1228             ewrt             = _mm_mul_pd(r10,ewtabscale);
1229             ewitab           = _mm_cvttpd_epi32(ewrt);
1230             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1231             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1232             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1233             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1234
1235             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1236
1237             fscal            = felec;
1238
1239             fscal            = _mm_and_pd(fscal,cutoff_mask);
1240
1241             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1242
1243             /* Calculate temporary vectorial force */
1244             tx               = _mm_mul_pd(fscal,dx10);
1245             ty               = _mm_mul_pd(fscal,dy10);
1246             tz               = _mm_mul_pd(fscal,dz10);
1247
1248             /* Update vectorial force */
1249             fix1             = _mm_add_pd(fix1,tx);
1250             fiy1             = _mm_add_pd(fiy1,ty);
1251             fiz1             = _mm_add_pd(fiz1,tz);
1252
1253             fjx0             = _mm_add_pd(fjx0,tx);
1254             fjy0             = _mm_add_pd(fjy0,ty);
1255             fjz0             = _mm_add_pd(fjz0,tz);
1256
1257             }
1258
1259             /**************************
1260              * CALCULATE INTERACTIONS *
1261              **************************/
1262
1263             if (gmx_mm_any_lt(rsq20,rcutoff2))
1264             {
1265
1266             r20              = _mm_mul_pd(rsq20,rinv20);
1267
1268             /* Compute parameters for interactions between i and j atoms */
1269             qq20             = _mm_mul_pd(iq2,jq0);
1270
1271             /* EWALD ELECTROSTATICS */
1272
1273             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1274             ewrt             = _mm_mul_pd(r20,ewtabscale);
1275             ewitab           = _mm_cvttpd_epi32(ewrt);
1276             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1277             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1278             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1279             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1280
1281             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1282
1283             fscal            = felec;
1284
1285             fscal            = _mm_and_pd(fscal,cutoff_mask);
1286
1287             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1288
1289             /* Calculate temporary vectorial force */
1290             tx               = _mm_mul_pd(fscal,dx20);
1291             ty               = _mm_mul_pd(fscal,dy20);
1292             tz               = _mm_mul_pd(fscal,dz20);
1293
1294             /* Update vectorial force */
1295             fix2             = _mm_add_pd(fix2,tx);
1296             fiy2             = _mm_add_pd(fiy2,ty);
1297             fiz2             = _mm_add_pd(fiz2,tz);
1298
1299             fjx0             = _mm_add_pd(fjx0,tx);
1300             fjy0             = _mm_add_pd(fjy0,ty);
1301             fjz0             = _mm_add_pd(fjz0,tz);
1302
1303             }
1304
1305             /**************************
1306              * CALCULATE INTERACTIONS *
1307              **************************/
1308
1309             if (gmx_mm_any_lt(rsq30,rcutoff2))
1310             {
1311
1312             r30              = _mm_mul_pd(rsq30,rinv30);
1313
1314             /* Compute parameters for interactions between i and j atoms */
1315             qq30             = _mm_mul_pd(iq3,jq0);
1316
1317             /* EWALD ELECTROSTATICS */
1318
1319             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1320             ewrt             = _mm_mul_pd(r30,ewtabscale);
1321             ewitab           = _mm_cvttpd_epi32(ewrt);
1322             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1323             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1324             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1325             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1326
1327             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1328
1329             fscal            = felec;
1330
1331             fscal            = _mm_and_pd(fscal,cutoff_mask);
1332
1333             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1334
1335             /* Calculate temporary vectorial force */
1336             tx               = _mm_mul_pd(fscal,dx30);
1337             ty               = _mm_mul_pd(fscal,dy30);
1338             tz               = _mm_mul_pd(fscal,dz30);
1339
1340             /* Update vectorial force */
1341             fix3             = _mm_add_pd(fix3,tx);
1342             fiy3             = _mm_add_pd(fiy3,ty);
1343             fiz3             = _mm_add_pd(fiz3,tz);
1344
1345             fjx0             = _mm_add_pd(fjx0,tx);
1346             fjy0             = _mm_add_pd(fjy0,ty);
1347             fjz0             = _mm_add_pd(fjz0,tz);
1348
1349             }
1350
1351             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1352
1353             /* Inner loop uses 150 flops */
1354         }
1355
1356         /* End of innermost loop */
1357
1358         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1359                                               f+i_coord_offset,fshift+i_shift_offset);
1360
1361         /* Increment number of inner iterations */
1362         inneriter                  += j_index_end - j_index_start;
1363
1364         /* Outer loop uses 24 flops */
1365     }
1366
1367     /* Increment number of outer iterations */
1368     outeriter        += nri;
1369
1370     /* Update outer/inner flops */
1371
1372     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*150);
1373 }