src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecEw_VdwLJEw_GeomW4P1_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_ElecEw_VdwLJEw_GeomW4P1_VF_sse4_1_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_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              vdwioffset3;
  89     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B;
  91     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 103     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 104     __m128d           c6grid_00;
 105     __m128d           c6grid_10;
 106     __m128d           c6grid_20;
 107     __m128d           c6grid_30;
 108     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 109     real             *vdwgridparam;
 110     __m128d           one_half = _mm_set1_pd(0.5);
 111     __m128d           minus_one = _mm_set1_pd(-1.0);
 112     __m128i          ewitab;
 113     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 114     real             *ewtab;
 115     __m128d          dummy_mask,cutoff_mask;
 116     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 117     __m128d          one     = _mm_set1_pd(1.0);
 118     __m128d          two     = _mm_set1_pd(2.0);
 119     x                = xx[0];
 120     f                = ff[0];
 121
 122     nri              = nlist->nri;
 123     iinr             = nlist->iinr;
 124     jindex           = nlist->jindex;
 125     jjnr             = nlist->jjnr;
 126     shiftidx         = nlist->shift;
 127     gid              = nlist->gid;
 128     shiftvec         = fr->shift_vec[0];
 129     fshift           = fr->fshift[0];
 130     facel            = _mm_set1_pd(fr->epsfac);
 131     charge           = mdatoms->chargeA;
 132     nvdwtype         = fr->ntype;
 133     vdwparam         = fr->nbfp;
 134     vdwtype          = mdatoms->typeA;
 135     vdwgridparam     = fr->ljpme_c6grid;
 136     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 137     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 138     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 139
 140     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 141     ewtab            = fr->ic->tabq_coul_FDV0;
 142     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 143     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 144
 145     /* Setup water-specific parameters */
 146     inr              = nlist->iinr[0];
 147     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 148     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 149     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 150     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 151
 152     /* Avoid stupid compiler warnings */
 153     jnrA = jnrB = 0;
 154     j_coord_offsetA = 0;
 155     j_coord_offsetB = 0;
 156
 157     outeriter        = 0;
 158     inneriter        = 0;
 159
 160     /* Start outer loop over neighborlists */
 161     for(iidx=0; iidx<nri; iidx++)
 162     {
 163         /* Load shift vector for this list */
 164         i_shift_offset   = DIM*shiftidx[iidx];
 165
 166         /* Load limits for loop over neighbors */
 167         j_index_start    = jindex[iidx];
 168         j_index_end      = jindex[iidx+1];
 169
 170         /* Get outer coordinate index */
 171         inr              = iinr[iidx];
 172         i_coord_offset   = DIM*inr;
 173
 174         /* Load i particle coords and add shift vector */
 175         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 176                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 177
 178         fix0             = _mm_setzero_pd();
 179         fiy0             = _mm_setzero_pd();
 180         fiz0             = _mm_setzero_pd();
 181         fix1             = _mm_setzero_pd();
 182         fiy1             = _mm_setzero_pd();
 183         fiz1             = _mm_setzero_pd();
 184         fix2             = _mm_setzero_pd();
 185         fiy2             = _mm_setzero_pd();
 186         fiz2             = _mm_setzero_pd();
 187         fix3             = _mm_setzero_pd();
 188         fiy3             = _mm_setzero_pd();
 189         fiz3             = _mm_setzero_pd();
 190
 191         /* Reset potential sums */
 192         velecsum         = _mm_setzero_pd();
 193         vvdwsum          = _mm_setzero_pd();
 194
 195         /* Start inner kernel loop */
 196         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 197         {
 198
 199             /* Get j neighbor index, and coordinate index */
 200             jnrA             = jjnr[jidx];
 201             jnrB             = jjnr[jidx+1];
 202             j_coord_offsetA  = DIM*jnrA;
 203             j_coord_offsetB  = DIM*jnrB;
 204
 205             /* load j atom coordinates */
 206             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 207                                               &jx0,&jy0,&jz0);
 208
 209             /* Calculate displacement vector */
 210             dx00             = _mm_sub_pd(ix0,jx0);
 211             dy00             = _mm_sub_pd(iy0,jy0);
 212             dz00             = _mm_sub_pd(iz0,jz0);
 213             dx10             = _mm_sub_pd(ix1,jx0);
 214             dy10             = _mm_sub_pd(iy1,jy0);
 215             dz10             = _mm_sub_pd(iz1,jz0);
 216             dx20             = _mm_sub_pd(ix2,jx0);
 217             dy20             = _mm_sub_pd(iy2,jy0);
 218             dz20             = _mm_sub_pd(iz2,jz0);
 219             dx30             = _mm_sub_pd(ix3,jx0);
 220             dy30             = _mm_sub_pd(iy3,jy0);
 221             dz30             = _mm_sub_pd(iz3,jz0);
 222
 223             /* Calculate squared distance and things based on it */
 224             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 225             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 226             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 227             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 228
 229             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 230             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 231             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 232             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 233
 234             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 235             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 236             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 237             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 238
 239             /* Load parameters for j particles */
 240             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 241             vdwjidx0A        = 2*vdwtype[jnrA+0];
 242             vdwjidx0B        = 2*vdwtype[jnrB+0];
 243
 244             fjx0             = _mm_setzero_pd();
 245             fjy0             = _mm_setzero_pd();
 246             fjz0             = _mm_setzero_pd();
 247
 248             /**************************
 249              * CALCULATE INTERACTIONS *
 250              **************************/
 251
 252             r00              = _mm_mul_pd(rsq00,rinv00);
 253
 254             /* Compute parameters for interactions between i and j atoms */
 255             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 256                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 257             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 258                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 259
 260             /* Analytical LJ-PME */
 261             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 262             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 263             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 264             exponent         = gmx_simd_exp_d(ewcljrsq);
 265             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 266             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 267             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 268             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 269             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 270             vvdw             = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
 271             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 272             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);
 273
 274             /* Update potential sum for this i atom from the interaction with this j atom. */
 275             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 276
 277             fscal            = fvdw;
 278
 279             /* Calculate temporary vectorial force */
 280             tx               = _mm_mul_pd(fscal,dx00);
 281             ty               = _mm_mul_pd(fscal,dy00);
 282             tz               = _mm_mul_pd(fscal,dz00);
 283
 284             /* Update vectorial force */
 285             fix0             = _mm_add_pd(fix0,tx);
 286             fiy0             = _mm_add_pd(fiy0,ty);
 287             fiz0             = _mm_add_pd(fiz0,tz);
 288
 289             fjx0             = _mm_add_pd(fjx0,tx);
 290             fjy0             = _mm_add_pd(fjy0,ty);
 291             fjz0             = _mm_add_pd(fjz0,tz);
 292
 293             /**************************
 294              * CALCULATE INTERACTIONS *
 295              **************************/
 296
 297             r10              = _mm_mul_pd(rsq10,rinv10);
 298
 299             /* Compute parameters for interactions between i and j atoms */
 300             qq10             = _mm_mul_pd(iq1,jq0);
 301
 302             /* EWALD ELECTROSTATICS */
 303
 304             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 305             ewrt             = _mm_mul_pd(r10,ewtabscale);
 306             ewitab           = _mm_cvttpd_epi32(ewrt);
 307             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 308             ewitab           = _mm_slli_epi32(ewitab,2);
 309             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 310             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 311             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 312             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 313             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 314             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 315             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 316             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 317             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 318             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 319
 320             /* Update potential sum for this i atom from the interaction with this j atom. */
 321             velecsum         = _mm_add_pd(velecsum,velec);
 322
 323             fscal            = felec;
 324
 325             /* Calculate temporary vectorial force */
 326             tx               = _mm_mul_pd(fscal,dx10);
 327             ty               = _mm_mul_pd(fscal,dy10);
 328             tz               = _mm_mul_pd(fscal,dz10);
 329
 330             /* Update vectorial force */
 331             fix1             = _mm_add_pd(fix1,tx);
 332             fiy1             = _mm_add_pd(fiy1,ty);
 333             fiz1             = _mm_add_pd(fiz1,tz);
 334
 335             fjx0             = _mm_add_pd(fjx0,tx);
 336             fjy0             = _mm_add_pd(fjy0,ty);
 337             fjz0             = _mm_add_pd(fjz0,tz);
 338
 339             /**************************
 340              * CALCULATE INTERACTIONS *
 341              **************************/
 342
 343             r20              = _mm_mul_pd(rsq20,rinv20);
 344
 345             /* Compute parameters for interactions between i and j atoms */
 346             qq20             = _mm_mul_pd(iq2,jq0);
 347
 348             /* EWALD ELECTROSTATICS */
 349
 350             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 351             ewrt             = _mm_mul_pd(r20,ewtabscale);
 352             ewitab           = _mm_cvttpd_epi32(ewrt);
 353             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 354             ewitab           = _mm_slli_epi32(ewitab,2);
 355             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 356             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 357             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 358             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 359             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 360             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 361             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 362             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 363             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 364             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 365
 366             /* Update potential sum for this i atom from the interaction with this j atom. */
 367             velecsum         = _mm_add_pd(velecsum,velec);
 368
 369             fscal            = felec;
 370
 371             /* Calculate temporary vectorial force */
 372             tx               = _mm_mul_pd(fscal,dx20);
 373             ty               = _mm_mul_pd(fscal,dy20);
 374             tz               = _mm_mul_pd(fscal,dz20);
 375
 376             /* Update vectorial force */
 377             fix2             = _mm_add_pd(fix2,tx);
 378             fiy2             = _mm_add_pd(fiy2,ty);
 379             fiz2             = _mm_add_pd(fiz2,tz);
 380
 381             fjx0             = _mm_add_pd(fjx0,tx);
 382             fjy0             = _mm_add_pd(fjy0,ty);
 383             fjz0             = _mm_add_pd(fjz0,tz);
 384
 385             /**************************
 386              * CALCULATE INTERACTIONS *
 387              **************************/
 388
 389             r30              = _mm_mul_pd(rsq30,rinv30);
 390
 391             /* Compute parameters for interactions between i and j atoms */
 392             qq30             = _mm_mul_pd(iq3,jq0);
 393
 394             /* EWALD ELECTROSTATICS */
 395
 396             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 397             ewrt             = _mm_mul_pd(r30,ewtabscale);
 398             ewitab           = _mm_cvttpd_epi32(ewrt);
 399             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 400             ewitab           = _mm_slli_epi32(ewitab,2);
 401             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 402             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 403             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 404             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 405             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 406             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 407             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 408             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 409             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 410             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 411
 412             /* Update potential sum for this i atom from the interaction with this j atom. */
 413             velecsum         = _mm_add_pd(velecsum,velec);
 414
 415             fscal            = felec;
 416
 417             /* Calculate temporary vectorial force */
 418             tx               = _mm_mul_pd(fscal,dx30);
 419             ty               = _mm_mul_pd(fscal,dy30);
 420             tz               = _mm_mul_pd(fscal,dz30);
 421
 422             /* Update vectorial force */
 423             fix3             = _mm_add_pd(fix3,tx);
 424             fiy3             = _mm_add_pd(fiy3,ty);
 425             fiz3             = _mm_add_pd(fiz3,tz);
 426
 427             fjx0             = _mm_add_pd(fjx0,tx);
 428             fjy0             = _mm_add_pd(fjy0,ty);
 429             fjz0             = _mm_add_pd(fjz0,tz);
 430
 431             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 432
 433             /* Inner loop uses 177 flops */
 434         }
 435
 436         if(jidx<j_index_end)
 437         {
 438
 439             jnrA             = jjnr[jidx];
 440             j_coord_offsetA  = DIM*jnrA;
 441
 442             /* load j atom coordinates */
 443             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 444                                               &jx0,&jy0,&jz0);
 445
 446             /* Calculate displacement vector */
 447             dx00             = _mm_sub_pd(ix0,jx0);
 448             dy00             = _mm_sub_pd(iy0,jy0);
 449             dz00             = _mm_sub_pd(iz0,jz0);
 450             dx10             = _mm_sub_pd(ix1,jx0);
 451             dy10             = _mm_sub_pd(iy1,jy0);
 452             dz10             = _mm_sub_pd(iz1,jz0);
 453             dx20             = _mm_sub_pd(ix2,jx0);
 454             dy20             = _mm_sub_pd(iy2,jy0);
 455             dz20             = _mm_sub_pd(iz2,jz0);
 456             dx30             = _mm_sub_pd(ix3,jx0);
 457             dy30             = _mm_sub_pd(iy3,jy0);
 458             dz30             = _mm_sub_pd(iz3,jz0);
 459
 460             /* Calculate squared distance and things based on it */
 461             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 462             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 463             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 464             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 465
 466             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 467             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 468             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 469             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 470
 471             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 472             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 473             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 474             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 475
 476             /* Load parameters for j particles */
 477             jq0              = _mm_load_sd(charge+jnrA+0);
 478             vdwjidx0A        = 2*vdwtype[jnrA+0];
 479
 480             fjx0             = _mm_setzero_pd();
 481             fjy0             = _mm_setzero_pd();
 482             fjz0             = _mm_setzero_pd();
 483
 484             /**************************
 485              * CALCULATE INTERACTIONS *
 486              **************************/
 487
 488             r00              = _mm_mul_pd(rsq00,rinv00);
 489
 490             /* Compute parameters for interactions between i and j atoms */
 491             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 492
 493             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 494
 495             /* Analytical LJ-PME */
 496             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 497             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 498             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 499             exponent         = gmx_simd_exp_d(ewcljrsq);
 500             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 501             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 502             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 503             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 504             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 505             vvdw             = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
 506             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 507             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);
 508
 509             /* Update potential sum for this i atom from the interaction with this j atom. */
 510             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 511             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 512
 513             fscal            = fvdw;
 514
 515             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 516
 517             /* Calculate temporary vectorial force */
 518             tx               = _mm_mul_pd(fscal,dx00);
 519             ty               = _mm_mul_pd(fscal,dy00);
 520             tz               = _mm_mul_pd(fscal,dz00);
 521
 522             /* Update vectorial force */
 523             fix0             = _mm_add_pd(fix0,tx);
 524             fiy0             = _mm_add_pd(fiy0,ty);
 525             fiz0             = _mm_add_pd(fiz0,tz);
 526
 527             fjx0             = _mm_add_pd(fjx0,tx);
 528             fjy0             = _mm_add_pd(fjy0,ty);
 529             fjz0             = _mm_add_pd(fjz0,tz);
 530
 531             /**************************
 532              * CALCULATE INTERACTIONS *
 533              **************************/
 534
 535             r10              = _mm_mul_pd(rsq10,rinv10);
 536
 537             /* Compute parameters for interactions between i and j atoms */
 538             qq10             = _mm_mul_pd(iq1,jq0);
 539
 540             /* EWALD ELECTROSTATICS */
 541
 542             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 543             ewrt             = _mm_mul_pd(r10,ewtabscale);
 544             ewitab           = _mm_cvttpd_epi32(ewrt);
 545             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 546             ewitab           = _mm_slli_epi32(ewitab,2);
 547             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 548             ewtabD           = _mm_setzero_pd();
 549             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 550             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 551             ewtabFn          = _mm_setzero_pd();
 552             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 553             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 554             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 555             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 556             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 557
 558             /* Update potential sum for this i atom from the interaction with this j atom. */
 559             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 560             velecsum         = _mm_add_pd(velecsum,velec);
 561
 562             fscal            = felec;
 563
 564             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 565
 566             /* Calculate temporary vectorial force */
 567             tx               = _mm_mul_pd(fscal,dx10);
 568             ty               = _mm_mul_pd(fscal,dy10);
 569             tz               = _mm_mul_pd(fscal,dz10);
 570
 571             /* Update vectorial force */
 572             fix1             = _mm_add_pd(fix1,tx);
 573             fiy1             = _mm_add_pd(fiy1,ty);
 574             fiz1             = _mm_add_pd(fiz1,tz);
 575
 576             fjx0             = _mm_add_pd(fjx0,tx);
 577             fjy0             = _mm_add_pd(fjy0,ty);
 578             fjz0             = _mm_add_pd(fjz0,tz);
 579
 580             /**************************
 581              * CALCULATE INTERACTIONS *
 582              **************************/
 583
 584             r20              = _mm_mul_pd(rsq20,rinv20);
 585
 586             /* Compute parameters for interactions between i and j atoms */
 587             qq20             = _mm_mul_pd(iq2,jq0);
 588
 589             /* EWALD ELECTROSTATICS */
 590
 591             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 592             ewrt             = _mm_mul_pd(r20,ewtabscale);
 593             ewitab           = _mm_cvttpd_epi32(ewrt);
 594             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 595             ewitab           = _mm_slli_epi32(ewitab,2);
 596             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 597             ewtabD           = _mm_setzero_pd();
 598             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 599             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 600             ewtabFn          = _mm_setzero_pd();
 601             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 602             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 603             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 604             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 605             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 606
 607             /* Update potential sum for this i atom from the interaction with this j atom. */
 608             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 609             velecsum         = _mm_add_pd(velecsum,velec);
 610
 611             fscal            = felec;
 612
 613             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 614
 615             /* Calculate temporary vectorial force */
 616             tx               = _mm_mul_pd(fscal,dx20);
 617             ty               = _mm_mul_pd(fscal,dy20);
 618             tz               = _mm_mul_pd(fscal,dz20);
 619
 620             /* Update vectorial force */
 621             fix2             = _mm_add_pd(fix2,tx);
 622             fiy2             = _mm_add_pd(fiy2,ty);
 623             fiz2             = _mm_add_pd(fiz2,tz);
 624
 625             fjx0             = _mm_add_pd(fjx0,tx);
 626             fjy0             = _mm_add_pd(fjy0,ty);
 627             fjz0             = _mm_add_pd(fjz0,tz);
 628
 629             /**************************
 630              * CALCULATE INTERACTIONS *
 631              **************************/
 632
 633             r30              = _mm_mul_pd(rsq30,rinv30);
 634
 635             /* Compute parameters for interactions between i and j atoms */
 636             qq30             = _mm_mul_pd(iq3,jq0);
 637
 638             /* EWALD ELECTROSTATICS */
 639
 640             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 641             ewrt             = _mm_mul_pd(r30,ewtabscale);
 642             ewitab           = _mm_cvttpd_epi32(ewrt);
 643             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 644             ewitab           = _mm_slli_epi32(ewitab,2);
 645             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 646             ewtabD           = _mm_setzero_pd();
 647             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 648             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 649             ewtabFn          = _mm_setzero_pd();
 650             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 651             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 652             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 653             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 654             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 655
 656             /* Update potential sum for this i atom from the interaction with this j atom. */
 657             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 658             velecsum         = _mm_add_pd(velecsum,velec);
 659
 660             fscal            = felec;
 661
 662             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 663
 664             /* Calculate temporary vectorial force */
 665             tx               = _mm_mul_pd(fscal,dx30);
 666             ty               = _mm_mul_pd(fscal,dy30);
 667             tz               = _mm_mul_pd(fscal,dz30);
 668
 669             /* Update vectorial force */
 670             fix3             = _mm_add_pd(fix3,tx);
 671             fiy3             = _mm_add_pd(fiy3,ty);
 672             fiz3             = _mm_add_pd(fiz3,tz);
 673
 674             fjx0             = _mm_add_pd(fjx0,tx);
 675             fjy0             = _mm_add_pd(fjy0,ty);
 676             fjz0             = _mm_add_pd(fjz0,tz);
 677
 678             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 679
 680             /* Inner loop uses 177 flops */
 681         }
 682
 683         /* End of innermost loop */
 684
 685         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 686                                               f+i_coord_offset,fshift+i_shift_offset);
 687
 688         ggid                        = gid[iidx];
 689         /* Update potential energies */
 690         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 691         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 692
 693         /* Increment number of inner iterations */
 694         inneriter                  += j_index_end - j_index_start;
 695
 696         /* Outer loop uses 26 flops */
 697     }
 698
 699     /* Increment number of outer iterations */
 700     outeriter        += nri;
 701
 702     /* Update outer/inner flops */
 703
 704     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*177);
 705 }
 706 /*
 707  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_sse4_1_double
 708  * Electrostatics interaction: Ewald
 709  * VdW interaction:            LJEwald
 710  * Geometry:                   Water4-Particle
 711  * Calculate force/pot:        Force
 712  */
 713 void
 714 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_sse4_1_double
 715                     (t_nblist                    * gmx_restrict       nlist,
 716                      rvec                        * gmx_restrict          xx,
 717                      rvec                        * gmx_restrict          ff,
 718                      t_forcerec                  * gmx_restrict          fr,
 719                      t_mdatoms                   * gmx_restrict     mdatoms,
 720                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 721                      t_nrnb                      * gmx_restrict        nrnb)
 722 {
 723     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 724      * just 0 for non-waters.
 725      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 726      * jnr indices corresponding to data put in the four positions in the SIMD register.
 727      */
 728     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 729     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 730     int              jnrA,jnrB;
 731     int              j_coord_offsetA,j_coord_offsetB;
 732     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 733     real             rcutoff_scalar;
 734     real             *shiftvec,*fshift,*x,*f;
 735     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 736     int              vdwioffset0;
 737     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 738     int              vdwioffset1;
 739     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 740     int              vdwioffset2;
 741     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 742     int              vdwioffset3;
 743     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 744     int              vdwjidx0A,vdwjidx0B;
 745     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 746     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 747     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 748     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 749     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 750     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 751     real             *charge;
 752     int              nvdwtype;
 753     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 754     int              *vdwtype;
 755     real             *vdwparam;
 756     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 757     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 758     __m128d           c6grid_00;
 759     __m128d           c6grid_10;
 760     __m128d           c6grid_20;
 761     __m128d           c6grid_30;
 762     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 763     real             *vdwgridparam;
 764     __m128d           one_half = _mm_set1_pd(0.5);
 765     __m128d           minus_one = _mm_set1_pd(-1.0);
 766     __m128i          ewitab;
 767     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 768     real             *ewtab;
 769     __m128d          dummy_mask,cutoff_mask;
 770     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 771     __m128d          one     = _mm_set1_pd(1.0);
 772     __m128d          two     = _mm_set1_pd(2.0);
 773     x                = xx[0];
 774     f                = ff[0];
 775
 776     nri              = nlist->nri;
 777     iinr             = nlist->iinr;
 778     jindex           = nlist->jindex;
 779     jjnr             = nlist->jjnr;
 780     shiftidx         = nlist->shift;
 781     gid              = nlist->gid;
 782     shiftvec         = fr->shift_vec[0];
 783     fshift           = fr->fshift[0];
 784     facel            = _mm_set1_pd(fr->epsfac);
 785     charge           = mdatoms->chargeA;
 786     nvdwtype         = fr->ntype;
 787     vdwparam         = fr->nbfp;
 788     vdwtype          = mdatoms->typeA;
 789     vdwgridparam     = fr->ljpme_c6grid;
 790     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 791     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 792     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 793
 794     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 795     ewtab            = fr->ic->tabq_coul_F;
 796     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 797     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 798
 799     /* Setup water-specific parameters */
 800     inr              = nlist->iinr[0];
 801     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 802     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 803     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 804     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 805
 806     /* Avoid stupid compiler warnings */
 807     jnrA = jnrB = 0;
 808     j_coord_offsetA = 0;
 809     j_coord_offsetB = 0;
 810
 811     outeriter        = 0;
 812     inneriter        = 0;
 813
 814     /* Start outer loop over neighborlists */
 815     for(iidx=0; iidx<nri; iidx++)
 816     {
 817         /* Load shift vector for this list */
 818         i_shift_offset   = DIM*shiftidx[iidx];
 819
 820         /* Load limits for loop over neighbors */
 821         j_index_start    = jindex[iidx];
 822         j_index_end      = jindex[iidx+1];
 823
 824         /* Get outer coordinate index */
 825         inr              = iinr[iidx];
 826         i_coord_offset   = DIM*inr;
 827
 828         /* Load i particle coords and add shift vector */
 829         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 830                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 831
 832         fix0             = _mm_setzero_pd();
 833         fiy0             = _mm_setzero_pd();
 834         fiz0             = _mm_setzero_pd();
 835         fix1             = _mm_setzero_pd();
 836         fiy1             = _mm_setzero_pd();
 837         fiz1             = _mm_setzero_pd();
 838         fix2             = _mm_setzero_pd();
 839         fiy2             = _mm_setzero_pd();
 840         fiz2             = _mm_setzero_pd();
 841         fix3             = _mm_setzero_pd();
 842         fiy3             = _mm_setzero_pd();
 843         fiz3             = _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             dx30             = _mm_sub_pd(ix3,jx0);
 870             dy30             = _mm_sub_pd(iy3,jy0);
 871             dz30             = _mm_sub_pd(iz3,jz0);
 872
 873             /* Calculate squared distance and things based on it */
 874             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 875             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 876             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 877             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 878
 879             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 880             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 881             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 882             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 883
 884             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 885             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 886             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 887             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 888
 889             /* Load parameters for j particles */
 890             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 891             vdwjidx0A        = 2*vdwtype[jnrA+0];
 892             vdwjidx0B        = 2*vdwtype[jnrB+0];
 893
 894             fjx0             = _mm_setzero_pd();
 895             fjy0             = _mm_setzero_pd();
 896             fjz0             = _mm_setzero_pd();
 897
 898             /**************************
 899              * CALCULATE INTERACTIONS *
 900              **************************/
 901
 902             r00              = _mm_mul_pd(rsq00,rinv00);
 903
 904             /* Compute parameters for interactions between i and j atoms */
 905             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 906                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 907             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 908                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 909
 910             /* Analytical LJ-PME */
 911             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 912             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 913             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 914             exponent         = gmx_simd_exp_d(ewcljrsq);
 915             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 916             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 917             /* f6A = 6 * C6grid * (1 - poly) */
 918             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 919             /* f6B = C6grid * exponent * beta^6 */
 920             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 921             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 922             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);
 923
 924             fscal            = fvdw;
 925
 926             /* Calculate temporary vectorial force */
 927             tx               = _mm_mul_pd(fscal,dx00);
 928             ty               = _mm_mul_pd(fscal,dy00);
 929             tz               = _mm_mul_pd(fscal,dz00);
 930
 931             /* Update vectorial force */
 932             fix0             = _mm_add_pd(fix0,tx);
 933             fiy0             = _mm_add_pd(fiy0,ty);
 934             fiz0             = _mm_add_pd(fiz0,tz);
 935
 936             fjx0             = _mm_add_pd(fjx0,tx);
 937             fjy0             = _mm_add_pd(fjy0,ty);
 938             fjz0             = _mm_add_pd(fjz0,tz);
 939
 940             /**************************
 941              * CALCULATE INTERACTIONS *
 942              **************************/
 943
 944             r10              = _mm_mul_pd(rsq10,rinv10);
 945
 946             /* Compute parameters for interactions between i and j atoms */
 947             qq10             = _mm_mul_pd(iq1,jq0);
 948
 949             /* EWALD ELECTROSTATICS */
 950
 951             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 952             ewrt             = _mm_mul_pd(r10,ewtabscale);
 953             ewitab           = _mm_cvttpd_epi32(ewrt);
 954             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 955             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 956                                          &ewtabF,&ewtabFn);
 957             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 958             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 959
 960             fscal            = felec;
 961
 962             /* Calculate temporary vectorial force */
 963             tx               = _mm_mul_pd(fscal,dx10);
 964             ty               = _mm_mul_pd(fscal,dy10);
 965             tz               = _mm_mul_pd(fscal,dz10);
 966
 967             /* Update vectorial force */
 968             fix1             = _mm_add_pd(fix1,tx);
 969             fiy1             = _mm_add_pd(fiy1,ty);
 970             fiz1             = _mm_add_pd(fiz1,tz);
 971
 972             fjx0             = _mm_add_pd(fjx0,tx);
 973             fjy0             = _mm_add_pd(fjy0,ty);
 974             fjz0             = _mm_add_pd(fjz0,tz);
 975
 976             /**************************
 977              * CALCULATE INTERACTIONS *
 978              **************************/
 979
 980             r20              = _mm_mul_pd(rsq20,rinv20);
 981
 982             /* Compute parameters for interactions between i and j atoms */
 983             qq20             = _mm_mul_pd(iq2,jq0);
 984
 985             /* EWALD ELECTROSTATICS */
 986
 987             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 988             ewrt             = _mm_mul_pd(r20,ewtabscale);
 989             ewitab           = _mm_cvttpd_epi32(ewrt);
 990             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 991             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 992                                          &ewtabF,&ewtabFn);
 993             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 994             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 995
 996             fscal            = felec;
 997
 998             /* Calculate temporary vectorial force */
 999             tx               = _mm_mul_pd(fscal,dx20);
1000             ty               = _mm_mul_pd(fscal,dy20);
1001             tz               = _mm_mul_pd(fscal,dz20);
1002
1003             /* Update vectorial force */
1004             fix2             = _mm_add_pd(fix2,tx);
1005             fiy2             = _mm_add_pd(fiy2,ty);
1006             fiz2             = _mm_add_pd(fiz2,tz);
1007
1008             fjx0             = _mm_add_pd(fjx0,tx);
1009             fjy0             = _mm_add_pd(fjy0,ty);
1010             fjz0             = _mm_add_pd(fjz0,tz);
1011
1012             /**************************
1013              * CALCULATE INTERACTIONS *
1014              **************************/
1015
1016             r30              = _mm_mul_pd(rsq30,rinv30);
1017
1018             /* Compute parameters for interactions between i and j atoms */
1019             qq30             = _mm_mul_pd(iq3,jq0);
1020
1021             /* EWALD ELECTROSTATICS */
1022
1023             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1024             ewrt             = _mm_mul_pd(r30,ewtabscale);
1025             ewitab           = _mm_cvttpd_epi32(ewrt);
1026             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1027             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1028                                          &ewtabF,&ewtabFn);
1029             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1030             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1031
1032             fscal            = felec;
1033
1034             /* Calculate temporary vectorial force */
1035             tx               = _mm_mul_pd(fscal,dx30);
1036             ty               = _mm_mul_pd(fscal,dy30);
1037             tz               = _mm_mul_pd(fscal,dz30);
1038
1039             /* Update vectorial force */
1040             fix3             = _mm_add_pd(fix3,tx);
1041             fiy3             = _mm_add_pd(fiy3,ty);
1042             fiz3             = _mm_add_pd(fiz3,tz);
1043
1044             fjx0             = _mm_add_pd(fjx0,tx);
1045             fjy0             = _mm_add_pd(fjy0,ty);
1046             fjz0             = _mm_add_pd(fjz0,tz);
1047
1048             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1049
1050             /* Inner loop uses 157 flops */
1051         }
1052
1053         if(jidx<j_index_end)
1054         {
1055
1056             jnrA             = jjnr[jidx];
1057             j_coord_offsetA  = DIM*jnrA;
1058
1059             /* load j atom coordinates */
1060             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1061                                               &jx0,&jy0,&jz0);
1062
1063             /* Calculate displacement vector */
1064             dx00             = _mm_sub_pd(ix0,jx0);
1065             dy00             = _mm_sub_pd(iy0,jy0);
1066             dz00             = _mm_sub_pd(iz0,jz0);
1067             dx10             = _mm_sub_pd(ix1,jx0);
1068             dy10             = _mm_sub_pd(iy1,jy0);
1069             dz10             = _mm_sub_pd(iz1,jz0);
1070             dx20             = _mm_sub_pd(ix2,jx0);
1071             dy20             = _mm_sub_pd(iy2,jy0);
1072             dz20             = _mm_sub_pd(iz2,jz0);
1073             dx30             = _mm_sub_pd(ix3,jx0);
1074             dy30             = _mm_sub_pd(iy3,jy0);
1075             dz30             = _mm_sub_pd(iz3,jz0);
1076
1077             /* Calculate squared distance and things based on it */
1078             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1079             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1080             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1081             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1082
1083             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1084             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1085             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1086             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1087
1088             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1089             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1090             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1091             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1092
1093             /* Load parameters for j particles */
1094             jq0              = _mm_load_sd(charge+jnrA+0);
1095             vdwjidx0A        = 2*vdwtype[jnrA+0];
1096
1097             fjx0             = _mm_setzero_pd();
1098             fjy0             = _mm_setzero_pd();
1099             fjz0             = _mm_setzero_pd();
1100
1101             /**************************
1102              * CALCULATE INTERACTIONS *
1103              **************************/
1104
1105             r00              = _mm_mul_pd(rsq00,rinv00);
1106
1107             /* Compute parameters for interactions between i and j atoms */
1108             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1109
1110             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
1111
1112             /* Analytical LJ-PME */
1113             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1114             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1115             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1116             exponent         = gmx_simd_exp_d(ewcljrsq);
1117             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1118             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1119             /* f6A = 6 * C6grid * (1 - poly) */
1120             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1121             /* f6B = C6grid * exponent * beta^6 */
1122             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1123             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1124             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);
1125
1126             fscal            = fvdw;
1127
1128             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1129
1130             /* Calculate temporary vectorial force */
1131             tx               = _mm_mul_pd(fscal,dx00);
1132             ty               = _mm_mul_pd(fscal,dy00);
1133             tz               = _mm_mul_pd(fscal,dz00);
1134
1135             /* Update vectorial force */
1136             fix0             = _mm_add_pd(fix0,tx);
1137             fiy0             = _mm_add_pd(fiy0,ty);
1138             fiz0             = _mm_add_pd(fiz0,tz);
1139
1140             fjx0             = _mm_add_pd(fjx0,tx);
1141             fjy0             = _mm_add_pd(fjy0,ty);
1142             fjz0             = _mm_add_pd(fjz0,tz);
1143
1144             /**************************
1145              * CALCULATE INTERACTIONS *
1146              **************************/
1147
1148             r10              = _mm_mul_pd(rsq10,rinv10);
1149
1150             /* Compute parameters for interactions between i and j atoms */
1151             qq10             = _mm_mul_pd(iq1,jq0);
1152
1153             /* EWALD ELECTROSTATICS */
1154
1155             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1156             ewrt             = _mm_mul_pd(r10,ewtabscale);
1157             ewitab           = _mm_cvttpd_epi32(ewrt);
1158             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1159             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1160             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1161             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1162
1163             fscal            = felec;
1164
1165             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1166
1167             /* Calculate temporary vectorial force */
1168             tx               = _mm_mul_pd(fscal,dx10);
1169             ty               = _mm_mul_pd(fscal,dy10);
1170             tz               = _mm_mul_pd(fscal,dz10);
1171
1172             /* Update vectorial force */
1173             fix1             = _mm_add_pd(fix1,tx);
1174             fiy1             = _mm_add_pd(fiy1,ty);
1175             fiz1             = _mm_add_pd(fiz1,tz);
1176
1177             fjx0             = _mm_add_pd(fjx0,tx);
1178             fjy0             = _mm_add_pd(fjy0,ty);
1179             fjz0             = _mm_add_pd(fjz0,tz);
1180
1181             /**************************
1182              * CALCULATE INTERACTIONS *
1183              **************************/
1184
1185             r20              = _mm_mul_pd(rsq20,rinv20);
1186
1187             /* Compute parameters for interactions between i and j atoms */
1188             qq20             = _mm_mul_pd(iq2,jq0);
1189
1190             /* EWALD ELECTROSTATICS */
1191
1192             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1193             ewrt             = _mm_mul_pd(r20,ewtabscale);
1194             ewitab           = _mm_cvttpd_epi32(ewrt);
1195             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1196             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1197             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1198             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1199
1200             fscal            = felec;
1201
1202             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1203
1204             /* Calculate temporary vectorial force */
1205             tx               = _mm_mul_pd(fscal,dx20);
1206             ty               = _mm_mul_pd(fscal,dy20);
1207             tz               = _mm_mul_pd(fscal,dz20);
1208
1209             /* Update vectorial force */
1210             fix2             = _mm_add_pd(fix2,tx);
1211             fiy2             = _mm_add_pd(fiy2,ty);
1212             fiz2             = _mm_add_pd(fiz2,tz);
1213
1214             fjx0             = _mm_add_pd(fjx0,tx);
1215             fjy0             = _mm_add_pd(fjy0,ty);
1216             fjz0             = _mm_add_pd(fjz0,tz);
1217
1218             /**************************
1219              * CALCULATE INTERACTIONS *
1220              **************************/
1221
1222             r30              = _mm_mul_pd(rsq30,rinv30);
1223
1224             /* Compute parameters for interactions between i and j atoms */
1225             qq30             = _mm_mul_pd(iq3,jq0);
1226
1227             /* EWALD ELECTROSTATICS */
1228
1229             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1230             ewrt             = _mm_mul_pd(r30,ewtabscale);
1231             ewitab           = _mm_cvttpd_epi32(ewrt);
1232             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1233             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1234             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1235             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1236
1237             fscal            = felec;
1238
1239             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1240
1241             /* Calculate temporary vectorial force */
1242             tx               = _mm_mul_pd(fscal,dx30);
1243             ty               = _mm_mul_pd(fscal,dy30);
1244             tz               = _mm_mul_pd(fscal,dz30);
1245
1246             /* Update vectorial force */
1247             fix3             = _mm_add_pd(fix3,tx);
1248             fiy3             = _mm_add_pd(fiy3,ty);
1249             fiz3             = _mm_add_pd(fiz3,tz);
1250
1251             fjx0             = _mm_add_pd(fjx0,tx);
1252             fjy0             = _mm_add_pd(fjy0,ty);
1253             fjz0             = _mm_add_pd(fjz0,tz);
1254
1255             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1256
1257             /* Inner loop uses 157 flops */
1258         }
1259
1260         /* End of innermost loop */
1261
1262         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1263                                               f+i_coord_offset,fshift+i_shift_offset);
1264
1265         /* Increment number of inner iterations */
1266         inneriter                  += j_index_end - j_index_start;
1267
1268         /* Outer loop uses 24 flops */
1269     }
1270
1271     /* Increment number of outer iterations */
1272     outeriter        += nri;
1273
1274     /* Update outer/inner flops */
1275
1276     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*157);
1277 }