src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_sse4_1_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_double.h"
  50 #include "kernelutil_x86_sse4_1_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_sse4_1_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_sse4_1_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 100     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 101     __m128d           c6grid_00;
 102     __m128d           c6grid_10;
 103     __m128d           c6grid_20;
 104     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 105     real             *vdwgridparam;
 106     __m128d           one_half = _mm_set1_pd(0.5);
 107     __m128d           minus_one = _mm_set1_pd(-1.0);
 108     __m128i          ewitab;
 109     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 110     real             *ewtab;
 111     __m128d          dummy_mask,cutoff_mask;
 112     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 113     __m128d          one     = _mm_set1_pd(1.0);
 114     __m128d          two     = _mm_set1_pd(2.0);
 115     x                = xx[0];
 116     f                = ff[0];
 117
 118     nri              = nlist->nri;
 119     iinr             = nlist->iinr;
 120     jindex           = nlist->jindex;
 121     jjnr             = nlist->jjnr;
 122     shiftidx         = nlist->shift;
 123     gid              = nlist->gid;
 124     shiftvec         = fr->shift_vec[0];
 125     fshift           = fr->fshift[0];
 126     facel            = _mm_set1_pd(fr->epsfac);
 127     charge           = mdatoms->chargeA;
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131     vdwgridparam     = fr->ljpme_c6grid;
 132     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 133     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 134     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 135
 136     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 137     ewtab            = fr->ic->tabq_coul_FDV0;
 138     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 139     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 140
 141     /* Setup water-specific parameters */
 142     inr              = nlist->iinr[0];
 143     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 144     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 145     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 146     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 147
 148     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 149     rcutoff_scalar   = fr->rcoulomb;
 150     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 151     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 152
 153     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 154     rvdw             = _mm_set1_pd(fr->rvdw);
 155
 156     /* Avoid stupid compiler warnings */
 157     jnrA = jnrB = 0;
 158     j_coord_offsetA = 0;
 159     j_coord_offsetB = 0;
 160
 161     outeriter        = 0;
 162     inneriter        = 0;
 163
 164     /* Start outer loop over neighborlists */
 165     for(iidx=0; iidx<nri; iidx++)
 166     {
 167         /* Load shift vector for this list */
 168         i_shift_offset   = DIM*shiftidx[iidx];
 169
 170         /* Load limits for loop over neighbors */
 171         j_index_start    = jindex[iidx];
 172         j_index_end      = jindex[iidx+1];
 173
 174         /* Get outer coordinate index */
 175         inr              = iinr[iidx];
 176         i_coord_offset   = DIM*inr;
 177
 178         /* Load i particle coords and add shift vector */
 179         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 180                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 181
 182         fix0             = _mm_setzero_pd();
 183         fiy0             = _mm_setzero_pd();
 184         fiz0             = _mm_setzero_pd();
 185         fix1             = _mm_setzero_pd();
 186         fiy1             = _mm_setzero_pd();
 187         fiz1             = _mm_setzero_pd();
 188         fix2             = _mm_setzero_pd();
 189         fiy2             = _mm_setzero_pd();
 190         fiz2             = _mm_setzero_pd();
 191
 192         /* Reset potential sums */
 193         velecsum         = _mm_setzero_pd();
 194         vvdwsum          = _mm_setzero_pd();
 195
 196         /* Start inner kernel loop */
 197         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 198         {
 199
 200             /* Get j neighbor index, and coordinate index */
 201             jnrA             = jjnr[jidx];
 202             jnrB             = jjnr[jidx+1];
 203             j_coord_offsetA  = DIM*jnrA;
 204             j_coord_offsetB  = DIM*jnrB;
 205
 206             /* load j atom coordinates */
 207             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 208                                               &jx0,&jy0,&jz0);
 209
 210             /* Calculate displacement vector */
 211             dx00             = _mm_sub_pd(ix0,jx0);
 212             dy00             = _mm_sub_pd(iy0,jy0);
 213             dz00             = _mm_sub_pd(iz0,jz0);
 214             dx10             = _mm_sub_pd(ix1,jx0);
 215             dy10             = _mm_sub_pd(iy1,jy0);
 216             dz10             = _mm_sub_pd(iz1,jz0);
 217             dx20             = _mm_sub_pd(ix2,jx0);
 218             dy20             = _mm_sub_pd(iy2,jy0);
 219             dz20             = _mm_sub_pd(iz2,jz0);
 220
 221             /* Calculate squared distance and things based on it */
 222             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 223             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 224             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 225
 226             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 227             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 228             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 229
 230             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 231             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 232             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 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             r00              = _mm_mul_pd(rsq00,rinv00);
 251
 252             /* Compute parameters for interactions between i and j atoms */
 253             qq00             = _mm_mul_pd(iq0,jq0);
 254             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 255                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 256             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 257                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 258
 259             /* EWALD ELECTROSTATICS */
 260
 261             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 262             ewrt             = _mm_mul_pd(r00,ewtabscale);
 263             ewitab           = _mm_cvttpd_epi32(ewrt);
 264             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 265             ewitab           = _mm_slli_epi32(ewitab,2);
 266             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 267             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 268             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 269             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 270             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 271             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 272             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 273             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 274             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 275             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 276
 277             /* Analytical LJ-PME */
 278             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 279             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 280             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 281             exponent         = gmx_simd_exp_d(ewcljrsq);
 282             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 283             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 284             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 285             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 286             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 287             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),
 288                                _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));
 289             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 290             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);
 291
 292             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 293
 294             /* Update potential sum for this i atom from the interaction with this j atom. */
 295             velec            = _mm_and_pd(velec,cutoff_mask);
 296             velecsum         = _mm_add_pd(velecsum,velec);
 297             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 298             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 299
 300             fscal            = _mm_add_pd(felec,fvdw);
 301
 302             fscal            = _mm_and_pd(fscal,cutoff_mask);
 303
 304             /* Calculate temporary vectorial force */
 305             tx               = _mm_mul_pd(fscal,dx00);
 306             ty               = _mm_mul_pd(fscal,dy00);
 307             tz               = _mm_mul_pd(fscal,dz00);
 308
 309             /* Update vectorial force */
 310             fix0             = _mm_add_pd(fix0,tx);
 311             fiy0             = _mm_add_pd(fiy0,ty);
 312             fiz0             = _mm_add_pd(fiz0,tz);
 313
 314             fjx0             = _mm_add_pd(fjx0,tx);
 315             fjy0             = _mm_add_pd(fjy0,ty);
 316             fjz0             = _mm_add_pd(fjz0,tz);
 317
 318             }
 319
 320             /**************************
 321              * CALCULATE INTERACTIONS *
 322              **************************/
 323
 324             if (gmx_mm_any_lt(rsq10,rcutoff2))
 325             {
 326
 327             r10              = _mm_mul_pd(rsq10,rinv10);
 328
 329             /* Compute parameters for interactions between i and j atoms */
 330             qq10             = _mm_mul_pd(iq1,jq0);
 331
 332             /* EWALD ELECTROSTATICS */
 333
 334             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 335             ewrt             = _mm_mul_pd(r10,ewtabscale);
 336             ewitab           = _mm_cvttpd_epi32(ewrt);
 337             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 338             ewitab           = _mm_slli_epi32(ewitab,2);
 339             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 340             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 341             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 342             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 343             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 344             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 345             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 346             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 347             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 348             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 349
 350             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 351
 352             /* Update potential sum for this i atom from the interaction with this j atom. */
 353             velec            = _mm_and_pd(velec,cutoff_mask);
 354             velecsum         = _mm_add_pd(velecsum,velec);
 355
 356             fscal            = felec;
 357
 358             fscal            = _mm_and_pd(fscal,cutoff_mask);
 359
 360             /* Calculate temporary vectorial force */
 361             tx               = _mm_mul_pd(fscal,dx10);
 362             ty               = _mm_mul_pd(fscal,dy10);
 363             tz               = _mm_mul_pd(fscal,dz10);
 364
 365             /* Update vectorial force */
 366             fix1             = _mm_add_pd(fix1,tx);
 367             fiy1             = _mm_add_pd(fiy1,ty);
 368             fiz1             = _mm_add_pd(fiz1,tz);
 369
 370             fjx0             = _mm_add_pd(fjx0,tx);
 371             fjy0             = _mm_add_pd(fjy0,ty);
 372             fjz0             = _mm_add_pd(fjz0,tz);
 373
 374             }
 375
 376             /**************************
 377              * CALCULATE INTERACTIONS *
 378              **************************/
 379
 380             if (gmx_mm_any_lt(rsq20,rcutoff2))
 381             {
 382
 383             r20              = _mm_mul_pd(rsq20,rinv20);
 384
 385             /* Compute parameters for interactions between i and j atoms */
 386             qq20             = _mm_mul_pd(iq2,jq0);
 387
 388             /* EWALD ELECTROSTATICS */
 389
 390             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 391             ewrt             = _mm_mul_pd(r20,ewtabscale);
 392             ewitab           = _mm_cvttpd_epi32(ewrt);
 393             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 394             ewitab           = _mm_slli_epi32(ewitab,2);
 395             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 396             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 397             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 398             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 399             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 400             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 401             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 402             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 403             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 404             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 405
 406             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 407
 408             /* Update potential sum for this i atom from the interaction with this j atom. */
 409             velec            = _mm_and_pd(velec,cutoff_mask);
 410             velecsum         = _mm_add_pd(velecsum,velec);
 411
 412             fscal            = felec;
 413
 414             fscal            = _mm_and_pd(fscal,cutoff_mask);
 415
 416             /* Calculate temporary vectorial force */
 417             tx               = _mm_mul_pd(fscal,dx20);
 418             ty               = _mm_mul_pd(fscal,dy20);
 419             tz               = _mm_mul_pd(fscal,dz20);
 420
 421             /* Update vectorial force */
 422             fix2             = _mm_add_pd(fix2,tx);
 423             fiy2             = _mm_add_pd(fiy2,ty);
 424             fiz2             = _mm_add_pd(fiz2,tz);
 425
 426             fjx0             = _mm_add_pd(fjx0,tx);
 427             fjy0             = _mm_add_pd(fjy0,ty);
 428             fjz0             = _mm_add_pd(fjz0,tz);
 429
 430             }
 431
 432             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 433
 434             /* Inner loop uses 176 flops */
 435         }
 436
 437         if(jidx<j_index_end)
 438         {
 439
 440             jnrA             = jjnr[jidx];
 441             j_coord_offsetA  = DIM*jnrA;
 442
 443             /* load j atom coordinates */
 444             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 445                                               &jx0,&jy0,&jz0);
 446
 447             /* Calculate displacement vector */
 448             dx00             = _mm_sub_pd(ix0,jx0);
 449             dy00             = _mm_sub_pd(iy0,jy0);
 450             dz00             = _mm_sub_pd(iz0,jz0);
 451             dx10             = _mm_sub_pd(ix1,jx0);
 452             dy10             = _mm_sub_pd(iy1,jy0);
 453             dz10             = _mm_sub_pd(iz1,jz0);
 454             dx20             = _mm_sub_pd(ix2,jx0);
 455             dy20             = _mm_sub_pd(iy2,jy0);
 456             dz20             = _mm_sub_pd(iz2,jz0);
 457
 458             /* Calculate squared distance and things based on it */
 459             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 460             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 461             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 462
 463             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 464             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 465             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 466
 467             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 468             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 469             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 470
 471             /* Load parameters for j particles */
 472             jq0              = _mm_load_sd(charge+jnrA+0);
 473             vdwjidx0A        = 2*vdwtype[jnrA+0];
 474
 475             fjx0             = _mm_setzero_pd();
 476             fjy0             = _mm_setzero_pd();
 477             fjz0             = _mm_setzero_pd();
 478
 479             /**************************
 480              * CALCULATE INTERACTIONS *
 481              **************************/
 482
 483             if (gmx_mm_any_lt(rsq00,rcutoff2))
 484             {
 485
 486             r00              = _mm_mul_pd(rsq00,rinv00);
 487
 488             /* Compute parameters for interactions between i and j atoms */
 489             qq00             = _mm_mul_pd(iq0,jq0);
 490             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 491
 492             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 493
 494             /* EWALD ELECTROSTATICS */
 495
 496             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 497             ewrt             = _mm_mul_pd(r00,ewtabscale);
 498             ewitab           = _mm_cvttpd_epi32(ewrt);
 499             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 500             ewitab           = _mm_slli_epi32(ewitab,2);
 501             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 502             ewtabD           = _mm_setzero_pd();
 503             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 504             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 505             ewtabFn          = _mm_setzero_pd();
 506             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 507             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 508             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 509             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 510             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 511
 512             /* Analytical LJ-PME */
 513             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 514             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 515             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 516             exponent         = gmx_simd_exp_d(ewcljrsq);
 517             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 518             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 519             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 520             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 521             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 522             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),
 523                                _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));
 524             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 525             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);
 526
 527             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 528
 529             /* Update potential sum for this i atom from the interaction with this j atom. */
 530             velec            = _mm_and_pd(velec,cutoff_mask);
 531             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 532             velecsum         = _mm_add_pd(velecsum,velec);
 533             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 534             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 535             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 536
 537             fscal            = _mm_add_pd(felec,fvdw);
 538
 539             fscal            = _mm_and_pd(fscal,cutoff_mask);
 540
 541             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 542
 543             /* Calculate temporary vectorial force */
 544             tx               = _mm_mul_pd(fscal,dx00);
 545             ty               = _mm_mul_pd(fscal,dy00);
 546             tz               = _mm_mul_pd(fscal,dz00);
 547
 548             /* Update vectorial force */
 549             fix0             = _mm_add_pd(fix0,tx);
 550             fiy0             = _mm_add_pd(fiy0,ty);
 551             fiz0             = _mm_add_pd(fiz0,tz);
 552
 553             fjx0             = _mm_add_pd(fjx0,tx);
 554             fjy0             = _mm_add_pd(fjy0,ty);
 555             fjz0             = _mm_add_pd(fjz0,tz);
 556
 557             }
 558
 559             /**************************
 560              * CALCULATE INTERACTIONS *
 561              **************************/
 562
 563             if (gmx_mm_any_lt(rsq10,rcutoff2))
 564             {
 565
 566             r10              = _mm_mul_pd(rsq10,rinv10);
 567
 568             /* Compute parameters for interactions between i and j atoms */
 569             qq10             = _mm_mul_pd(iq1,jq0);
 570
 571             /* EWALD ELECTROSTATICS */
 572
 573             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 574             ewrt             = _mm_mul_pd(r10,ewtabscale);
 575             ewitab           = _mm_cvttpd_epi32(ewrt);
 576             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 577             ewitab           = _mm_slli_epi32(ewitab,2);
 578             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 579             ewtabD           = _mm_setzero_pd();
 580             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 581             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 582             ewtabFn          = _mm_setzero_pd();
 583             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 584             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 585             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 586             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 587             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 588
 589             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 590
 591             /* Update potential sum for this i atom from the interaction with this j atom. */
 592             velec            = _mm_and_pd(velec,cutoff_mask);
 593             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 594             velecsum         = _mm_add_pd(velecsum,velec);
 595
 596             fscal            = felec;
 597
 598             fscal            = _mm_and_pd(fscal,cutoff_mask);
 599
 600             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 601
 602             /* Calculate temporary vectorial force */
 603             tx               = _mm_mul_pd(fscal,dx10);
 604             ty               = _mm_mul_pd(fscal,dy10);
 605             tz               = _mm_mul_pd(fscal,dz10);
 606
 607             /* Update vectorial force */
 608             fix1             = _mm_add_pd(fix1,tx);
 609             fiy1             = _mm_add_pd(fiy1,ty);
 610             fiz1             = _mm_add_pd(fiz1,tz);
 611
 612             fjx0             = _mm_add_pd(fjx0,tx);
 613             fjy0             = _mm_add_pd(fjy0,ty);
 614             fjz0             = _mm_add_pd(fjz0,tz);
 615
 616             }
 617
 618             /**************************
 619              * CALCULATE INTERACTIONS *
 620              **************************/
 621
 622             if (gmx_mm_any_lt(rsq20,rcutoff2))
 623             {
 624
 625             r20              = _mm_mul_pd(rsq20,rinv20);
 626
 627             /* Compute parameters for interactions between i and j atoms */
 628             qq20             = _mm_mul_pd(iq2,jq0);
 629
 630             /* EWALD ELECTROSTATICS */
 631
 632             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 633             ewrt             = _mm_mul_pd(r20,ewtabscale);
 634             ewitab           = _mm_cvttpd_epi32(ewrt);
 635             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 636             ewitab           = _mm_slli_epi32(ewitab,2);
 637             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 638             ewtabD           = _mm_setzero_pd();
 639             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 640             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 641             ewtabFn          = _mm_setzero_pd();
 642             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 643             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 644             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 645             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 646             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 647
 648             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 649
 650             /* Update potential sum for this i atom from the interaction with this j atom. */
 651             velec            = _mm_and_pd(velec,cutoff_mask);
 652             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 653             velecsum         = _mm_add_pd(velecsum,velec);
 654
 655             fscal            = felec;
 656
 657             fscal            = _mm_and_pd(fscal,cutoff_mask);
 658
 659             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 660
 661             /* Calculate temporary vectorial force */
 662             tx               = _mm_mul_pd(fscal,dx20);
 663             ty               = _mm_mul_pd(fscal,dy20);
 664             tz               = _mm_mul_pd(fscal,dz20);
 665
 666             /* Update vectorial force */
 667             fix2             = _mm_add_pd(fix2,tx);
 668             fiy2             = _mm_add_pd(fiy2,ty);
 669             fiz2             = _mm_add_pd(fiz2,tz);
 670
 671             fjx0             = _mm_add_pd(fjx0,tx);
 672             fjy0             = _mm_add_pd(fjy0,ty);
 673             fjz0             = _mm_add_pd(fjz0,tz);
 674
 675             }
 676
 677             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 678
 679             /* Inner loop uses 176 flops */
 680         }
 681
 682         /* End of innermost loop */
 683
 684         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 685                                               f+i_coord_offset,fshift+i_shift_offset);
 686
 687         ggid                        = gid[iidx];
 688         /* Update potential energies */
 689         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 690         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 691
 692         /* Increment number of inner iterations */
 693         inneriter                  += j_index_end - j_index_start;
 694
 695         /* Outer loop uses 20 flops */
 696     }
 697
 698     /* Increment number of outer iterations */
 699     outeriter        += nri;
 700
 701     /* Update outer/inner flops */
 702
 703     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*176);
 704 }
 705 /*
 706  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_sse4_1_double
 707  * Electrostatics interaction: Ewald
 708  * VdW interaction:            LJEwald
 709  * Geometry:                   Water3-Particle
 710  * Calculate force/pot:        Force
 711  */
 712 void
 713 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_sse4_1_double
 714                     (t_nblist                    * gmx_restrict       nlist,
 715                      rvec                        * gmx_restrict          xx,
 716                      rvec                        * gmx_restrict          ff,
 717                      t_forcerec                  * gmx_restrict          fr,
 718                      t_mdatoms                   * gmx_restrict     mdatoms,
 719                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 720                      t_nrnb                      * gmx_restrict        nrnb)
 721 {
 722     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 723      * just 0 for non-waters.
 724      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 725      * jnr indices corresponding to data put in the four positions in the SIMD register.
 726      */
 727     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 728     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 729     int              jnrA,jnrB;
 730     int              j_coord_offsetA,j_coord_offsetB;
 731     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 732     real             rcutoff_scalar;
 733     real             *shiftvec,*fshift,*x,*f;
 734     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 735     int              vdwioffset0;
 736     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 737     int              vdwioffset1;
 738     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 739     int              vdwioffset2;
 740     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 741     int              vdwjidx0A,vdwjidx0B;
 742     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 743     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 744     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 745     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 746     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 747     real             *charge;
 748     int              nvdwtype;
 749     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 750     int              *vdwtype;
 751     real             *vdwparam;
 752     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 753     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 754     __m128d           c6grid_00;
 755     __m128d           c6grid_10;
 756     __m128d           c6grid_20;
 757     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 758     real             *vdwgridparam;
 759     __m128d           one_half = _mm_set1_pd(0.5);
 760     __m128d           minus_one = _mm_set1_pd(-1.0);
 761     __m128i          ewitab;
 762     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 763     real             *ewtab;
 764     __m128d          dummy_mask,cutoff_mask;
 765     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 766     __m128d          one     = _mm_set1_pd(1.0);
 767     __m128d          two     = _mm_set1_pd(2.0);
 768     x                = xx[0];
 769     f                = ff[0];
 770
 771     nri              = nlist->nri;
 772     iinr             = nlist->iinr;
 773     jindex           = nlist->jindex;
 774     jjnr             = nlist->jjnr;
 775     shiftidx         = nlist->shift;
 776     gid              = nlist->gid;
 777     shiftvec         = fr->shift_vec[0];
 778     fshift           = fr->fshift[0];
 779     facel            = _mm_set1_pd(fr->epsfac);
 780     charge           = mdatoms->chargeA;
 781     nvdwtype         = fr->ntype;
 782     vdwparam         = fr->nbfp;
 783     vdwtype          = mdatoms->typeA;
 784     vdwgridparam     = fr->ljpme_c6grid;
 785     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 786     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 787     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 788
 789     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 790     ewtab            = fr->ic->tabq_coul_F;
 791     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 792     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 793
 794     /* Setup water-specific parameters */
 795     inr              = nlist->iinr[0];
 796     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 797     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 798     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 799     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 800
 801     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 802     rcutoff_scalar   = fr->rcoulomb;
 803     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 804     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 805
 806     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 807     rvdw             = _mm_set1_pd(fr->rvdw);
 808
 809     /* Avoid stupid compiler warnings */
 810     jnrA = jnrB = 0;
 811     j_coord_offsetA = 0;
 812     j_coord_offsetB = 0;
 813
 814     outeriter        = 0;
 815     inneriter        = 0;
 816
 817     /* Start outer loop over neighborlists */
 818     for(iidx=0; iidx<nri; iidx++)
 819     {
 820         /* Load shift vector for this list */
 821         i_shift_offset   = DIM*shiftidx[iidx];
 822
 823         /* Load limits for loop over neighbors */
 824         j_index_start    = jindex[iidx];
 825         j_index_end      = jindex[iidx+1];
 826
 827         /* Get outer coordinate index */
 828         inr              = iinr[iidx];
 829         i_coord_offset   = DIM*inr;
 830
 831         /* Load i particle coords and add shift vector */
 832         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 833                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 834
 835         fix0             = _mm_setzero_pd();
 836         fiy0             = _mm_setzero_pd();
 837         fiz0             = _mm_setzero_pd();
 838         fix1             = _mm_setzero_pd();
 839         fiy1             = _mm_setzero_pd();
 840         fiz1             = _mm_setzero_pd();
 841         fix2             = _mm_setzero_pd();
 842         fiy2             = _mm_setzero_pd();
 843         fiz2             = _mm_setzero_pd();
 844
 845         /* Start inner kernel loop */
 846         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 847         {
 848
 849             /* Get j neighbor index, and coordinate index */
 850             jnrA             = jjnr[jidx];
 851             jnrB             = jjnr[jidx+1];
 852             j_coord_offsetA  = DIM*jnrA;
 853             j_coord_offsetB  = DIM*jnrB;
 854
 855             /* load j atom coordinates */
 856             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 857                                               &jx0,&jy0,&jz0);
 858
 859             /* Calculate displacement vector */
 860             dx00             = _mm_sub_pd(ix0,jx0);
 861             dy00             = _mm_sub_pd(iy0,jy0);
 862             dz00             = _mm_sub_pd(iz0,jz0);
 863             dx10             = _mm_sub_pd(ix1,jx0);
 864             dy10             = _mm_sub_pd(iy1,jy0);
 865             dz10             = _mm_sub_pd(iz1,jz0);
 866             dx20             = _mm_sub_pd(ix2,jx0);
 867             dy20             = _mm_sub_pd(iy2,jy0);
 868             dz20             = _mm_sub_pd(iz2,jz0);
 869
 870             /* Calculate squared distance and things based on it */
 871             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 872             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 873             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 874
 875             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 876             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 877             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 878
 879             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 880             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 881             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 882
 883             /* Load parameters for j particles */
 884             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 885             vdwjidx0A        = 2*vdwtype[jnrA+0];
 886             vdwjidx0B        = 2*vdwtype[jnrB+0];
 887
 888             fjx0             = _mm_setzero_pd();
 889             fjy0             = _mm_setzero_pd();
 890             fjz0             = _mm_setzero_pd();
 891
 892             /**************************
 893              * CALCULATE INTERACTIONS *
 894              **************************/
 895
 896             if (gmx_mm_any_lt(rsq00,rcutoff2))
 897             {
 898
 899             r00              = _mm_mul_pd(rsq00,rinv00);
 900
 901             /* Compute parameters for interactions between i and j atoms */
 902             qq00             = _mm_mul_pd(iq0,jq0);
 903             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 904                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 905             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 906                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 907
 908             /* EWALD ELECTROSTATICS */
 909
 910             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 911             ewrt             = _mm_mul_pd(r00,ewtabscale);
 912             ewitab           = _mm_cvttpd_epi32(ewrt);
 913             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 914             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 915                                          &ewtabF,&ewtabFn);
 916             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 917             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 918
 919             /* Analytical LJ-PME */
 920             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 921             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 922             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 923             exponent         = gmx_simd_exp_d(ewcljrsq);
 924             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 925             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 926             /* f6A = 6 * C6grid * (1 - poly) */
 927             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 928             /* f6B = C6grid * exponent * beta^6 */
 929             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 930             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 931             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);
 932
 933             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 934
 935             fscal            = _mm_add_pd(felec,fvdw);
 936
 937             fscal            = _mm_and_pd(fscal,cutoff_mask);
 938
 939             /* Calculate temporary vectorial force */
 940             tx               = _mm_mul_pd(fscal,dx00);
 941             ty               = _mm_mul_pd(fscal,dy00);
 942             tz               = _mm_mul_pd(fscal,dz00);
 943
 944             /* Update vectorial force */
 945             fix0             = _mm_add_pd(fix0,tx);
 946             fiy0             = _mm_add_pd(fiy0,ty);
 947             fiz0             = _mm_add_pd(fiz0,tz);
 948
 949             fjx0             = _mm_add_pd(fjx0,tx);
 950             fjy0             = _mm_add_pd(fjy0,ty);
 951             fjz0             = _mm_add_pd(fjz0,tz);
 952
 953             }
 954
 955             /**************************
 956              * CALCULATE INTERACTIONS *
 957              **************************/
 958
 959             if (gmx_mm_any_lt(rsq10,rcutoff2))
 960             {
 961
 962             r10              = _mm_mul_pd(rsq10,rinv10);
 963
 964             /* Compute parameters for interactions between i and j atoms */
 965             qq10             = _mm_mul_pd(iq1,jq0);
 966
 967             /* EWALD ELECTROSTATICS */
 968
 969             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 970             ewrt             = _mm_mul_pd(r10,ewtabscale);
 971             ewitab           = _mm_cvttpd_epi32(ewrt);
 972             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 973             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 974                                          &ewtabF,&ewtabFn);
 975             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 976             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 977
 978             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 979
 980             fscal            = felec;
 981
 982             fscal            = _mm_and_pd(fscal,cutoff_mask);
 983
 984             /* Calculate temporary vectorial force */
 985             tx               = _mm_mul_pd(fscal,dx10);
 986             ty               = _mm_mul_pd(fscal,dy10);
 987             tz               = _mm_mul_pd(fscal,dz10);
 988
 989             /* Update vectorial force */
 990             fix1             = _mm_add_pd(fix1,tx);
 991             fiy1             = _mm_add_pd(fiy1,ty);
 992             fiz1             = _mm_add_pd(fiz1,tz);
 993
 994             fjx0             = _mm_add_pd(fjx0,tx);
 995             fjy0             = _mm_add_pd(fjy0,ty);
 996             fjz0             = _mm_add_pd(fjz0,tz);
 997
 998             }
 999
1000             /**************************
1001              * CALCULATE INTERACTIONS *
1002              **************************/
1003
1004             if (gmx_mm_any_lt(rsq20,rcutoff2))
1005             {
1006
1007             r20              = _mm_mul_pd(rsq20,rinv20);
1008
1009             /* Compute parameters for interactions between i and j atoms */
1010             qq20             = _mm_mul_pd(iq2,jq0);
1011
1012             /* EWALD ELECTROSTATICS */
1013
1014             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1015             ewrt             = _mm_mul_pd(r20,ewtabscale);
1016             ewitab           = _mm_cvttpd_epi32(ewrt);
1017             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1018             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1019                                          &ewtabF,&ewtabFn);
1020             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1021             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1022
1023             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1024
1025             fscal            = felec;
1026
1027             fscal            = _mm_and_pd(fscal,cutoff_mask);
1028
1029             /* Calculate temporary vectorial force */
1030             tx               = _mm_mul_pd(fscal,dx20);
1031             ty               = _mm_mul_pd(fscal,dy20);
1032             tz               = _mm_mul_pd(fscal,dz20);
1033
1034             /* Update vectorial force */
1035             fix2             = _mm_add_pd(fix2,tx);
1036             fiy2             = _mm_add_pd(fiy2,ty);
1037             fiz2             = _mm_add_pd(fiz2,tz);
1038
1039             fjx0             = _mm_add_pd(fjx0,tx);
1040             fjy0             = _mm_add_pd(fjy0,ty);
1041             fjz0             = _mm_add_pd(fjz0,tz);
1042
1043             }
1044
1045             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1046
1047             /* Inner loop uses 143 flops */
1048         }
1049
1050         if(jidx<j_index_end)
1051         {
1052
1053             jnrA             = jjnr[jidx];
1054             j_coord_offsetA  = DIM*jnrA;
1055
1056             /* load j atom coordinates */
1057             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1058                                               &jx0,&jy0,&jz0);
1059
1060             /* Calculate displacement vector */
1061             dx00             = _mm_sub_pd(ix0,jx0);
1062             dy00             = _mm_sub_pd(iy0,jy0);
1063             dz00             = _mm_sub_pd(iz0,jz0);
1064             dx10             = _mm_sub_pd(ix1,jx0);
1065             dy10             = _mm_sub_pd(iy1,jy0);
1066             dz10             = _mm_sub_pd(iz1,jz0);
1067             dx20             = _mm_sub_pd(ix2,jx0);
1068             dy20             = _mm_sub_pd(iy2,jy0);
1069             dz20             = _mm_sub_pd(iz2,jz0);
1070
1071             /* Calculate squared distance and things based on it */
1072             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1073             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1074             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1075
1076             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1077             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1078             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1079
1080             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1081             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1082             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1083
1084             /* Load parameters for j particles */
1085             jq0              = _mm_load_sd(charge+jnrA+0);
1086             vdwjidx0A        = 2*vdwtype[jnrA+0];
1087
1088             fjx0             = _mm_setzero_pd();
1089             fjy0             = _mm_setzero_pd();
1090             fjz0             = _mm_setzero_pd();
1091
1092             /**************************
1093              * CALCULATE INTERACTIONS *
1094              **************************/
1095
1096             if (gmx_mm_any_lt(rsq00,rcutoff2))
1097             {
1098
1099             r00              = _mm_mul_pd(rsq00,rinv00);
1100
1101             /* Compute parameters for interactions between i and j atoms */
1102             qq00             = _mm_mul_pd(iq0,jq0);
1103             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1104
1105             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
1106
1107             /* EWALD ELECTROSTATICS */
1108
1109             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1110             ewrt             = _mm_mul_pd(r00,ewtabscale);
1111             ewitab           = _mm_cvttpd_epi32(ewrt);
1112             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1113             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1114             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1115             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1116
1117             /* Analytical LJ-PME */
1118             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1119             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1120             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1121             exponent         = gmx_simd_exp_d(ewcljrsq);
1122             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1123             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1124             /* f6A = 6 * C6grid * (1 - poly) */
1125             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1126             /* f6B = C6grid * exponent * beta^6 */
1127             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1128             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1129             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);
1130
1131             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1132
1133             fscal            = _mm_add_pd(felec,fvdw);
1134
1135             fscal            = _mm_and_pd(fscal,cutoff_mask);
1136
1137             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1138
1139             /* Calculate temporary vectorial force */
1140             tx               = _mm_mul_pd(fscal,dx00);
1141             ty               = _mm_mul_pd(fscal,dy00);
1142             tz               = _mm_mul_pd(fscal,dz00);
1143
1144             /* Update vectorial force */
1145             fix0             = _mm_add_pd(fix0,tx);
1146             fiy0             = _mm_add_pd(fiy0,ty);
1147             fiz0             = _mm_add_pd(fiz0,tz);
1148
1149             fjx0             = _mm_add_pd(fjx0,tx);
1150             fjy0             = _mm_add_pd(fjy0,ty);
1151             fjz0             = _mm_add_pd(fjz0,tz);
1152
1153             }
1154
1155             /**************************
1156              * CALCULATE INTERACTIONS *
1157              **************************/
1158
1159             if (gmx_mm_any_lt(rsq10,rcutoff2))
1160             {
1161
1162             r10              = _mm_mul_pd(rsq10,rinv10);
1163
1164             /* Compute parameters for interactions between i and j atoms */
1165             qq10             = _mm_mul_pd(iq1,jq0);
1166
1167             /* EWALD ELECTROSTATICS */
1168
1169             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1170             ewrt             = _mm_mul_pd(r10,ewtabscale);
1171             ewitab           = _mm_cvttpd_epi32(ewrt);
1172             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1173             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1174             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1175             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1176
1177             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1178
1179             fscal            = felec;
1180
1181             fscal            = _mm_and_pd(fscal,cutoff_mask);
1182
1183             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1184
1185             /* Calculate temporary vectorial force */
1186             tx               = _mm_mul_pd(fscal,dx10);
1187             ty               = _mm_mul_pd(fscal,dy10);
1188             tz               = _mm_mul_pd(fscal,dz10);
1189
1190             /* Update vectorial force */
1191             fix1             = _mm_add_pd(fix1,tx);
1192             fiy1             = _mm_add_pd(fiy1,ty);
1193             fiz1             = _mm_add_pd(fiz1,tz);
1194
1195             fjx0             = _mm_add_pd(fjx0,tx);
1196             fjy0             = _mm_add_pd(fjy0,ty);
1197             fjz0             = _mm_add_pd(fjz0,tz);
1198
1199             }
1200
1201             /**************************
1202              * CALCULATE INTERACTIONS *
1203              **************************/
1204
1205             if (gmx_mm_any_lt(rsq20,rcutoff2))
1206             {
1207
1208             r20              = _mm_mul_pd(rsq20,rinv20);
1209
1210             /* Compute parameters for interactions between i and j atoms */
1211             qq20             = _mm_mul_pd(iq2,jq0);
1212
1213             /* EWALD ELECTROSTATICS */
1214
1215             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1216             ewrt             = _mm_mul_pd(r20,ewtabscale);
1217             ewitab           = _mm_cvttpd_epi32(ewrt);
1218             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1219             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1220             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1221             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1222
1223             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1224
1225             fscal            = felec;
1226
1227             fscal            = _mm_and_pd(fscal,cutoff_mask);
1228
1229             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1230
1231             /* Calculate temporary vectorial force */
1232             tx               = _mm_mul_pd(fscal,dx20);
1233             ty               = _mm_mul_pd(fscal,dy20);
1234             tz               = _mm_mul_pd(fscal,dz20);
1235
1236             /* Update vectorial force */
1237             fix2             = _mm_add_pd(fix2,tx);
1238             fiy2             = _mm_add_pd(fiy2,ty);
1239             fiz2             = _mm_add_pd(fiz2,tz);
1240
1241             fjx0             = _mm_add_pd(fjx0,tx);
1242             fjy0             = _mm_add_pd(fjy0,ty);
1243             fjz0             = _mm_add_pd(fjz0,tz);
1244
1245             }
1246
1247             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1248
1249             /* Inner loop uses 143 flops */
1250         }
1251
1252         /* End of innermost loop */
1253
1254         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1255                                               f+i_coord_offset,fshift+i_shift_offset);
1256
1257         /* Increment number of inner iterations */
1258         inneriter                  += j_index_end - j_index_start;
1259
1260         /* Outer loop uses 18 flops */
1261     }
1262
1263     /* Increment number of outer iterations */
1264     outeriter        += nri;
1265
1266     /* Update outer/inner flops */
1267
1268     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*143);
1269 }