src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecEw_VdwLJEw_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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "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_ElecEw_VdwLJEw_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_ElecEw_VdwLJEw_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     /* Avoid stupid compiler warnings */
 149     jnrA = jnrB = 0;
 150     j_coord_offsetA = 0;
 151     j_coord_offsetB = 0;
 152
 153     outeriter        = 0;
 154     inneriter        = 0;
 155
 156     /* Start outer loop over neighborlists */
 157     for(iidx=0; iidx<nri; iidx++)
 158     {
 159         /* Load shift vector for this list */
 160         i_shift_offset   = DIM*shiftidx[iidx];
 161
 162         /* Load limits for loop over neighbors */
 163         j_index_start    = jindex[iidx];
 164         j_index_end      = jindex[iidx+1];
 165
 166         /* Get outer coordinate index */
 167         inr              = iinr[iidx];
 168         i_coord_offset   = DIM*inr;
 169
 170         /* Load i particle coords and add shift vector */
 171         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 172                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 173
 174         fix0             = _mm_setzero_pd();
 175         fiy0             = _mm_setzero_pd();
 176         fiz0             = _mm_setzero_pd();
 177         fix1             = _mm_setzero_pd();
 178         fiy1             = _mm_setzero_pd();
 179         fiz1             = _mm_setzero_pd();
 180         fix2             = _mm_setzero_pd();
 181         fiy2             = _mm_setzero_pd();
 182         fiz2             = _mm_setzero_pd();
 183
 184         /* Reset potential sums */
 185         velecsum         = _mm_setzero_pd();
 186         vvdwsum          = _mm_setzero_pd();
 187
 188         /* Start inner kernel loop */
 189         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 190         {
 191
 192             /* Get j neighbor index, and coordinate index */
 193             jnrA             = jjnr[jidx];
 194             jnrB             = jjnr[jidx+1];
 195             j_coord_offsetA  = DIM*jnrA;
 196             j_coord_offsetB  = DIM*jnrB;
 197
 198             /* load j atom coordinates */
 199             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 200                                               &jx0,&jy0,&jz0);
 201
 202             /* Calculate displacement vector */
 203             dx00             = _mm_sub_pd(ix0,jx0);
 204             dy00             = _mm_sub_pd(iy0,jy0);
 205             dz00             = _mm_sub_pd(iz0,jz0);
 206             dx10             = _mm_sub_pd(ix1,jx0);
 207             dy10             = _mm_sub_pd(iy1,jy0);
 208             dz10             = _mm_sub_pd(iz1,jz0);
 209             dx20             = _mm_sub_pd(ix2,jx0);
 210             dy20             = _mm_sub_pd(iy2,jy0);
 211             dz20             = _mm_sub_pd(iz2,jz0);
 212
 213             /* Calculate squared distance and things based on it */
 214             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 215             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 216             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 217
 218             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 219             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 220             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 221
 222             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 223             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 224             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 225
 226             /* Load parameters for j particles */
 227             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 228             vdwjidx0A        = 2*vdwtype[jnrA+0];
 229             vdwjidx0B        = 2*vdwtype[jnrB+0];
 230
 231             fjx0             = _mm_setzero_pd();
 232             fjy0             = _mm_setzero_pd();
 233             fjz0             = _mm_setzero_pd();
 234
 235             /**************************
 236              * CALCULATE INTERACTIONS *
 237              **************************/
 238
 239             r00              = _mm_mul_pd(rsq00,rinv00);
 240
 241             /* Compute parameters for interactions between i and j atoms */
 242             qq00             = _mm_mul_pd(iq0,jq0);
 243             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 244                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 245             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 246                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 247
 248             /* EWALD ELECTROSTATICS */
 249
 250             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 251             ewrt             = _mm_mul_pd(r00,ewtabscale);
 252             ewitab           = _mm_cvttpd_epi32(ewrt);
 253             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 254             ewitab           = _mm_slli_epi32(ewitab,2);
 255             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 256             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 257             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 258             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 259             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 260             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 261             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 262             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 263             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 264             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 265
 266             /* Analytical LJ-PME */
 267             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 268             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 269             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 270             exponent         = gmx_simd_exp_d(ewcljrsq);
 271             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 272             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 273             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 274             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 275             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 276             vvdw             = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
 277             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 278             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);
 279
 280             /* Update potential sum for this i atom from the interaction with this j atom. */
 281             velecsum         = _mm_add_pd(velecsum,velec);
 282             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 283
 284             fscal            = _mm_add_pd(felec,fvdw);
 285
 286             /* Calculate temporary vectorial force */
 287             tx               = _mm_mul_pd(fscal,dx00);
 288             ty               = _mm_mul_pd(fscal,dy00);
 289             tz               = _mm_mul_pd(fscal,dz00);
 290
 291             /* Update vectorial force */
 292             fix0             = _mm_add_pd(fix0,tx);
 293             fiy0             = _mm_add_pd(fiy0,ty);
 294             fiz0             = _mm_add_pd(fiz0,tz);
 295
 296             fjx0             = _mm_add_pd(fjx0,tx);
 297             fjy0             = _mm_add_pd(fjy0,ty);
 298             fjz0             = _mm_add_pd(fjz0,tz);
 299
 300             /**************************
 301              * CALCULATE INTERACTIONS *
 302              **************************/
 303
 304             r10              = _mm_mul_pd(rsq10,rinv10);
 305
 306             /* Compute parameters for interactions between i and j atoms */
 307             qq10             = _mm_mul_pd(iq1,jq0);
 308
 309             /* EWALD ELECTROSTATICS */
 310
 311             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 312             ewrt             = _mm_mul_pd(r10,ewtabscale);
 313             ewitab           = _mm_cvttpd_epi32(ewrt);
 314             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 315             ewitab           = _mm_slli_epi32(ewitab,2);
 316             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 317             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 318             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 319             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 320             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 321             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 322             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 323             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 324             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 325             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 326
 327             /* Update potential sum for this i atom from the interaction with this j atom. */
 328             velecsum         = _mm_add_pd(velecsum,velec);
 329
 330             fscal            = felec;
 331
 332             /* Calculate temporary vectorial force */
 333             tx               = _mm_mul_pd(fscal,dx10);
 334             ty               = _mm_mul_pd(fscal,dy10);
 335             tz               = _mm_mul_pd(fscal,dz10);
 336
 337             /* Update vectorial force */
 338             fix1             = _mm_add_pd(fix1,tx);
 339             fiy1             = _mm_add_pd(fiy1,ty);
 340             fiz1             = _mm_add_pd(fiz1,tz);
 341
 342             fjx0             = _mm_add_pd(fjx0,tx);
 343             fjy0             = _mm_add_pd(fjy0,ty);
 344             fjz0             = _mm_add_pd(fjz0,tz);
 345
 346             /**************************
 347              * CALCULATE INTERACTIONS *
 348              **************************/
 349
 350             r20              = _mm_mul_pd(rsq20,rinv20);
 351
 352             /* Compute parameters for interactions between i and j atoms */
 353             qq20             = _mm_mul_pd(iq2,jq0);
 354
 355             /* EWALD ELECTROSTATICS */
 356
 357             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 358             ewrt             = _mm_mul_pd(r20,ewtabscale);
 359             ewitab           = _mm_cvttpd_epi32(ewrt);
 360             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 361             ewitab           = _mm_slli_epi32(ewitab,2);
 362             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 363             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 364             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 365             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 366             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 367             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 368             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 369             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 370             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 371             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 372
 373             /* Update potential sum for this i atom from the interaction with this j atom. */
 374             velecsum         = _mm_add_pd(velecsum,velec);
 375
 376             fscal            = felec;
 377
 378             /* Calculate temporary vectorial force */
 379             tx               = _mm_mul_pd(fscal,dx20);
 380             ty               = _mm_mul_pd(fscal,dy20);
 381             tz               = _mm_mul_pd(fscal,dz20);
 382
 383             /* Update vectorial force */
 384             fix2             = _mm_add_pd(fix2,tx);
 385             fiy2             = _mm_add_pd(fiy2,ty);
 386             fiz2             = _mm_add_pd(fiz2,tz);
 387
 388             fjx0             = _mm_add_pd(fjx0,tx);
 389             fjy0             = _mm_add_pd(fjy0,ty);
 390             fjz0             = _mm_add_pd(fjz0,tz);
 391
 392             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 393
 394             /* Inner loop uses 154 flops */
 395         }
 396
 397         if(jidx<j_index_end)
 398         {
 399
 400             jnrA             = jjnr[jidx];
 401             j_coord_offsetA  = DIM*jnrA;
 402
 403             /* load j atom coordinates */
 404             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 405                                               &jx0,&jy0,&jz0);
 406
 407             /* Calculate displacement vector */
 408             dx00             = _mm_sub_pd(ix0,jx0);
 409             dy00             = _mm_sub_pd(iy0,jy0);
 410             dz00             = _mm_sub_pd(iz0,jz0);
 411             dx10             = _mm_sub_pd(ix1,jx0);
 412             dy10             = _mm_sub_pd(iy1,jy0);
 413             dz10             = _mm_sub_pd(iz1,jz0);
 414             dx20             = _mm_sub_pd(ix2,jx0);
 415             dy20             = _mm_sub_pd(iy2,jy0);
 416             dz20             = _mm_sub_pd(iz2,jz0);
 417
 418             /* Calculate squared distance and things based on it */
 419             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 420             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 421             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 422
 423             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 424             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 425             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 426
 427             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 428             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 429             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 430
 431             /* Load parameters for j particles */
 432             jq0              = _mm_load_sd(charge+jnrA+0);
 433             vdwjidx0A        = 2*vdwtype[jnrA+0];
 434
 435             fjx0             = _mm_setzero_pd();
 436             fjy0             = _mm_setzero_pd();
 437             fjz0             = _mm_setzero_pd();
 438
 439             /**************************
 440              * CALCULATE INTERACTIONS *
 441              **************************/
 442
 443             r00              = _mm_mul_pd(rsq00,rinv00);
 444
 445             /* Compute parameters for interactions between i and j atoms */
 446             qq00             = _mm_mul_pd(iq0,jq0);
 447             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 448
 449             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 450
 451             /* EWALD ELECTROSTATICS */
 452
 453             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 454             ewrt             = _mm_mul_pd(r00,ewtabscale);
 455             ewitab           = _mm_cvttpd_epi32(ewrt);
 456             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 457             ewitab           = _mm_slli_epi32(ewitab,2);
 458             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 459             ewtabD           = _mm_setzero_pd();
 460             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 461             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 462             ewtabFn          = _mm_setzero_pd();
 463             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 464             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 465             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 466             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 467             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 468
 469             /* Analytical LJ-PME */
 470             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 471             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 472             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 473             exponent         = gmx_simd_exp_d(ewcljrsq);
 474             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 475             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 476             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 477             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 478             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 479             vvdw             = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
 480             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 481             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);
 482
 483             /* Update potential sum for this i atom from the interaction with this j atom. */
 484             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 485             velecsum         = _mm_add_pd(velecsum,velec);
 486             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 487             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 488
 489             fscal            = _mm_add_pd(felec,fvdw);
 490
 491             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 492
 493             /* Calculate temporary vectorial force */
 494             tx               = _mm_mul_pd(fscal,dx00);
 495             ty               = _mm_mul_pd(fscal,dy00);
 496             tz               = _mm_mul_pd(fscal,dz00);
 497
 498             /* Update vectorial force */
 499             fix0             = _mm_add_pd(fix0,tx);
 500             fiy0             = _mm_add_pd(fiy0,ty);
 501             fiz0             = _mm_add_pd(fiz0,tz);
 502
 503             fjx0             = _mm_add_pd(fjx0,tx);
 504             fjy0             = _mm_add_pd(fjy0,ty);
 505             fjz0             = _mm_add_pd(fjz0,tz);
 506
 507             /**************************
 508              * CALCULATE INTERACTIONS *
 509              **************************/
 510
 511             r10              = _mm_mul_pd(rsq10,rinv10);
 512
 513             /* Compute parameters for interactions between i and j atoms */
 514             qq10             = _mm_mul_pd(iq1,jq0);
 515
 516             /* EWALD ELECTROSTATICS */
 517
 518             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 519             ewrt             = _mm_mul_pd(r10,ewtabscale);
 520             ewitab           = _mm_cvttpd_epi32(ewrt);
 521             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 522             ewitab           = _mm_slli_epi32(ewitab,2);
 523             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 524             ewtabD           = _mm_setzero_pd();
 525             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 526             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 527             ewtabFn          = _mm_setzero_pd();
 528             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 529             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 530             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 531             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 532             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 533
 534             /* Update potential sum for this i atom from the interaction with this j atom. */
 535             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 536             velecsum         = _mm_add_pd(velecsum,velec);
 537
 538             fscal            = felec;
 539
 540             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 541
 542             /* Calculate temporary vectorial force */
 543             tx               = _mm_mul_pd(fscal,dx10);
 544             ty               = _mm_mul_pd(fscal,dy10);
 545             tz               = _mm_mul_pd(fscal,dz10);
 546
 547             /* Update vectorial force */
 548             fix1             = _mm_add_pd(fix1,tx);
 549             fiy1             = _mm_add_pd(fiy1,ty);
 550             fiz1             = _mm_add_pd(fiz1,tz);
 551
 552             fjx0             = _mm_add_pd(fjx0,tx);
 553             fjy0             = _mm_add_pd(fjy0,ty);
 554             fjz0             = _mm_add_pd(fjz0,tz);
 555
 556             /**************************
 557              * CALCULATE INTERACTIONS *
 558              **************************/
 559
 560             r20              = _mm_mul_pd(rsq20,rinv20);
 561
 562             /* Compute parameters for interactions between i and j atoms */
 563             qq20             = _mm_mul_pd(iq2,jq0);
 564
 565             /* EWALD ELECTROSTATICS */
 566
 567             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 568             ewrt             = _mm_mul_pd(r20,ewtabscale);
 569             ewitab           = _mm_cvttpd_epi32(ewrt);
 570             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 571             ewitab           = _mm_slli_epi32(ewitab,2);
 572             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 573             ewtabD           = _mm_setzero_pd();
 574             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 575             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 576             ewtabFn          = _mm_setzero_pd();
 577             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 578             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 579             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 580             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 581             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 582
 583             /* Update potential sum for this i atom from the interaction with this j atom. */
 584             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 585             velecsum         = _mm_add_pd(velecsum,velec);
 586
 587             fscal            = felec;
 588
 589             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 590
 591             /* Calculate temporary vectorial force */
 592             tx               = _mm_mul_pd(fscal,dx20);
 593             ty               = _mm_mul_pd(fscal,dy20);
 594             tz               = _mm_mul_pd(fscal,dz20);
 595
 596             /* Update vectorial force */
 597             fix2             = _mm_add_pd(fix2,tx);
 598             fiy2             = _mm_add_pd(fiy2,ty);
 599             fiz2             = _mm_add_pd(fiz2,tz);
 600
 601             fjx0             = _mm_add_pd(fjx0,tx);
 602             fjy0             = _mm_add_pd(fjy0,ty);
 603             fjz0             = _mm_add_pd(fjz0,tz);
 604
 605             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 606
 607             /* Inner loop uses 154 flops */
 608         }
 609
 610         /* End of innermost loop */
 611
 612         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 613                                               f+i_coord_offset,fshift+i_shift_offset);
 614
 615         ggid                        = gid[iidx];
 616         /* Update potential energies */
 617         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 618         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 619
 620         /* Increment number of inner iterations */
 621         inneriter                  += j_index_end - j_index_start;
 622
 623         /* Outer loop uses 20 flops */
 624     }
 625
 626     /* Increment number of outer iterations */
 627     outeriter        += nri;
 628
 629     /* Update outer/inner flops */
 630
 631     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*154);
 632 }
 633 /*
 634  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_sse4_1_double
 635  * Electrostatics interaction: Ewald
 636  * VdW interaction:            LJEwald
 637  * Geometry:                   Water3-Particle
 638  * Calculate force/pot:        Force
 639  */
 640 void
 641 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_sse4_1_double
 642                     (t_nblist                    * gmx_restrict       nlist,
 643                      rvec                        * gmx_restrict          xx,
 644                      rvec                        * gmx_restrict          ff,
 645                      t_forcerec                  * gmx_restrict          fr,
 646                      t_mdatoms                   * gmx_restrict     mdatoms,
 647                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 648                      t_nrnb                      * gmx_restrict        nrnb)
 649 {
 650     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 651      * just 0 for non-waters.
 652      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 653      * jnr indices corresponding to data put in the four positions in the SIMD register.
 654      */
 655     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 656     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 657     int              jnrA,jnrB;
 658     int              j_coord_offsetA,j_coord_offsetB;
 659     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 660     real             rcutoff_scalar;
 661     real             *shiftvec,*fshift,*x,*f;
 662     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 663     int              vdwioffset0;
 664     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 665     int              vdwioffset1;
 666     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 667     int              vdwioffset2;
 668     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 669     int              vdwjidx0A,vdwjidx0B;
 670     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 671     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 672     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 673     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 674     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 675     real             *charge;
 676     int              nvdwtype;
 677     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 678     int              *vdwtype;
 679     real             *vdwparam;
 680     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 681     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 682     __m128d           c6grid_00;
 683     __m128d           c6grid_10;
 684     __m128d           c6grid_20;
 685     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 686     real             *vdwgridparam;
 687     __m128d           one_half = _mm_set1_pd(0.5);
 688     __m128d           minus_one = _mm_set1_pd(-1.0);
 689     __m128i          ewitab;
 690     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 691     real             *ewtab;
 692     __m128d          dummy_mask,cutoff_mask;
 693     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 694     __m128d          one     = _mm_set1_pd(1.0);
 695     __m128d          two     = _mm_set1_pd(2.0);
 696     x                = xx[0];
 697     f                = ff[0];
 698
 699     nri              = nlist->nri;
 700     iinr             = nlist->iinr;
 701     jindex           = nlist->jindex;
 702     jjnr             = nlist->jjnr;
 703     shiftidx         = nlist->shift;
 704     gid              = nlist->gid;
 705     shiftvec         = fr->shift_vec[0];
 706     fshift           = fr->fshift[0];
 707     facel            = _mm_set1_pd(fr->epsfac);
 708     charge           = mdatoms->chargeA;
 709     nvdwtype         = fr->ntype;
 710     vdwparam         = fr->nbfp;
 711     vdwtype          = mdatoms->typeA;
 712     vdwgridparam     = fr->ljpme_c6grid;
 713     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 714     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 715     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 716
 717     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 718     ewtab            = fr->ic->tabq_coul_F;
 719     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 720     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 721
 722     /* Setup water-specific parameters */
 723     inr              = nlist->iinr[0];
 724     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 725     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 726     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 727     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 728
 729     /* Avoid stupid compiler warnings */
 730     jnrA = jnrB = 0;
 731     j_coord_offsetA = 0;
 732     j_coord_offsetB = 0;
 733
 734     outeriter        = 0;
 735     inneriter        = 0;
 736
 737     /* Start outer loop over neighborlists */
 738     for(iidx=0; iidx<nri; iidx++)
 739     {
 740         /* Load shift vector for this list */
 741         i_shift_offset   = DIM*shiftidx[iidx];
 742
 743         /* Load limits for loop over neighbors */
 744         j_index_start    = jindex[iidx];
 745         j_index_end      = jindex[iidx+1];
 746
 747         /* Get outer coordinate index */
 748         inr              = iinr[iidx];
 749         i_coord_offset   = DIM*inr;
 750
 751         /* Load i particle coords and add shift vector */
 752         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 753                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 754
 755         fix0             = _mm_setzero_pd();
 756         fiy0             = _mm_setzero_pd();
 757         fiz0             = _mm_setzero_pd();
 758         fix1             = _mm_setzero_pd();
 759         fiy1             = _mm_setzero_pd();
 760         fiz1             = _mm_setzero_pd();
 761         fix2             = _mm_setzero_pd();
 762         fiy2             = _mm_setzero_pd();
 763         fiz2             = _mm_setzero_pd();
 764
 765         /* Start inner kernel loop */
 766         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 767         {
 768
 769             /* Get j neighbor index, and coordinate index */
 770             jnrA             = jjnr[jidx];
 771             jnrB             = jjnr[jidx+1];
 772             j_coord_offsetA  = DIM*jnrA;
 773             j_coord_offsetB  = DIM*jnrB;
 774
 775             /* load j atom coordinates */
 776             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 777                                               &jx0,&jy0,&jz0);
 778
 779             /* Calculate displacement vector */
 780             dx00             = _mm_sub_pd(ix0,jx0);
 781             dy00             = _mm_sub_pd(iy0,jy0);
 782             dz00             = _mm_sub_pd(iz0,jz0);
 783             dx10             = _mm_sub_pd(ix1,jx0);
 784             dy10             = _mm_sub_pd(iy1,jy0);
 785             dz10             = _mm_sub_pd(iz1,jz0);
 786             dx20             = _mm_sub_pd(ix2,jx0);
 787             dy20             = _mm_sub_pd(iy2,jy0);
 788             dz20             = _mm_sub_pd(iz2,jz0);
 789
 790             /* Calculate squared distance and things based on it */
 791             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 792             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 793             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 794
 795             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 796             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 797             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 798
 799             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 800             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 801             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 802
 803             /* Load parameters for j particles */
 804             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 805             vdwjidx0A        = 2*vdwtype[jnrA+0];
 806             vdwjidx0B        = 2*vdwtype[jnrB+0];
 807
 808             fjx0             = _mm_setzero_pd();
 809             fjy0             = _mm_setzero_pd();
 810             fjz0             = _mm_setzero_pd();
 811
 812             /**************************
 813              * CALCULATE INTERACTIONS *
 814              **************************/
 815
 816             r00              = _mm_mul_pd(rsq00,rinv00);
 817
 818             /* Compute parameters for interactions between i and j atoms */
 819             qq00             = _mm_mul_pd(iq0,jq0);
 820             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 821                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 822             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 823                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 824
 825             /* EWALD ELECTROSTATICS */
 826
 827             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 828             ewrt             = _mm_mul_pd(r00,ewtabscale);
 829             ewitab           = _mm_cvttpd_epi32(ewrt);
 830             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 831             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 832                                          &ewtabF,&ewtabFn);
 833             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 834             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 835
 836             /* Analytical LJ-PME */
 837             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 838             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 839             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 840             exponent         = gmx_simd_exp_d(ewcljrsq);
 841             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 842             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 843             /* f6A = 6 * C6grid * (1 - poly) */
 844             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 845             /* f6B = C6grid * exponent * beta^6 */
 846             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 847             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 848             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);
 849
 850             fscal            = _mm_add_pd(felec,fvdw);
 851
 852             /* Calculate temporary vectorial force */
 853             tx               = _mm_mul_pd(fscal,dx00);
 854             ty               = _mm_mul_pd(fscal,dy00);
 855             tz               = _mm_mul_pd(fscal,dz00);
 856
 857             /* Update vectorial force */
 858             fix0             = _mm_add_pd(fix0,tx);
 859             fiy0             = _mm_add_pd(fiy0,ty);
 860             fiz0             = _mm_add_pd(fiz0,tz);
 861
 862             fjx0             = _mm_add_pd(fjx0,tx);
 863             fjy0             = _mm_add_pd(fjy0,ty);
 864             fjz0             = _mm_add_pd(fjz0,tz);
 865
 866             /**************************
 867              * CALCULATE INTERACTIONS *
 868              **************************/
 869
 870             r10              = _mm_mul_pd(rsq10,rinv10);
 871
 872             /* Compute parameters for interactions between i and j atoms */
 873             qq10             = _mm_mul_pd(iq1,jq0);
 874
 875             /* EWALD ELECTROSTATICS */
 876
 877             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 878             ewrt             = _mm_mul_pd(r10,ewtabscale);
 879             ewitab           = _mm_cvttpd_epi32(ewrt);
 880             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 881             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 882                                          &ewtabF,&ewtabFn);
 883             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 884             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 885
 886             fscal            = felec;
 887
 888             /* Calculate temporary vectorial force */
 889             tx               = _mm_mul_pd(fscal,dx10);
 890             ty               = _mm_mul_pd(fscal,dy10);
 891             tz               = _mm_mul_pd(fscal,dz10);
 892
 893             /* Update vectorial force */
 894             fix1             = _mm_add_pd(fix1,tx);
 895             fiy1             = _mm_add_pd(fiy1,ty);
 896             fiz1             = _mm_add_pd(fiz1,tz);
 897
 898             fjx0             = _mm_add_pd(fjx0,tx);
 899             fjy0             = _mm_add_pd(fjy0,ty);
 900             fjz0             = _mm_add_pd(fjz0,tz);
 901
 902             /**************************
 903              * CALCULATE INTERACTIONS *
 904              **************************/
 905
 906             r20              = _mm_mul_pd(rsq20,rinv20);
 907
 908             /* Compute parameters for interactions between i and j atoms */
 909             qq20             = _mm_mul_pd(iq2,jq0);
 910
 911             /* EWALD ELECTROSTATICS */
 912
 913             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 914             ewrt             = _mm_mul_pd(r20,ewtabscale);
 915             ewitab           = _mm_cvttpd_epi32(ewrt);
 916             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 917             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 918                                          &ewtabF,&ewtabFn);
 919             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 920             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 921
 922             fscal            = felec;
 923
 924             /* Calculate temporary vectorial force */
 925             tx               = _mm_mul_pd(fscal,dx20);
 926             ty               = _mm_mul_pd(fscal,dy20);
 927             tz               = _mm_mul_pd(fscal,dz20);
 928
 929             /* Update vectorial force */
 930             fix2             = _mm_add_pd(fix2,tx);
 931             fiy2             = _mm_add_pd(fiy2,ty);
 932             fiz2             = _mm_add_pd(fiz2,tz);
 933
 934             fjx0             = _mm_add_pd(fjx0,tx);
 935             fjy0             = _mm_add_pd(fjy0,ty);
 936             fjz0             = _mm_add_pd(fjz0,tz);
 937
 938             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 939
 940             /* Inner loop uses 134 flops */
 941         }
 942
 943         if(jidx<j_index_end)
 944         {
 945
 946             jnrA             = jjnr[jidx];
 947             j_coord_offsetA  = DIM*jnrA;
 948
 949             /* load j atom coordinates */
 950             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 951                                               &jx0,&jy0,&jz0);
 952
 953             /* Calculate displacement vector */
 954             dx00             = _mm_sub_pd(ix0,jx0);
 955             dy00             = _mm_sub_pd(iy0,jy0);
 956             dz00             = _mm_sub_pd(iz0,jz0);
 957             dx10             = _mm_sub_pd(ix1,jx0);
 958             dy10             = _mm_sub_pd(iy1,jy0);
 959             dz10             = _mm_sub_pd(iz1,jz0);
 960             dx20             = _mm_sub_pd(ix2,jx0);
 961             dy20             = _mm_sub_pd(iy2,jy0);
 962             dz20             = _mm_sub_pd(iz2,jz0);
 963
 964             /* Calculate squared distance and things based on it */
 965             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 966             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 967             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 968
 969             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 970             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 971             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 972
 973             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 974             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 975             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 976
 977             /* Load parameters for j particles */
 978             jq0              = _mm_load_sd(charge+jnrA+0);
 979             vdwjidx0A        = 2*vdwtype[jnrA+0];
 980
 981             fjx0             = _mm_setzero_pd();
 982             fjy0             = _mm_setzero_pd();
 983             fjz0             = _mm_setzero_pd();
 984
 985             /**************************
 986              * CALCULATE INTERACTIONS *
 987              **************************/
 988
 989             r00              = _mm_mul_pd(rsq00,rinv00);
 990
 991             /* Compute parameters for interactions between i and j atoms */
 992             qq00             = _mm_mul_pd(iq0,jq0);
 993             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 994
 995             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 996
 997             /* EWALD ELECTROSTATICS */
 998
 999             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1000             ewrt             = _mm_mul_pd(r00,ewtabscale);
1001             ewitab           = _mm_cvttpd_epi32(ewrt);
1002             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1003             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1004             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1005             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1006
1007             /* Analytical LJ-PME */
1008             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1009             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1010             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1011             exponent         = gmx_simd_exp_d(ewcljrsq);
1012             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1013             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1014             /* f6A = 6 * C6grid * (1 - poly) */
1015             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1016             /* f6B = C6grid * exponent * beta^6 */
1017             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1018             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1019             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);
1020
1021             fscal            = _mm_add_pd(felec,fvdw);
1022
1023             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1024
1025             /* Calculate temporary vectorial force */
1026             tx               = _mm_mul_pd(fscal,dx00);
1027             ty               = _mm_mul_pd(fscal,dy00);
1028             tz               = _mm_mul_pd(fscal,dz00);
1029
1030             /* Update vectorial force */
1031             fix0             = _mm_add_pd(fix0,tx);
1032             fiy0             = _mm_add_pd(fiy0,ty);
1033             fiz0             = _mm_add_pd(fiz0,tz);
1034
1035             fjx0             = _mm_add_pd(fjx0,tx);
1036             fjy0             = _mm_add_pd(fjy0,ty);
1037             fjz0             = _mm_add_pd(fjz0,tz);
1038
1039             /**************************
1040              * CALCULATE INTERACTIONS *
1041              **************************/
1042
1043             r10              = _mm_mul_pd(rsq10,rinv10);
1044
1045             /* Compute parameters for interactions between i and j atoms */
1046             qq10             = _mm_mul_pd(iq1,jq0);
1047
1048             /* EWALD ELECTROSTATICS */
1049
1050             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1051             ewrt             = _mm_mul_pd(r10,ewtabscale);
1052             ewitab           = _mm_cvttpd_epi32(ewrt);
1053             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1054             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1055             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1056             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1057
1058             fscal            = felec;
1059
1060             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1061
1062             /* Calculate temporary vectorial force */
1063             tx               = _mm_mul_pd(fscal,dx10);
1064             ty               = _mm_mul_pd(fscal,dy10);
1065             tz               = _mm_mul_pd(fscal,dz10);
1066
1067             /* Update vectorial force */
1068             fix1             = _mm_add_pd(fix1,tx);
1069             fiy1             = _mm_add_pd(fiy1,ty);
1070             fiz1             = _mm_add_pd(fiz1,tz);
1071
1072             fjx0             = _mm_add_pd(fjx0,tx);
1073             fjy0             = _mm_add_pd(fjy0,ty);
1074             fjz0             = _mm_add_pd(fjz0,tz);
1075
1076             /**************************
1077              * CALCULATE INTERACTIONS *
1078              **************************/
1079
1080             r20              = _mm_mul_pd(rsq20,rinv20);
1081
1082             /* Compute parameters for interactions between i and j atoms */
1083             qq20             = _mm_mul_pd(iq2,jq0);
1084
1085             /* EWALD ELECTROSTATICS */
1086
1087             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1088             ewrt             = _mm_mul_pd(r20,ewtabscale);
1089             ewitab           = _mm_cvttpd_epi32(ewrt);
1090             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1091             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1092             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1093             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1094
1095             fscal            = felec;
1096
1097             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1098
1099             /* Calculate temporary vectorial force */
1100             tx               = _mm_mul_pd(fscal,dx20);
1101             ty               = _mm_mul_pd(fscal,dy20);
1102             tz               = _mm_mul_pd(fscal,dz20);
1103
1104             /* Update vectorial force */
1105             fix2             = _mm_add_pd(fix2,tx);
1106             fiy2             = _mm_add_pd(fiy2,ty);
1107             fiz2             = _mm_add_pd(fiz2,tz);
1108
1109             fjx0             = _mm_add_pd(fjx0,tx);
1110             fjy0             = _mm_add_pd(fjy0,ty);
1111             fjz0             = _mm_add_pd(fjz0,tz);
1112
1113             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1114
1115             /* Inner loop uses 134 flops */
1116         }
1117
1118         /* End of innermost loop */
1119
1120         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1121                                               f+i_coord_offset,fshift+i_shift_offset);
1122
1123         /* Increment number of inner iterations */
1124         inneriter                  += j_index_end - j_index_start;
1125
1126         /* Outer loop uses 18 flops */
1127     }
1128
1129     /* Increment number of outer iterations */
1130     outeriter        += nri;
1131
1132     /* Update outer/inner flops */
1133
1134     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*134);
1135 }