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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_double.h"
  48 #include "kernelutil_x86_sse4_1_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW4P1_VF_sse4_1_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_VF_sse4_1_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwioffset3;
  87     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  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          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  94     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 101     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 102     __m128d           c6grid_00;
 103     __m128d           c6grid_10;
 104     __m128d           c6grid_20;
 105     __m128d           c6grid_30;
 106     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 107     real             *vdwgridparam;
 108     __m128d           one_half = _mm_set1_pd(0.5);
 109     __m128d           minus_one = _mm_set1_pd(-1.0);
 110     __m128i          ewitab;
 111     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 112     real             *ewtab;
 113     __m128d          dummy_mask,cutoff_mask;
 114     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 115     __m128d          one     = _mm_set1_pd(1.0);
 116     __m128d          two     = _mm_set1_pd(2.0);
 117     x                = xx[0];
 118     f                = ff[0];
 119
 120     nri              = nlist->nri;
 121     iinr             = nlist->iinr;
 122     jindex           = nlist->jindex;
 123     jjnr             = nlist->jjnr;
 124     shiftidx         = nlist->shift;
 125     gid              = nlist->gid;
 126     shiftvec         = fr->shift_vec[0];
 127     fshift           = fr->fshift[0];
 128     facel            = _mm_set1_pd(fr->epsfac);
 129     charge           = mdatoms->chargeA;
 130     nvdwtype         = fr->ntype;
 131     vdwparam         = fr->nbfp;
 132     vdwtype          = mdatoms->typeA;
 133     vdwgridparam     = fr->ljpme_c6grid;
 134     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 135     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 136     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 137
 138     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 139     ewtab            = fr->ic->tabq_coul_FDV0;
 140     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 141     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 142
 143     /* Setup water-specific parameters */
 144     inr              = nlist->iinr[0];
 145     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 146     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 147     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 148     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 149
 150     /* Avoid stupid compiler warnings */
 151     jnrA = jnrB = 0;
 152     j_coord_offsetA = 0;
 153     j_coord_offsetB = 0;
 154
 155     outeriter        = 0;
 156     inneriter        = 0;
 157
 158     /* Start outer loop over neighborlists */
 159     for(iidx=0; iidx<nri; iidx++)
 160     {
 161         /* Load shift vector for this list */
 162         i_shift_offset   = DIM*shiftidx[iidx];
 163
 164         /* Load limits for loop over neighbors */
 165         j_index_start    = jindex[iidx];
 166         j_index_end      = jindex[iidx+1];
 167
 168         /* Get outer coordinate index */
 169         inr              = iinr[iidx];
 170         i_coord_offset   = DIM*inr;
 171
 172         /* Load i particle coords and add shift vector */
 173         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 174                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 175
 176         fix0             = _mm_setzero_pd();
 177         fiy0             = _mm_setzero_pd();
 178         fiz0             = _mm_setzero_pd();
 179         fix1             = _mm_setzero_pd();
 180         fiy1             = _mm_setzero_pd();
 181         fiz1             = _mm_setzero_pd();
 182         fix2             = _mm_setzero_pd();
 183         fiy2             = _mm_setzero_pd();
 184         fiz2             = _mm_setzero_pd();
 185         fix3             = _mm_setzero_pd();
 186         fiy3             = _mm_setzero_pd();
 187         fiz3             = _mm_setzero_pd();
 188
 189         /* Reset potential sums */
 190         velecsum         = _mm_setzero_pd();
 191         vvdwsum          = _mm_setzero_pd();
 192
 193         /* Start inner kernel loop */
 194         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 195         {
 196
 197             /* Get j neighbor index, and coordinate index */
 198             jnrA             = jjnr[jidx];
 199             jnrB             = jjnr[jidx+1];
 200             j_coord_offsetA  = DIM*jnrA;
 201             j_coord_offsetB  = DIM*jnrB;
 202
 203             /* load j atom coordinates */
 204             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 205                                               &jx0,&jy0,&jz0);
 206
 207             /* Calculate displacement vector */
 208             dx00             = _mm_sub_pd(ix0,jx0);
 209             dy00             = _mm_sub_pd(iy0,jy0);
 210             dz00             = _mm_sub_pd(iz0,jz0);
 211             dx10             = _mm_sub_pd(ix1,jx0);
 212             dy10             = _mm_sub_pd(iy1,jy0);
 213             dz10             = _mm_sub_pd(iz1,jz0);
 214             dx20             = _mm_sub_pd(ix2,jx0);
 215             dy20             = _mm_sub_pd(iy2,jy0);
 216             dz20             = _mm_sub_pd(iz2,jz0);
 217             dx30             = _mm_sub_pd(ix3,jx0);
 218             dy30             = _mm_sub_pd(iy3,jy0);
 219             dz30             = _mm_sub_pd(iz3,jz0);
 220
 221             /* Calculate squared distance and things based on it */
 222             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 223             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 224             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 225             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 226
 227             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 228             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 229             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 230             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 231
 232             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 233             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 234             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 235             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 236
 237             /* Load parameters for j particles */
 238             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 239             vdwjidx0A        = 2*vdwtype[jnrA+0];
 240             vdwjidx0B        = 2*vdwtype[jnrB+0];
 241
 242             fjx0             = _mm_setzero_pd();
 243             fjy0             = _mm_setzero_pd();
 244             fjz0             = _mm_setzero_pd();
 245
 246             /**************************
 247              * CALCULATE INTERACTIONS *
 248              **************************/
 249
 250             r00              = _mm_mul_pd(rsq00,rinv00);
 251
 252             /* Compute parameters for interactions between i and j atoms */
 253             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 254                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 255             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 256                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 257
 258             /* Analytical LJ-PME */
 259             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 260             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 261             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 262             exponent         = gmx_simd_exp_d(ewcljrsq);
 263             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 264             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 265             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 266             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 267             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 268             vvdw             = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
 269             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 270             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);
 271
 272             /* Update potential sum for this i atom from the interaction with this j atom. */
 273             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 274
 275             fscal            = fvdw;
 276
 277             /* Calculate temporary vectorial force */
 278             tx               = _mm_mul_pd(fscal,dx00);
 279             ty               = _mm_mul_pd(fscal,dy00);
 280             tz               = _mm_mul_pd(fscal,dz00);
 281
 282             /* Update vectorial force */
 283             fix0             = _mm_add_pd(fix0,tx);
 284             fiy0             = _mm_add_pd(fiy0,ty);
 285             fiz0             = _mm_add_pd(fiz0,tz);
 286
 287             fjx0             = _mm_add_pd(fjx0,tx);
 288             fjy0             = _mm_add_pd(fjy0,ty);
 289             fjz0             = _mm_add_pd(fjz0,tz);
 290
 291             /**************************
 292              * CALCULATE INTERACTIONS *
 293              **************************/
 294
 295             r10              = _mm_mul_pd(rsq10,rinv10);
 296
 297             /* Compute parameters for interactions between i and j atoms */
 298             qq10             = _mm_mul_pd(iq1,jq0);
 299
 300             /* EWALD ELECTROSTATICS */
 301
 302             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 303             ewrt             = _mm_mul_pd(r10,ewtabscale);
 304             ewitab           = _mm_cvttpd_epi32(ewrt);
 305             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 306             ewitab           = _mm_slli_epi32(ewitab,2);
 307             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 308             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 309             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 310             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 311             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 312             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 313             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 314             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 315             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 316             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 317
 318             /* Update potential sum for this i atom from the interaction with this j atom. */
 319             velecsum         = _mm_add_pd(velecsum,velec);
 320
 321             fscal            = felec;
 322
 323             /* Calculate temporary vectorial force */
 324             tx               = _mm_mul_pd(fscal,dx10);
 325             ty               = _mm_mul_pd(fscal,dy10);
 326             tz               = _mm_mul_pd(fscal,dz10);
 327
 328             /* Update vectorial force */
 329             fix1             = _mm_add_pd(fix1,tx);
 330             fiy1             = _mm_add_pd(fiy1,ty);
 331             fiz1             = _mm_add_pd(fiz1,tz);
 332
 333             fjx0             = _mm_add_pd(fjx0,tx);
 334             fjy0             = _mm_add_pd(fjy0,ty);
 335             fjz0             = _mm_add_pd(fjz0,tz);
 336
 337             /**************************
 338              * CALCULATE INTERACTIONS *
 339              **************************/
 340
 341             r20              = _mm_mul_pd(rsq20,rinv20);
 342
 343             /* Compute parameters for interactions between i and j atoms */
 344             qq20             = _mm_mul_pd(iq2,jq0);
 345
 346             /* EWALD ELECTROSTATICS */
 347
 348             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 349             ewrt             = _mm_mul_pd(r20,ewtabscale);
 350             ewitab           = _mm_cvttpd_epi32(ewrt);
 351             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 352             ewitab           = _mm_slli_epi32(ewitab,2);
 353             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 354             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 355             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 356             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 357             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 358             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 359             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 360             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 361             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 362             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 363
 364             /* Update potential sum for this i atom from the interaction with this j atom. */
 365             velecsum         = _mm_add_pd(velecsum,velec);
 366
 367             fscal            = felec;
 368
 369             /* Calculate temporary vectorial force */
 370             tx               = _mm_mul_pd(fscal,dx20);
 371             ty               = _mm_mul_pd(fscal,dy20);
 372             tz               = _mm_mul_pd(fscal,dz20);
 373
 374             /* Update vectorial force */
 375             fix2             = _mm_add_pd(fix2,tx);
 376             fiy2             = _mm_add_pd(fiy2,ty);
 377             fiz2             = _mm_add_pd(fiz2,tz);
 378
 379             fjx0             = _mm_add_pd(fjx0,tx);
 380             fjy0             = _mm_add_pd(fjy0,ty);
 381             fjz0             = _mm_add_pd(fjz0,tz);
 382
 383             /**************************
 384              * CALCULATE INTERACTIONS *
 385              **************************/
 386
 387             r30              = _mm_mul_pd(rsq30,rinv30);
 388
 389             /* Compute parameters for interactions between i and j atoms */
 390             qq30             = _mm_mul_pd(iq3,jq0);
 391
 392             /* EWALD ELECTROSTATICS */
 393
 394             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 395             ewrt             = _mm_mul_pd(r30,ewtabscale);
 396             ewitab           = _mm_cvttpd_epi32(ewrt);
 397             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 398             ewitab           = _mm_slli_epi32(ewitab,2);
 399             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 400             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 401             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 402             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 403             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 404             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 405             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 406             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 407             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 408             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 409
 410             /* Update potential sum for this i atom from the interaction with this j atom. */
 411             velecsum         = _mm_add_pd(velecsum,velec);
 412
 413             fscal            = felec;
 414
 415             /* Calculate temporary vectorial force */
 416             tx               = _mm_mul_pd(fscal,dx30);
 417             ty               = _mm_mul_pd(fscal,dy30);
 418             tz               = _mm_mul_pd(fscal,dz30);
 419
 420             /* Update vectorial force */
 421             fix3             = _mm_add_pd(fix3,tx);
 422             fiy3             = _mm_add_pd(fiy3,ty);
 423             fiz3             = _mm_add_pd(fiz3,tz);
 424
 425             fjx0             = _mm_add_pd(fjx0,tx);
 426             fjy0             = _mm_add_pd(fjy0,ty);
 427             fjz0             = _mm_add_pd(fjz0,tz);
 428
 429             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 430
 431             /* Inner loop uses 177 flops */
 432         }
 433
 434         if(jidx<j_index_end)
 435         {
 436
 437             jnrA             = jjnr[jidx];
 438             j_coord_offsetA  = DIM*jnrA;
 439
 440             /* load j atom coordinates */
 441             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 442                                               &jx0,&jy0,&jz0);
 443
 444             /* Calculate displacement vector */
 445             dx00             = _mm_sub_pd(ix0,jx0);
 446             dy00             = _mm_sub_pd(iy0,jy0);
 447             dz00             = _mm_sub_pd(iz0,jz0);
 448             dx10             = _mm_sub_pd(ix1,jx0);
 449             dy10             = _mm_sub_pd(iy1,jy0);
 450             dz10             = _mm_sub_pd(iz1,jz0);
 451             dx20             = _mm_sub_pd(ix2,jx0);
 452             dy20             = _mm_sub_pd(iy2,jy0);
 453             dz20             = _mm_sub_pd(iz2,jz0);
 454             dx30             = _mm_sub_pd(ix3,jx0);
 455             dy30             = _mm_sub_pd(iy3,jy0);
 456             dz30             = _mm_sub_pd(iz3,jz0);
 457
 458             /* Calculate squared distance and things based on it */
 459             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 460             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 461             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 462             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 463
 464             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 465             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 466             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 467             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 468
 469             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 470             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 471             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 472             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 473
 474             /* Load parameters for j particles */
 475             jq0              = _mm_load_sd(charge+jnrA+0);
 476             vdwjidx0A        = 2*vdwtype[jnrA+0];
 477
 478             fjx0             = _mm_setzero_pd();
 479             fjy0             = _mm_setzero_pd();
 480             fjz0             = _mm_setzero_pd();
 481
 482             /**************************
 483              * CALCULATE INTERACTIONS *
 484              **************************/
 485
 486             r00              = _mm_mul_pd(rsq00,rinv00);
 487
 488             /* Compute parameters for interactions between i and j atoms */
 489             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 490
 491             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 492
 493             /* Analytical LJ-PME */
 494             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 495             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 496             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 497             exponent         = gmx_simd_exp_d(ewcljrsq);
 498             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 499             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 500             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 501             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 502             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 503             vvdw             = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
 504             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 505             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);
 506
 507             /* Update potential sum for this i atom from the interaction with this j atom. */
 508             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 509             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 510
 511             fscal            = fvdw;
 512
 513             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 514
 515             /* Calculate temporary vectorial force */
 516             tx               = _mm_mul_pd(fscal,dx00);
 517             ty               = _mm_mul_pd(fscal,dy00);
 518             tz               = _mm_mul_pd(fscal,dz00);
 519
 520             /* Update vectorial force */
 521             fix0             = _mm_add_pd(fix0,tx);
 522             fiy0             = _mm_add_pd(fiy0,ty);
 523             fiz0             = _mm_add_pd(fiz0,tz);
 524
 525             fjx0             = _mm_add_pd(fjx0,tx);
 526             fjy0             = _mm_add_pd(fjy0,ty);
 527             fjz0             = _mm_add_pd(fjz0,tz);
 528
 529             /**************************
 530              * CALCULATE INTERACTIONS *
 531              **************************/
 532
 533             r10              = _mm_mul_pd(rsq10,rinv10);
 534
 535             /* Compute parameters for interactions between i and j atoms */
 536             qq10             = _mm_mul_pd(iq1,jq0);
 537
 538             /* EWALD ELECTROSTATICS */
 539
 540             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 541             ewrt             = _mm_mul_pd(r10,ewtabscale);
 542             ewitab           = _mm_cvttpd_epi32(ewrt);
 543             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 544             ewitab           = _mm_slli_epi32(ewitab,2);
 545             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 546             ewtabD           = _mm_setzero_pd();
 547             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 548             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 549             ewtabFn          = _mm_setzero_pd();
 550             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 551             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 552             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 553             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 554             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 555
 556             /* Update potential sum for this i atom from the interaction with this j atom. */
 557             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 558             velecsum         = _mm_add_pd(velecsum,velec);
 559
 560             fscal            = felec;
 561
 562             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 563
 564             /* Calculate temporary vectorial force */
 565             tx               = _mm_mul_pd(fscal,dx10);
 566             ty               = _mm_mul_pd(fscal,dy10);
 567             tz               = _mm_mul_pd(fscal,dz10);
 568
 569             /* Update vectorial force */
 570             fix1             = _mm_add_pd(fix1,tx);
 571             fiy1             = _mm_add_pd(fiy1,ty);
 572             fiz1             = _mm_add_pd(fiz1,tz);
 573
 574             fjx0             = _mm_add_pd(fjx0,tx);
 575             fjy0             = _mm_add_pd(fjy0,ty);
 576             fjz0             = _mm_add_pd(fjz0,tz);
 577
 578             /**************************
 579              * CALCULATE INTERACTIONS *
 580              **************************/
 581
 582             r20              = _mm_mul_pd(rsq20,rinv20);
 583
 584             /* Compute parameters for interactions between i and j atoms */
 585             qq20             = _mm_mul_pd(iq2,jq0);
 586
 587             /* EWALD ELECTROSTATICS */
 588
 589             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 590             ewrt             = _mm_mul_pd(r20,ewtabscale);
 591             ewitab           = _mm_cvttpd_epi32(ewrt);
 592             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 593             ewitab           = _mm_slli_epi32(ewitab,2);
 594             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 595             ewtabD           = _mm_setzero_pd();
 596             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 597             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 598             ewtabFn          = _mm_setzero_pd();
 599             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 600             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 601             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 602             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 603             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 604
 605             /* Update potential sum for this i atom from the interaction with this j atom. */
 606             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 607             velecsum         = _mm_add_pd(velecsum,velec);
 608
 609             fscal            = felec;
 610
 611             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 612
 613             /* Calculate temporary vectorial force */
 614             tx               = _mm_mul_pd(fscal,dx20);
 615             ty               = _mm_mul_pd(fscal,dy20);
 616             tz               = _mm_mul_pd(fscal,dz20);
 617
 618             /* Update vectorial force */
 619             fix2             = _mm_add_pd(fix2,tx);
 620             fiy2             = _mm_add_pd(fiy2,ty);
 621             fiz2             = _mm_add_pd(fiz2,tz);
 622
 623             fjx0             = _mm_add_pd(fjx0,tx);
 624             fjy0             = _mm_add_pd(fjy0,ty);
 625             fjz0             = _mm_add_pd(fjz0,tz);
 626
 627             /**************************
 628              * CALCULATE INTERACTIONS *
 629              **************************/
 630
 631             r30              = _mm_mul_pd(rsq30,rinv30);
 632
 633             /* Compute parameters for interactions between i and j atoms */
 634             qq30             = _mm_mul_pd(iq3,jq0);
 635
 636             /* EWALD ELECTROSTATICS */
 637
 638             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 639             ewrt             = _mm_mul_pd(r30,ewtabscale);
 640             ewitab           = _mm_cvttpd_epi32(ewrt);
 641             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 642             ewitab           = _mm_slli_epi32(ewitab,2);
 643             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 644             ewtabD           = _mm_setzero_pd();
 645             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 646             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 647             ewtabFn          = _mm_setzero_pd();
 648             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 649             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 650             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 651             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 652             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 653
 654             /* Update potential sum for this i atom from the interaction with this j atom. */
 655             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 656             velecsum         = _mm_add_pd(velecsum,velec);
 657
 658             fscal            = felec;
 659
 660             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 661
 662             /* Calculate temporary vectorial force */
 663             tx               = _mm_mul_pd(fscal,dx30);
 664             ty               = _mm_mul_pd(fscal,dy30);
 665             tz               = _mm_mul_pd(fscal,dz30);
 666
 667             /* Update vectorial force */
 668             fix3             = _mm_add_pd(fix3,tx);
 669             fiy3             = _mm_add_pd(fiy3,ty);
 670             fiz3             = _mm_add_pd(fiz3,tz);
 671
 672             fjx0             = _mm_add_pd(fjx0,tx);
 673             fjy0             = _mm_add_pd(fjy0,ty);
 674             fjz0             = _mm_add_pd(fjz0,tz);
 675
 676             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 677
 678             /* Inner loop uses 177 flops */
 679         }
 680
 681         /* End of innermost loop */
 682
 683         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 684                                               f+i_coord_offset,fshift+i_shift_offset);
 685
 686         ggid                        = gid[iidx];
 687         /* Update potential energies */
 688         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 689         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 690
 691         /* Increment number of inner iterations */
 692         inneriter                  += j_index_end - j_index_start;
 693
 694         /* Outer loop uses 26 flops */
 695     }
 696
 697     /* Increment number of outer iterations */
 698     outeriter        += nri;
 699
 700     /* Update outer/inner flops */
 701
 702     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*177);
 703 }
 704 /*
 705  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_sse4_1_double
 706  * Electrostatics interaction: Ewald
 707  * VdW interaction:            LJEwald
 708  * Geometry:                   Water4-Particle
 709  * Calculate force/pot:        Force
 710  */
 711 void
 712 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_sse4_1_double
 713                     (t_nblist                    * gmx_restrict       nlist,
 714                      rvec                        * gmx_restrict          xx,
 715                      rvec                        * gmx_restrict          ff,
 716                      t_forcerec                  * gmx_restrict          fr,
 717                      t_mdatoms                   * gmx_restrict     mdatoms,
 718                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 719                      t_nrnb                      * gmx_restrict        nrnb)
 720 {
 721     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 722      * just 0 for non-waters.
 723      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 724      * jnr indices corresponding to data put in the four positions in the SIMD register.
 725      */
 726     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 727     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 728     int              jnrA,jnrB;
 729     int              j_coord_offsetA,j_coord_offsetB;
 730     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 731     real             rcutoff_scalar;
 732     real             *shiftvec,*fshift,*x,*f;
 733     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 734     int              vdwioffset0;
 735     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 736     int              vdwioffset1;
 737     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 738     int              vdwioffset2;
 739     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 740     int              vdwioffset3;
 741     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 742     int              vdwjidx0A,vdwjidx0B;
 743     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 744     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 745     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 746     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 747     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 748     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 749     real             *charge;
 750     int              nvdwtype;
 751     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 752     int              *vdwtype;
 753     real             *vdwparam;
 754     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 755     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 756     __m128d           c6grid_00;
 757     __m128d           c6grid_10;
 758     __m128d           c6grid_20;
 759     __m128d           c6grid_30;
 760     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 761     real             *vdwgridparam;
 762     __m128d           one_half = _mm_set1_pd(0.5);
 763     __m128d           minus_one = _mm_set1_pd(-1.0);
 764     __m128i          ewitab;
 765     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 766     real             *ewtab;
 767     __m128d          dummy_mask,cutoff_mask;
 768     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 769     __m128d          one     = _mm_set1_pd(1.0);
 770     __m128d          two     = _mm_set1_pd(2.0);
 771     x                = xx[0];
 772     f                = ff[0];
 773
 774     nri              = nlist->nri;
 775     iinr             = nlist->iinr;
 776     jindex           = nlist->jindex;
 777     jjnr             = nlist->jjnr;
 778     shiftidx         = nlist->shift;
 779     gid              = nlist->gid;
 780     shiftvec         = fr->shift_vec[0];
 781     fshift           = fr->fshift[0];
 782     facel            = _mm_set1_pd(fr->epsfac);
 783     charge           = mdatoms->chargeA;
 784     nvdwtype         = fr->ntype;
 785     vdwparam         = fr->nbfp;
 786     vdwtype          = mdatoms->typeA;
 787     vdwgridparam     = fr->ljpme_c6grid;
 788     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 789     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 790     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 791
 792     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 793     ewtab            = fr->ic->tabq_coul_F;
 794     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 795     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 796
 797     /* Setup water-specific parameters */
 798     inr              = nlist->iinr[0];
 799     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 800     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 801     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 802     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 803
 804     /* Avoid stupid compiler warnings */
 805     jnrA = jnrB = 0;
 806     j_coord_offsetA = 0;
 807     j_coord_offsetB = 0;
 808
 809     outeriter        = 0;
 810     inneriter        = 0;
 811
 812     /* Start outer loop over neighborlists */
 813     for(iidx=0; iidx<nri; iidx++)
 814     {
 815         /* Load shift vector for this list */
 816         i_shift_offset   = DIM*shiftidx[iidx];
 817
 818         /* Load limits for loop over neighbors */
 819         j_index_start    = jindex[iidx];
 820         j_index_end      = jindex[iidx+1];
 821
 822         /* Get outer coordinate index */
 823         inr              = iinr[iidx];
 824         i_coord_offset   = DIM*inr;
 825
 826         /* Load i particle coords and add shift vector */
 827         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 828                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 829
 830         fix0             = _mm_setzero_pd();
 831         fiy0             = _mm_setzero_pd();
 832         fiz0             = _mm_setzero_pd();
 833         fix1             = _mm_setzero_pd();
 834         fiy1             = _mm_setzero_pd();
 835         fiz1             = _mm_setzero_pd();
 836         fix2             = _mm_setzero_pd();
 837         fiy2             = _mm_setzero_pd();
 838         fiz2             = _mm_setzero_pd();
 839         fix3             = _mm_setzero_pd();
 840         fiy3             = _mm_setzero_pd();
 841         fiz3             = _mm_setzero_pd();
 842
 843         /* Start inner kernel loop */
 844         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 845         {
 846
 847             /* Get j neighbor index, and coordinate index */
 848             jnrA             = jjnr[jidx];
 849             jnrB             = jjnr[jidx+1];
 850             j_coord_offsetA  = DIM*jnrA;
 851             j_coord_offsetB  = DIM*jnrB;
 852
 853             /* load j atom coordinates */
 854             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 855                                               &jx0,&jy0,&jz0);
 856
 857             /* Calculate displacement vector */
 858             dx00             = _mm_sub_pd(ix0,jx0);
 859             dy00             = _mm_sub_pd(iy0,jy0);
 860             dz00             = _mm_sub_pd(iz0,jz0);
 861             dx10             = _mm_sub_pd(ix1,jx0);
 862             dy10             = _mm_sub_pd(iy1,jy0);
 863             dz10             = _mm_sub_pd(iz1,jz0);
 864             dx20             = _mm_sub_pd(ix2,jx0);
 865             dy20             = _mm_sub_pd(iy2,jy0);
 866             dz20             = _mm_sub_pd(iz2,jz0);
 867             dx30             = _mm_sub_pd(ix3,jx0);
 868             dy30             = _mm_sub_pd(iy3,jy0);
 869             dz30             = _mm_sub_pd(iz3,jz0);
 870
 871             /* Calculate squared distance and things based on it */
 872             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 873             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 874             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 875             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 876
 877             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 878             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 879             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 880             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 881
 882             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 883             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 884             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 885             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 886
 887             /* Load parameters for j particles */
 888             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 889             vdwjidx0A        = 2*vdwtype[jnrA+0];
 890             vdwjidx0B        = 2*vdwtype[jnrB+0];
 891
 892             fjx0             = _mm_setzero_pd();
 893             fjy0             = _mm_setzero_pd();
 894             fjz0             = _mm_setzero_pd();
 895
 896             /**************************
 897              * CALCULATE INTERACTIONS *
 898              **************************/
 899
 900             r00              = _mm_mul_pd(rsq00,rinv00);
 901
 902             /* Compute parameters for interactions between i and j atoms */
 903             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 904                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 905             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 906                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 907
 908             /* Analytical LJ-PME */
 909             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 910             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 911             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 912             exponent         = gmx_simd_exp_d(ewcljrsq);
 913             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 914             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 915             /* f6A = 6 * C6grid * (1 - poly) */
 916             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 917             /* f6B = C6grid * exponent * beta^6 */
 918             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 919             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 920             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);
 921
 922             fscal            = fvdw;
 923
 924             /* Calculate temporary vectorial force */
 925             tx               = _mm_mul_pd(fscal,dx00);
 926             ty               = _mm_mul_pd(fscal,dy00);
 927             tz               = _mm_mul_pd(fscal,dz00);
 928
 929             /* Update vectorial force */
 930             fix0             = _mm_add_pd(fix0,tx);
 931             fiy0             = _mm_add_pd(fiy0,ty);
 932             fiz0             = _mm_add_pd(fiz0,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             /**************************
 939              * CALCULATE INTERACTIONS *
 940              **************************/
 941
 942             r10              = _mm_mul_pd(rsq10,rinv10);
 943
 944             /* Compute parameters for interactions between i and j atoms */
 945             qq10             = _mm_mul_pd(iq1,jq0);
 946
 947             /* EWALD ELECTROSTATICS */
 948
 949             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 950             ewrt             = _mm_mul_pd(r10,ewtabscale);
 951             ewitab           = _mm_cvttpd_epi32(ewrt);
 952             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 953             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 954                                          &ewtabF,&ewtabFn);
 955             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 956             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 957
 958             fscal            = felec;
 959
 960             /* Calculate temporary vectorial force */
 961             tx               = _mm_mul_pd(fscal,dx10);
 962             ty               = _mm_mul_pd(fscal,dy10);
 963             tz               = _mm_mul_pd(fscal,dz10);
 964
 965             /* Update vectorial force */
 966             fix1             = _mm_add_pd(fix1,tx);
 967             fiy1             = _mm_add_pd(fiy1,ty);
 968             fiz1             = _mm_add_pd(fiz1,tz);
 969
 970             fjx0             = _mm_add_pd(fjx0,tx);
 971             fjy0             = _mm_add_pd(fjy0,ty);
 972             fjz0             = _mm_add_pd(fjz0,tz);
 973
 974             /**************************
 975              * CALCULATE INTERACTIONS *
 976              **************************/
 977
 978             r20              = _mm_mul_pd(rsq20,rinv20);
 979
 980             /* Compute parameters for interactions between i and j atoms */
 981             qq20             = _mm_mul_pd(iq2,jq0);
 982
 983             /* EWALD ELECTROSTATICS */
 984
 985             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 986             ewrt             = _mm_mul_pd(r20,ewtabscale);
 987             ewitab           = _mm_cvttpd_epi32(ewrt);
 988             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 989             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 990                                          &ewtabF,&ewtabFn);
 991             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 992             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 993
 994             fscal            = felec;
 995
 996             /* Calculate temporary vectorial force */
 997             tx               = _mm_mul_pd(fscal,dx20);
 998             ty               = _mm_mul_pd(fscal,dy20);
 999             tz               = _mm_mul_pd(fscal,dz20);
1000
1001             /* Update vectorial force */
1002             fix2             = _mm_add_pd(fix2,tx);
1003             fiy2             = _mm_add_pd(fiy2,ty);
1004             fiz2             = _mm_add_pd(fiz2,tz);
1005
1006             fjx0             = _mm_add_pd(fjx0,tx);
1007             fjy0             = _mm_add_pd(fjy0,ty);
1008             fjz0             = _mm_add_pd(fjz0,tz);
1009
1010             /**************************
1011              * CALCULATE INTERACTIONS *
1012              **************************/
1013
1014             r30              = _mm_mul_pd(rsq30,rinv30);
1015
1016             /* Compute parameters for interactions between i and j atoms */
1017             qq30             = _mm_mul_pd(iq3,jq0);
1018
1019             /* EWALD ELECTROSTATICS */
1020
1021             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1022             ewrt             = _mm_mul_pd(r30,ewtabscale);
1023             ewitab           = _mm_cvttpd_epi32(ewrt);
1024             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1025             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1026                                          &ewtabF,&ewtabFn);
1027             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1028             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1029
1030             fscal            = felec;
1031
1032             /* Calculate temporary vectorial force */
1033             tx               = _mm_mul_pd(fscal,dx30);
1034             ty               = _mm_mul_pd(fscal,dy30);
1035             tz               = _mm_mul_pd(fscal,dz30);
1036
1037             /* Update vectorial force */
1038             fix3             = _mm_add_pd(fix3,tx);
1039             fiy3             = _mm_add_pd(fiy3,ty);
1040             fiz3             = _mm_add_pd(fiz3,tz);
1041
1042             fjx0             = _mm_add_pd(fjx0,tx);
1043             fjy0             = _mm_add_pd(fjy0,ty);
1044             fjz0             = _mm_add_pd(fjz0,tz);
1045
1046             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1047
1048             /* Inner loop uses 157 flops */
1049         }
1050
1051         if(jidx<j_index_end)
1052         {
1053
1054             jnrA             = jjnr[jidx];
1055             j_coord_offsetA  = DIM*jnrA;
1056
1057             /* load j atom coordinates */
1058             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1059                                               &jx0,&jy0,&jz0);
1060
1061             /* Calculate displacement vector */
1062             dx00             = _mm_sub_pd(ix0,jx0);
1063             dy00             = _mm_sub_pd(iy0,jy0);
1064             dz00             = _mm_sub_pd(iz0,jz0);
1065             dx10             = _mm_sub_pd(ix1,jx0);
1066             dy10             = _mm_sub_pd(iy1,jy0);
1067             dz10             = _mm_sub_pd(iz1,jz0);
1068             dx20             = _mm_sub_pd(ix2,jx0);
1069             dy20             = _mm_sub_pd(iy2,jy0);
1070             dz20             = _mm_sub_pd(iz2,jz0);
1071             dx30             = _mm_sub_pd(ix3,jx0);
1072             dy30             = _mm_sub_pd(iy3,jy0);
1073             dz30             = _mm_sub_pd(iz3,jz0);
1074
1075             /* Calculate squared distance and things based on it */
1076             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1077             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1078             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1079             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1080
1081             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1082             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1083             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1084             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1085
1086             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1087             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1088             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1089             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1090
1091             /* Load parameters for j particles */
1092             jq0              = _mm_load_sd(charge+jnrA+0);
1093             vdwjidx0A        = 2*vdwtype[jnrA+0];
1094
1095             fjx0             = _mm_setzero_pd();
1096             fjy0             = _mm_setzero_pd();
1097             fjz0             = _mm_setzero_pd();
1098
1099             /**************************
1100              * CALCULATE INTERACTIONS *
1101              **************************/
1102
1103             r00              = _mm_mul_pd(rsq00,rinv00);
1104
1105             /* Compute parameters for interactions between i and j atoms */
1106             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1107
1108             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
1109
1110             /* Analytical LJ-PME */
1111             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1112             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1113             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1114             exponent         = gmx_simd_exp_d(ewcljrsq);
1115             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1116             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1117             /* f6A = 6 * C6grid * (1 - poly) */
1118             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1119             /* f6B = C6grid * exponent * beta^6 */
1120             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1121             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1122             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);
1123
1124             fscal            = fvdw;
1125
1126             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1127
1128             /* Calculate temporary vectorial force */
1129             tx               = _mm_mul_pd(fscal,dx00);
1130             ty               = _mm_mul_pd(fscal,dy00);
1131             tz               = _mm_mul_pd(fscal,dz00);
1132
1133             /* Update vectorial force */
1134             fix0             = _mm_add_pd(fix0,tx);
1135             fiy0             = _mm_add_pd(fiy0,ty);
1136             fiz0             = _mm_add_pd(fiz0,tz);
1137
1138             fjx0             = _mm_add_pd(fjx0,tx);
1139             fjy0             = _mm_add_pd(fjy0,ty);
1140             fjz0             = _mm_add_pd(fjz0,tz);
1141
1142             /**************************
1143              * CALCULATE INTERACTIONS *
1144              **************************/
1145
1146             r10              = _mm_mul_pd(rsq10,rinv10);
1147
1148             /* Compute parameters for interactions between i and j atoms */
1149             qq10             = _mm_mul_pd(iq1,jq0);
1150
1151             /* EWALD ELECTROSTATICS */
1152
1153             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1154             ewrt             = _mm_mul_pd(r10,ewtabscale);
1155             ewitab           = _mm_cvttpd_epi32(ewrt);
1156             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1157             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1158             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1159             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1160
1161             fscal            = felec;
1162
1163             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1164
1165             /* Calculate temporary vectorial force */
1166             tx               = _mm_mul_pd(fscal,dx10);
1167             ty               = _mm_mul_pd(fscal,dy10);
1168             tz               = _mm_mul_pd(fscal,dz10);
1169
1170             /* Update vectorial force */
1171             fix1             = _mm_add_pd(fix1,tx);
1172             fiy1             = _mm_add_pd(fiy1,ty);
1173             fiz1             = _mm_add_pd(fiz1,tz);
1174
1175             fjx0             = _mm_add_pd(fjx0,tx);
1176             fjy0             = _mm_add_pd(fjy0,ty);
1177             fjz0             = _mm_add_pd(fjz0,tz);
1178
1179             /**************************
1180              * CALCULATE INTERACTIONS *
1181              **************************/
1182
1183             r20              = _mm_mul_pd(rsq20,rinv20);
1184
1185             /* Compute parameters for interactions between i and j atoms */
1186             qq20             = _mm_mul_pd(iq2,jq0);
1187
1188             /* EWALD ELECTROSTATICS */
1189
1190             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1191             ewrt             = _mm_mul_pd(r20,ewtabscale);
1192             ewitab           = _mm_cvttpd_epi32(ewrt);
1193             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1194             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1195             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1196             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1197
1198             fscal            = felec;
1199
1200             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1201
1202             /* Calculate temporary vectorial force */
1203             tx               = _mm_mul_pd(fscal,dx20);
1204             ty               = _mm_mul_pd(fscal,dy20);
1205             tz               = _mm_mul_pd(fscal,dz20);
1206
1207             /* Update vectorial force */
1208             fix2             = _mm_add_pd(fix2,tx);
1209             fiy2             = _mm_add_pd(fiy2,ty);
1210             fiz2             = _mm_add_pd(fiz2,tz);
1211
1212             fjx0             = _mm_add_pd(fjx0,tx);
1213             fjy0             = _mm_add_pd(fjy0,ty);
1214             fjz0             = _mm_add_pd(fjz0,tz);
1215
1216             /**************************
1217              * CALCULATE INTERACTIONS *
1218              **************************/
1219
1220             r30              = _mm_mul_pd(rsq30,rinv30);
1221
1222             /* Compute parameters for interactions between i and j atoms */
1223             qq30             = _mm_mul_pd(iq3,jq0);
1224
1225             /* EWALD ELECTROSTATICS */
1226
1227             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1228             ewrt             = _mm_mul_pd(r30,ewtabscale);
1229             ewitab           = _mm_cvttpd_epi32(ewrt);
1230             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1231             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1232             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1233             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1234
1235             fscal            = felec;
1236
1237             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1238
1239             /* Calculate temporary vectorial force */
1240             tx               = _mm_mul_pd(fscal,dx30);
1241             ty               = _mm_mul_pd(fscal,dy30);
1242             tz               = _mm_mul_pd(fscal,dz30);
1243
1244             /* Update vectorial force */
1245             fix3             = _mm_add_pd(fix3,tx);
1246             fiy3             = _mm_add_pd(fiy3,ty);
1247             fiz3             = _mm_add_pd(fiz3,tz);
1248
1249             fjx0             = _mm_add_pd(fjx0,tx);
1250             fjy0             = _mm_add_pd(fjy0,ty);
1251             fjz0             = _mm_add_pd(fjz0,tz);
1252
1253             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1254
1255             /* Inner loop uses 157 flops */
1256         }
1257
1258         /* End of innermost loop */
1259
1260         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1261                                               f+i_coord_offset,fshift+i_shift_offset);
1262
1263         /* Increment number of inner iterations */
1264         inneriter                  += j_index_end - j_index_start;
1265
1266         /* Outer loop uses 24 flops */
1267     }
1268
1269     /* Increment number of outer iterations */
1270     outeriter        += nri;
1271
1272     /* Update outer/inner flops */
1273
1274     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*157);
1275 }