src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_sse2_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_sse2_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LJEwald
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_sse2_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwioffset1;
  82     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  83     int              vdwioffset2;
  84     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  85     int              vdwjidx0A,vdwjidx0B;
  86     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  89     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  90     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     int              nvdwtype;
  93     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  97     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  98     __m128d           c6grid_00;
  99     __m128d           c6grid_10;
 100     __m128d           c6grid_20;
 101     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 102     real             *vdwgridparam;
 103     __m128d           one_half = _mm_set1_pd(0.5);
 104     __m128d           minus_one = _mm_set1_pd(-1.0);
 105     __m128i          ewitab;
 106     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 107     real             *ewtab;
 108     __m128d          dummy_mask,cutoff_mask;
 109     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 110     __m128d          one     = _mm_set1_pd(1.0);
 111     __m128d          two     = _mm_set1_pd(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm_set1_pd(fr->ic->epsfac);
 124     charge           = mdatoms->chargeA;
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128     vdwgridparam     = fr->ljpme_c6grid;
 129     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 130     ewclj            = _mm_set1_pd(fr->ic->ewaldcoeff_lj);
 131     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 132
 133     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 134     ewtab            = fr->ic->tabq_coul_FDV0;
 135     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 136     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 137
 138     /* Setup water-specific parameters */
 139     inr              = nlist->iinr[0];
 140     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 141     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 142     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 143     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 144
 145     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 146     rcutoff_scalar   = fr->ic->rcoulomb;
 147     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 148     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 149
 150     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 151     rvdw             = _mm_set1_pd(fr->ic->rvdw);
 152
 153     /* Avoid stupid compiler warnings */
 154     jnrA = jnrB = 0;
 155     j_coord_offsetA = 0;
 156     j_coord_offsetB = 0;
 157
 158     outeriter        = 0;
 159     inneriter        = 0;
 160
 161     /* Start outer loop over neighborlists */
 162     for(iidx=0; iidx<nri; iidx++)
 163     {
 164         /* Load shift vector for this list */
 165         i_shift_offset   = DIM*shiftidx[iidx];
 166
 167         /* Load limits for loop over neighbors */
 168         j_index_start    = jindex[iidx];
 169         j_index_end      = jindex[iidx+1];
 170
 171         /* Get outer coordinate index */
 172         inr              = iinr[iidx];
 173         i_coord_offset   = DIM*inr;
 174
 175         /* Load i particle coords and add shift vector */
 176         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 177                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 178
 179         fix0             = _mm_setzero_pd();
 180         fiy0             = _mm_setzero_pd();
 181         fiz0             = _mm_setzero_pd();
 182         fix1             = _mm_setzero_pd();
 183         fiy1             = _mm_setzero_pd();
 184         fiz1             = _mm_setzero_pd();
 185         fix2             = _mm_setzero_pd();
 186         fiy2             = _mm_setzero_pd();
 187         fiz2             = _mm_setzero_pd();
 188
 189         /* Reset potential sums */
 190         velecsum         = _mm_setzero_pd();
 191         vvdwsum          = _mm_setzero_pd();
 192
 193         /* Start inner kernel loop */
 194         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 195         {
 196
 197             /* Get j neighbor index, and coordinate index */
 198             jnrA             = jjnr[jidx];
 199             jnrB             = jjnr[jidx+1];
 200             j_coord_offsetA  = DIM*jnrA;
 201             j_coord_offsetB  = DIM*jnrB;
 202
 203             /* load j atom coordinates */
 204             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 205                                               &jx0,&jy0,&jz0);
 206
 207             /* Calculate displacement vector */
 208             dx00             = _mm_sub_pd(ix0,jx0);
 209             dy00             = _mm_sub_pd(iy0,jy0);
 210             dz00             = _mm_sub_pd(iz0,jz0);
 211             dx10             = _mm_sub_pd(ix1,jx0);
 212             dy10             = _mm_sub_pd(iy1,jy0);
 213             dz10             = _mm_sub_pd(iz1,jz0);
 214             dx20             = _mm_sub_pd(ix2,jx0);
 215             dy20             = _mm_sub_pd(iy2,jy0);
 216             dz20             = _mm_sub_pd(iz2,jz0);
 217
 218             /* Calculate squared distance and things based on it */
 219             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 220             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 221             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 222
 223             rinv00           = sse2_invsqrt_d(rsq00);
 224             rinv10           = sse2_invsqrt_d(rsq10);
 225             rinv20           = sse2_invsqrt_d(rsq20);
 226
 227             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 228             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 229             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 230
 231             /* Load parameters for j particles */
 232             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 233             vdwjidx0A        = 2*vdwtype[jnrA+0];
 234             vdwjidx0B        = 2*vdwtype[jnrB+0];
 235
 236             fjx0             = _mm_setzero_pd();
 237             fjy0             = _mm_setzero_pd();
 238             fjz0             = _mm_setzero_pd();
 239
 240             /**************************
 241              * CALCULATE INTERACTIONS *
 242              **************************/
 243
 244             if (gmx_mm_any_lt(rsq00,rcutoff2))
 245             {
 246
 247             r00              = _mm_mul_pd(rsq00,rinv00);
 248
 249             /* Compute parameters for interactions between i and j atoms */
 250             qq00             = _mm_mul_pd(iq0,jq0);
 251             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 252                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 253
 254             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 255                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 256
 257             /* EWALD ELECTROSTATICS */
 258
 259             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 260             ewrt             = _mm_mul_pd(r00,ewtabscale);
 261             ewitab           = _mm_cvttpd_epi32(ewrt);
 262             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 263             ewitab           = _mm_slli_epi32(ewitab,2);
 264             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 265             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 266             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 267             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 268             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 269             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 270             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 271             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 272             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 273             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 274
 275             /* Analytical LJ-PME */
 276             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 277             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 278             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 279             exponent         = sse2_exp_d(ewcljrsq);
 280             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 281             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 282             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 283             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 284             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 285             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),
 286                                _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));
 287             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 288             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);
 289
 290             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 291
 292             /* Update potential sum for this i atom from the interaction with this j atom. */
 293             velec            = _mm_and_pd(velec,cutoff_mask);
 294             velecsum         = _mm_add_pd(velecsum,velec);
 295             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 296             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 297
 298             fscal            = _mm_add_pd(felec,fvdw);
 299
 300             fscal            = _mm_and_pd(fscal,cutoff_mask);
 301
 302             /* Calculate temporary vectorial force */
 303             tx               = _mm_mul_pd(fscal,dx00);
 304             ty               = _mm_mul_pd(fscal,dy00);
 305             tz               = _mm_mul_pd(fscal,dz00);
 306
 307             /* Update vectorial force */
 308             fix0             = _mm_add_pd(fix0,tx);
 309             fiy0             = _mm_add_pd(fiy0,ty);
 310             fiz0             = _mm_add_pd(fiz0,tz);
 311
 312             fjx0             = _mm_add_pd(fjx0,tx);
 313             fjy0             = _mm_add_pd(fjy0,ty);
 314             fjz0             = _mm_add_pd(fjz0,tz);
 315
 316             }
 317
 318             /**************************
 319              * CALCULATE INTERACTIONS *
 320              **************************/
 321
 322             if (gmx_mm_any_lt(rsq10,rcutoff2))
 323             {
 324
 325             r10              = _mm_mul_pd(rsq10,rinv10);
 326
 327             /* Compute parameters for interactions between i and j atoms */
 328             qq10             = _mm_mul_pd(iq1,jq0);
 329
 330             /* EWALD ELECTROSTATICS */
 331
 332             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 333             ewrt             = _mm_mul_pd(r10,ewtabscale);
 334             ewitab           = _mm_cvttpd_epi32(ewrt);
 335             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 336             ewitab           = _mm_slli_epi32(ewitab,2);
 337             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 338             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 339             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 340             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 341             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 342             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 343             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 344             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 345             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 346             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 347
 348             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 349
 350             /* Update potential sum for this i atom from the interaction with this j atom. */
 351             velec            = _mm_and_pd(velec,cutoff_mask);
 352             velecsum         = _mm_add_pd(velecsum,velec);
 353
 354             fscal            = felec;
 355
 356             fscal            = _mm_and_pd(fscal,cutoff_mask);
 357
 358             /* Calculate temporary vectorial force */
 359             tx               = _mm_mul_pd(fscal,dx10);
 360             ty               = _mm_mul_pd(fscal,dy10);
 361             tz               = _mm_mul_pd(fscal,dz10);
 362
 363             /* Update vectorial force */
 364             fix1             = _mm_add_pd(fix1,tx);
 365             fiy1             = _mm_add_pd(fiy1,ty);
 366             fiz1             = _mm_add_pd(fiz1,tz);
 367
 368             fjx0             = _mm_add_pd(fjx0,tx);
 369             fjy0             = _mm_add_pd(fjy0,ty);
 370             fjz0             = _mm_add_pd(fjz0,tz);
 371
 372             }
 373
 374             /**************************
 375              * CALCULATE INTERACTIONS *
 376              **************************/
 377
 378             if (gmx_mm_any_lt(rsq20,rcutoff2))
 379             {
 380
 381             r20              = _mm_mul_pd(rsq20,rinv20);
 382
 383             /* Compute parameters for interactions between i and j atoms */
 384             qq20             = _mm_mul_pd(iq2,jq0);
 385
 386             /* EWALD ELECTROSTATICS */
 387
 388             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 389             ewrt             = _mm_mul_pd(r20,ewtabscale);
 390             ewitab           = _mm_cvttpd_epi32(ewrt);
 391             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 392             ewitab           = _mm_slli_epi32(ewitab,2);
 393             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 394             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 395             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 396             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 397             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 398             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 399             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 400             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 401             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 402             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 403
 404             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 405
 406             /* Update potential sum for this i atom from the interaction with this j atom. */
 407             velec            = _mm_and_pd(velec,cutoff_mask);
 408             velecsum         = _mm_add_pd(velecsum,velec);
 409
 410             fscal            = felec;
 411
 412             fscal            = _mm_and_pd(fscal,cutoff_mask);
 413
 414             /* Calculate temporary vectorial force */
 415             tx               = _mm_mul_pd(fscal,dx20);
 416             ty               = _mm_mul_pd(fscal,dy20);
 417             tz               = _mm_mul_pd(fscal,dz20);
 418
 419             /* Update vectorial force */
 420             fix2             = _mm_add_pd(fix2,tx);
 421             fiy2             = _mm_add_pd(fiy2,ty);
 422             fiz2             = _mm_add_pd(fiz2,tz);
 423
 424             fjx0             = _mm_add_pd(fjx0,tx);
 425             fjy0             = _mm_add_pd(fjy0,ty);
 426             fjz0             = _mm_add_pd(fjz0,tz);
 427
 428             }
 429
 430             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 431
 432             /* Inner loop uses 177 flops */
 433         }
 434
 435         if(jidx<j_index_end)
 436         {
 437
 438             jnrA             = jjnr[jidx];
 439             j_coord_offsetA  = DIM*jnrA;
 440
 441             /* load j atom coordinates */
 442             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 443                                               &jx0,&jy0,&jz0);
 444
 445             /* Calculate displacement vector */
 446             dx00             = _mm_sub_pd(ix0,jx0);
 447             dy00             = _mm_sub_pd(iy0,jy0);
 448             dz00             = _mm_sub_pd(iz0,jz0);
 449             dx10             = _mm_sub_pd(ix1,jx0);
 450             dy10             = _mm_sub_pd(iy1,jy0);
 451             dz10             = _mm_sub_pd(iz1,jz0);
 452             dx20             = _mm_sub_pd(ix2,jx0);
 453             dy20             = _mm_sub_pd(iy2,jy0);
 454             dz20             = _mm_sub_pd(iz2,jz0);
 455
 456             /* Calculate squared distance and things based on it */
 457             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 458             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 459             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 460
 461             rinv00           = sse2_invsqrt_d(rsq00);
 462             rinv10           = sse2_invsqrt_d(rsq10);
 463             rinv20           = sse2_invsqrt_d(rsq20);
 464
 465             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 466             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 467             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 468
 469             /* Load parameters for j particles */
 470             jq0              = _mm_load_sd(charge+jnrA+0);
 471             vdwjidx0A        = 2*vdwtype[jnrA+0];
 472
 473             fjx0             = _mm_setzero_pd();
 474             fjy0             = _mm_setzero_pd();
 475             fjz0             = _mm_setzero_pd();
 476
 477             /**************************
 478              * CALCULATE INTERACTIONS *
 479              **************************/
 480
 481             if (gmx_mm_any_lt(rsq00,rcutoff2))
 482             {
 483
 484             r00              = _mm_mul_pd(rsq00,rinv00);
 485
 486             /* Compute parameters for interactions between i and j atoms */
 487             qq00             = _mm_mul_pd(iq0,jq0);
 488             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 489
 490             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 491
 492             /* EWALD ELECTROSTATICS */
 493
 494             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 495             ewrt             = _mm_mul_pd(r00,ewtabscale);
 496             ewitab           = _mm_cvttpd_epi32(ewrt);
 497             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 498             ewitab           = _mm_slli_epi32(ewitab,2);
 499             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 500             ewtabD           = _mm_setzero_pd();
 501             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 502             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 503             ewtabFn          = _mm_setzero_pd();
 504             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 505             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 506             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 507             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 508             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 509
 510             /* Analytical LJ-PME */
 511             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 512             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 513             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 514             exponent         = sse2_exp_d(ewcljrsq);
 515             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 516             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 517             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 518             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 519             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 520             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),
 521                                _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));
 522             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 523             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);
 524
 525             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 526
 527             /* Update potential sum for this i atom from the interaction with this j atom. */
 528             velec            = _mm_and_pd(velec,cutoff_mask);
 529             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 530             velecsum         = _mm_add_pd(velecsum,velec);
 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            = _mm_add_pd(felec,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             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 676
 677             /* Inner loop uses 177 flops */
 678         }
 679
 680         /* End of innermost loop */
 681
 682         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 683                                               f+i_coord_offset,fshift+i_shift_offset);
 684
 685         ggid                        = gid[iidx];
 686         /* Update potential energies */
 687         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 688         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 689
 690         /* Increment number of inner iterations */
 691         inneriter                  += j_index_end - j_index_start;
 692
 693         /* Outer loop uses 20 flops */
 694     }
 695
 696     /* Increment number of outer iterations */
 697     outeriter        += nri;
 698
 699     /* Update outer/inner flops */
 700
 701     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*177);
 702 }
 703 /*
 704  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_sse2_double
 705  * Electrostatics interaction: Ewald
 706  * VdW interaction:            LJEwald
 707  * Geometry:                   Water3-Particle
 708  * Calculate force/pot:        Force
 709  */
 710 void
 711 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_sse2_double
 712                     (t_nblist                    * gmx_restrict       nlist,
 713                      rvec                        * gmx_restrict          xx,
 714                      rvec                        * gmx_restrict          ff,
 715                      struct t_forcerec           * gmx_restrict          fr,
 716                      t_mdatoms                   * gmx_restrict     mdatoms,
 717                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 718                      t_nrnb                      * gmx_restrict        nrnb)
 719 {
 720     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 721      * just 0 for non-waters.
 722      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 723      * jnr indices corresponding to data put in the four positions in the SIMD register.
 724      */
 725     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 726     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 727     int              jnrA,jnrB;
 728     int              j_coord_offsetA,j_coord_offsetB;
 729     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 730     real             rcutoff_scalar;
 731     real             *shiftvec,*fshift,*x,*f;
 732     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 733     int              vdwioffset0;
 734     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 735     int              vdwioffset1;
 736     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 737     int              vdwioffset2;
 738     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 739     int              vdwjidx0A,vdwjidx0B;
 740     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 741     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 742     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 743     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 744     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 745     real             *charge;
 746     int              nvdwtype;
 747     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 748     int              *vdwtype;
 749     real             *vdwparam;
 750     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 751     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 752     __m128d           c6grid_00;
 753     __m128d           c6grid_10;
 754     __m128d           c6grid_20;
 755     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 756     real             *vdwgridparam;
 757     __m128d           one_half = _mm_set1_pd(0.5);
 758     __m128d           minus_one = _mm_set1_pd(-1.0);
 759     __m128i          ewitab;
 760     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 761     real             *ewtab;
 762     __m128d          dummy_mask,cutoff_mask;
 763     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 764     __m128d          one     = _mm_set1_pd(1.0);
 765     __m128d          two     = _mm_set1_pd(2.0);
 766     x                = xx[0];
 767     f                = ff[0];
 768
 769     nri              = nlist->nri;
 770     iinr             = nlist->iinr;
 771     jindex           = nlist->jindex;
 772     jjnr             = nlist->jjnr;
 773     shiftidx         = nlist->shift;
 774     gid              = nlist->gid;
 775     shiftvec         = fr->shift_vec[0];
 776     fshift           = fr->fshift[0];
 777     facel            = _mm_set1_pd(fr->ic->epsfac);
 778     charge           = mdatoms->chargeA;
 779     nvdwtype         = fr->ntype;
 780     vdwparam         = fr->nbfp;
 781     vdwtype          = mdatoms->typeA;
 782     vdwgridparam     = fr->ljpme_c6grid;
 783     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 784     ewclj            = _mm_set1_pd(fr->ic->ewaldcoeff_lj);
 785     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 786
 787     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 788     ewtab            = fr->ic->tabq_coul_F;
 789     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 790     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 791
 792     /* Setup water-specific parameters */
 793     inr              = nlist->iinr[0];
 794     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 795     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 796     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 797     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 798
 799     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 800     rcutoff_scalar   = fr->ic->rcoulomb;
 801     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 802     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 803
 804     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 805     rvdw             = _mm_set1_pd(fr->ic->rvdw);
 806
 807     /* Avoid stupid compiler warnings */
 808     jnrA = jnrB = 0;
 809     j_coord_offsetA = 0;
 810     j_coord_offsetB = 0;
 811
 812     outeriter        = 0;
 813     inneriter        = 0;
 814
 815     /* Start outer loop over neighborlists */
 816     for(iidx=0; iidx<nri; iidx++)
 817     {
 818         /* Load shift vector for this list */
 819         i_shift_offset   = DIM*shiftidx[iidx];
 820
 821         /* Load limits for loop over neighbors */
 822         j_index_start    = jindex[iidx];
 823         j_index_end      = jindex[iidx+1];
 824
 825         /* Get outer coordinate index */
 826         inr              = iinr[iidx];
 827         i_coord_offset   = DIM*inr;
 828
 829         /* Load i particle coords and add shift vector */
 830         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 831                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 832
 833         fix0             = _mm_setzero_pd();
 834         fiy0             = _mm_setzero_pd();
 835         fiz0             = _mm_setzero_pd();
 836         fix1             = _mm_setzero_pd();
 837         fiy1             = _mm_setzero_pd();
 838         fiz1             = _mm_setzero_pd();
 839         fix2             = _mm_setzero_pd();
 840         fiy2             = _mm_setzero_pd();
 841         fiz2             = _mm_setzero_pd();
 842
 843         /* Start inner kernel loop */
 844         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 845         {
 846
 847             /* Get j neighbor index, and coordinate index */
 848             jnrA             = jjnr[jidx];
 849             jnrB             = jjnr[jidx+1];
 850             j_coord_offsetA  = DIM*jnrA;
 851             j_coord_offsetB  = DIM*jnrB;
 852
 853             /* load j atom coordinates */
 854             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 855                                               &jx0,&jy0,&jz0);
 856
 857             /* Calculate displacement vector */
 858             dx00             = _mm_sub_pd(ix0,jx0);
 859             dy00             = _mm_sub_pd(iy0,jy0);
 860             dz00             = _mm_sub_pd(iz0,jz0);
 861             dx10             = _mm_sub_pd(ix1,jx0);
 862             dy10             = _mm_sub_pd(iy1,jy0);
 863             dz10             = _mm_sub_pd(iz1,jz0);
 864             dx20             = _mm_sub_pd(ix2,jx0);
 865             dy20             = _mm_sub_pd(iy2,jy0);
 866             dz20             = _mm_sub_pd(iz2,jz0);
 867
 868             /* Calculate squared distance and things based on it */
 869             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 870             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 871             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 872
 873             rinv00           = sse2_invsqrt_d(rsq00);
 874             rinv10           = sse2_invsqrt_d(rsq10);
 875             rinv20           = sse2_invsqrt_d(rsq20);
 876
 877             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 878             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 879             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 880
 881             /* Load parameters for j particles */
 882             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 883             vdwjidx0A        = 2*vdwtype[jnrA+0];
 884             vdwjidx0B        = 2*vdwtype[jnrB+0];
 885
 886             fjx0             = _mm_setzero_pd();
 887             fjy0             = _mm_setzero_pd();
 888             fjz0             = _mm_setzero_pd();
 889
 890             /**************************
 891              * CALCULATE INTERACTIONS *
 892              **************************/
 893
 894             if (gmx_mm_any_lt(rsq00,rcutoff2))
 895             {
 896
 897             r00              = _mm_mul_pd(rsq00,rinv00);
 898
 899             /* Compute parameters for interactions between i and j atoms */
 900             qq00             = _mm_mul_pd(iq0,jq0);
 901             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 902                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 903
 904             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 905                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 906
 907             /* EWALD ELECTROSTATICS */
 908
 909             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 910             ewrt             = _mm_mul_pd(r00,ewtabscale);
 911             ewitab           = _mm_cvttpd_epi32(ewrt);
 912             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 913             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 914                                          &ewtabF,&ewtabFn);
 915             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 916             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 917
 918             /* Analytical LJ-PME */
 919             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 920             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 921             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 922             exponent         = sse2_exp_d(ewcljrsq);
 923             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 924             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 925             /* f6A = 6 * C6grid * (1 - poly) */
 926             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 927             /* f6B = C6grid * exponent * beta^6 */
 928             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 929             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 930             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);
 931
 932             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 933
 934             fscal            = _mm_add_pd(felec,fvdw);
 935
 936             fscal            = _mm_and_pd(fscal,cutoff_mask);
 937
 938             /* Calculate temporary vectorial force */
 939             tx               = _mm_mul_pd(fscal,dx00);
 940             ty               = _mm_mul_pd(fscal,dy00);
 941             tz               = _mm_mul_pd(fscal,dz00);
 942
 943             /* Update vectorial force */
 944             fix0             = _mm_add_pd(fix0,tx);
 945             fiy0             = _mm_add_pd(fiy0,ty);
 946             fiz0             = _mm_add_pd(fiz0,tz);
 947
 948             fjx0             = _mm_add_pd(fjx0,tx);
 949             fjy0             = _mm_add_pd(fjy0,ty);
 950             fjz0             = _mm_add_pd(fjz0,tz);
 951
 952             }
 953
 954             /**************************
 955              * CALCULATE INTERACTIONS *
 956              **************************/
 957
 958             if (gmx_mm_any_lt(rsq10,rcutoff2))
 959             {
 960
 961             r10              = _mm_mul_pd(rsq10,rinv10);
 962
 963             /* Compute parameters for interactions between i and j atoms */
 964             qq10             = _mm_mul_pd(iq1,jq0);
 965
 966             /* EWALD ELECTROSTATICS */
 967
 968             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 969             ewrt             = _mm_mul_pd(r10,ewtabscale);
 970             ewitab           = _mm_cvttpd_epi32(ewrt);
 971             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 972             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 973                                          &ewtabF,&ewtabFn);
 974             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 975             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 976
 977             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 978
 979             fscal            = felec;
 980
 981             fscal            = _mm_and_pd(fscal,cutoff_mask);
 982
 983             /* Calculate temporary vectorial force */
 984             tx               = _mm_mul_pd(fscal,dx10);
 985             ty               = _mm_mul_pd(fscal,dy10);
 986             tz               = _mm_mul_pd(fscal,dz10);
 987
 988             /* Update vectorial force */
 989             fix1             = _mm_add_pd(fix1,tx);
 990             fiy1             = _mm_add_pd(fiy1,ty);
 991             fiz1             = _mm_add_pd(fiz1,tz);
 992
 993             fjx0             = _mm_add_pd(fjx0,tx);
 994             fjy0             = _mm_add_pd(fjy0,ty);
 995             fjz0             = _mm_add_pd(fjz0,tz);
 996
 997             }
 998
 999             /**************************
1000              * CALCULATE INTERACTIONS *
1001              **************************/
1002
1003             if (gmx_mm_any_lt(rsq20,rcutoff2))
1004             {
1005
1006             r20              = _mm_mul_pd(rsq20,rinv20);
1007
1008             /* Compute parameters for interactions between i and j atoms */
1009             qq20             = _mm_mul_pd(iq2,jq0);
1010
1011             /* EWALD ELECTROSTATICS */
1012
1013             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1014             ewrt             = _mm_mul_pd(r20,ewtabscale);
1015             ewitab           = _mm_cvttpd_epi32(ewrt);
1016             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1017             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1018                                          &ewtabF,&ewtabFn);
1019             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1020             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1021
1022             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1023
1024             fscal            = felec;
1025
1026             fscal            = _mm_and_pd(fscal,cutoff_mask);
1027
1028             /* Calculate temporary vectorial force */
1029             tx               = _mm_mul_pd(fscal,dx20);
1030             ty               = _mm_mul_pd(fscal,dy20);
1031             tz               = _mm_mul_pd(fscal,dz20);
1032
1033             /* Update vectorial force */
1034             fix2             = _mm_add_pd(fix2,tx);
1035             fiy2             = _mm_add_pd(fiy2,ty);
1036             fiz2             = _mm_add_pd(fiz2,tz);
1037
1038             fjx0             = _mm_add_pd(fjx0,tx);
1039             fjy0             = _mm_add_pd(fjy0,ty);
1040             fjz0             = _mm_add_pd(fjz0,tz);
1041
1042             }
1043
1044             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1045
1046             /* Inner loop uses 143 flops */
1047         }
1048
1049         if(jidx<j_index_end)
1050         {
1051
1052             jnrA             = jjnr[jidx];
1053             j_coord_offsetA  = DIM*jnrA;
1054
1055             /* load j atom coordinates */
1056             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1057                                               &jx0,&jy0,&jz0);
1058
1059             /* Calculate displacement vector */
1060             dx00             = _mm_sub_pd(ix0,jx0);
1061             dy00             = _mm_sub_pd(iy0,jy0);
1062             dz00             = _mm_sub_pd(iz0,jz0);
1063             dx10             = _mm_sub_pd(ix1,jx0);
1064             dy10             = _mm_sub_pd(iy1,jy0);
1065             dz10             = _mm_sub_pd(iz1,jz0);
1066             dx20             = _mm_sub_pd(ix2,jx0);
1067             dy20             = _mm_sub_pd(iy2,jy0);
1068             dz20             = _mm_sub_pd(iz2,jz0);
1069
1070             /* Calculate squared distance and things based on it */
1071             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1072             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1073             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1074
1075             rinv00           = sse2_invsqrt_d(rsq00);
1076             rinv10           = sse2_invsqrt_d(rsq10);
1077             rinv20           = sse2_invsqrt_d(rsq20);
1078
1079             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1080             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1081             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1082
1083             /* Load parameters for j particles */
1084             jq0              = _mm_load_sd(charge+jnrA+0);
1085             vdwjidx0A        = 2*vdwtype[jnrA+0];
1086
1087             fjx0             = _mm_setzero_pd();
1088             fjy0             = _mm_setzero_pd();
1089             fjz0             = _mm_setzero_pd();
1090
1091             /**************************
1092              * CALCULATE INTERACTIONS *
1093              **************************/
1094
1095             if (gmx_mm_any_lt(rsq00,rcutoff2))
1096             {
1097
1098             r00              = _mm_mul_pd(rsq00,rinv00);
1099
1100             /* Compute parameters for interactions between i and j atoms */
1101             qq00             = _mm_mul_pd(iq0,jq0);
1102             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1103
1104             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
1105
1106             /* EWALD ELECTROSTATICS */
1107
1108             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1109             ewrt             = _mm_mul_pd(r00,ewtabscale);
1110             ewitab           = _mm_cvttpd_epi32(ewrt);
1111             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1112             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1113             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1114             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1115
1116             /* Analytical LJ-PME */
1117             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1118             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1119             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1120             exponent         = sse2_exp_d(ewcljrsq);
1121             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1122             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1123             /* f6A = 6 * C6grid * (1 - poly) */
1124             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1125             /* f6B = C6grid * exponent * beta^6 */
1126             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1127             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1128             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);
1129
1130             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1131
1132             fscal            = _mm_add_pd(felec,fvdw);
1133
1134             fscal            = _mm_and_pd(fscal,cutoff_mask);
1135
1136             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1137
1138             /* Calculate temporary vectorial force */
1139             tx               = _mm_mul_pd(fscal,dx00);
1140             ty               = _mm_mul_pd(fscal,dy00);
1141             tz               = _mm_mul_pd(fscal,dz00);
1142
1143             /* Update vectorial force */
1144             fix0             = _mm_add_pd(fix0,tx);
1145             fiy0             = _mm_add_pd(fiy0,ty);
1146             fiz0             = _mm_add_pd(fiz0,tz);
1147
1148             fjx0             = _mm_add_pd(fjx0,tx);
1149             fjy0             = _mm_add_pd(fjy0,ty);
1150             fjz0             = _mm_add_pd(fjz0,tz);
1151
1152             }
1153
1154             /**************************
1155              * CALCULATE INTERACTIONS *
1156              **************************/
1157
1158             if (gmx_mm_any_lt(rsq10,rcutoff2))
1159             {
1160
1161             r10              = _mm_mul_pd(rsq10,rinv10);
1162
1163             /* Compute parameters for interactions between i and j atoms */
1164             qq10             = _mm_mul_pd(iq1,jq0);
1165
1166             /* EWALD ELECTROSTATICS */
1167
1168             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1169             ewrt             = _mm_mul_pd(r10,ewtabscale);
1170             ewitab           = _mm_cvttpd_epi32(ewrt);
1171             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1172             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1173             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1174             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1175
1176             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1177
1178             fscal            = felec;
1179
1180             fscal            = _mm_and_pd(fscal,cutoff_mask);
1181
1182             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1183
1184             /* Calculate temporary vectorial force */
1185             tx               = _mm_mul_pd(fscal,dx10);
1186             ty               = _mm_mul_pd(fscal,dy10);
1187             tz               = _mm_mul_pd(fscal,dz10);
1188
1189             /* Update vectorial force */
1190             fix1             = _mm_add_pd(fix1,tx);
1191             fiy1             = _mm_add_pd(fiy1,ty);
1192             fiz1             = _mm_add_pd(fiz1,tz);
1193
1194             fjx0             = _mm_add_pd(fjx0,tx);
1195             fjy0             = _mm_add_pd(fjy0,ty);
1196             fjz0             = _mm_add_pd(fjz0,tz);
1197
1198             }
1199
1200             /**************************
1201              * CALCULATE INTERACTIONS *
1202              **************************/
1203
1204             if (gmx_mm_any_lt(rsq20,rcutoff2))
1205             {
1206
1207             r20              = _mm_mul_pd(rsq20,rinv20);
1208
1209             /* Compute parameters for interactions between i and j atoms */
1210             qq20             = _mm_mul_pd(iq2,jq0);
1211
1212             /* EWALD ELECTROSTATICS */
1213
1214             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1215             ewrt             = _mm_mul_pd(r20,ewtabscale);
1216             ewitab           = _mm_cvttpd_epi32(ewrt);
1217             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1218             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1219             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1220             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1221
1222             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1223
1224             fscal            = felec;
1225
1226             fscal            = _mm_and_pd(fscal,cutoff_mask);
1227
1228             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1229
1230             /* Calculate temporary vectorial force */
1231             tx               = _mm_mul_pd(fscal,dx20);
1232             ty               = _mm_mul_pd(fscal,dy20);
1233             tz               = _mm_mul_pd(fscal,dz20);
1234
1235             /* Update vectorial force */
1236             fix2             = _mm_add_pd(fix2,tx);
1237             fiy2             = _mm_add_pd(fiy2,ty);
1238             fiz2             = _mm_add_pd(fiz2,tz);
1239
1240             fjx0             = _mm_add_pd(fjx0,tx);
1241             fjy0             = _mm_add_pd(fjy0,ty);
1242             fjz0             = _mm_add_pd(fjz0,tz);
1243
1244             }
1245
1246             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1247
1248             /* Inner loop uses 143 flops */
1249         }
1250
1251         /* End of innermost loop */
1252
1253         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1254                                               f+i_coord_offset,fshift+i_shift_offset);
1255
1256         /* Increment number of inner iterations */
1257         inneriter                  += j_index_end - j_index_start;
1258
1259         /* Outer loop uses 18 flops */
1260     }
1261
1262     /* Increment number of outer iterations */
1263     outeriter        += nri;
1264
1265     /* Update outer/inner flops */
1266
1267     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*143);
1268 }