src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecEw_VdwCSTab_GeomW4P1_sse4_1_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse4_1_double.h"
  34 #include "kernelutil_x86_sse4_1_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4P1_VF_sse4_1_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_VF_sse4_1_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB;
  61     int              j_coord_offsetA,j_coord_offsetB;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwioffset1;
  69     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  70     int              vdwioffset2;
  71     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  72     int              vdwioffset3;
  73     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  74     int              vdwjidx0A,vdwjidx0B;
  75     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  76     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  77     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  78     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  79     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  80     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  87     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  88     __m128i          vfitab;
  89     __m128i          ifour       = _mm_set1_epi32(4);
  90     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  91     real             *vftab;
  92     __m128i          ewitab;
  93     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  94     real             *ewtab;
  95     __m128d          dummy_mask,cutoff_mask;
  96     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  97     __m128d          one     = _mm_set1_pd(1.0);
  98     __m128d          two     = _mm_set1_pd(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     facel            = _mm_set1_pd(fr->epsfac);
 111     charge           = mdatoms->chargeA;
 112     nvdwtype         = fr->ntype;
 113     vdwparam         = fr->nbfp;
 114     vdwtype          = mdatoms->typeA;
 115
 116     vftab            = kernel_data->table_vdw->data;
 117     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 118
 119     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 120     ewtab            = fr->ic->tabq_coul_FDV0;
 121     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 122     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 123
 124     /* Setup water-specific parameters */
 125     inr              = nlist->iinr[0];
 126     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 127     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 128     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 129     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 130
 131     /* Avoid stupid compiler warnings */
 132     jnrA = jnrB = 0;
 133     j_coord_offsetA = 0;
 134     j_coord_offsetB = 0;
 135
 136     outeriter        = 0;
 137     inneriter        = 0;
 138
 139     /* Start outer loop over neighborlists */
 140     for(iidx=0; iidx<nri; iidx++)
 141     {
 142         /* Load shift vector for this list */
 143         i_shift_offset   = DIM*shiftidx[iidx];
 144
 145         /* Load limits for loop over neighbors */
 146         j_index_start    = jindex[iidx];
 147         j_index_end      = jindex[iidx+1];
 148
 149         /* Get outer coordinate index */
 150         inr              = iinr[iidx];
 151         i_coord_offset   = DIM*inr;
 152
 153         /* Load i particle coords and add shift vector */
 154         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 155                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 156
 157         fix0             = _mm_setzero_pd();
 158         fiy0             = _mm_setzero_pd();
 159         fiz0             = _mm_setzero_pd();
 160         fix1             = _mm_setzero_pd();
 161         fiy1             = _mm_setzero_pd();
 162         fiz1             = _mm_setzero_pd();
 163         fix2             = _mm_setzero_pd();
 164         fiy2             = _mm_setzero_pd();
 165         fiz2             = _mm_setzero_pd();
 166         fix3             = _mm_setzero_pd();
 167         fiy3             = _mm_setzero_pd();
 168         fiz3             = _mm_setzero_pd();
 169
 170         /* Reset potential sums */
 171         velecsum         = _mm_setzero_pd();
 172         vvdwsum          = _mm_setzero_pd();
 173
 174         /* Start inner kernel loop */
 175         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 176         {
 177
 178             /* Get j neighbor index, and coordinate index */
 179             jnrA             = jjnr[jidx];
 180             jnrB             = jjnr[jidx+1];
 181             j_coord_offsetA  = DIM*jnrA;
 182             j_coord_offsetB  = DIM*jnrB;
 183
 184             /* load j atom coordinates */
 185             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 186                                               &jx0,&jy0,&jz0);
 187
 188             /* Calculate displacement vector */
 189             dx00             = _mm_sub_pd(ix0,jx0);
 190             dy00             = _mm_sub_pd(iy0,jy0);
 191             dz00             = _mm_sub_pd(iz0,jz0);
 192             dx10             = _mm_sub_pd(ix1,jx0);
 193             dy10             = _mm_sub_pd(iy1,jy0);
 194             dz10             = _mm_sub_pd(iz1,jz0);
 195             dx20             = _mm_sub_pd(ix2,jx0);
 196             dy20             = _mm_sub_pd(iy2,jy0);
 197             dz20             = _mm_sub_pd(iz2,jz0);
 198             dx30             = _mm_sub_pd(ix3,jx0);
 199             dy30             = _mm_sub_pd(iy3,jy0);
 200             dz30             = _mm_sub_pd(iz3,jz0);
 201
 202             /* Calculate squared distance and things based on it */
 203             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 204             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 205             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 206             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 207
 208             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 209             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 210             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 211             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 212
 213             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 214             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 215             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 216
 217             /* Load parameters for j particles */
 218             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 219             vdwjidx0A        = 2*vdwtype[jnrA+0];
 220             vdwjidx0B        = 2*vdwtype[jnrB+0];
 221
 222             fjx0             = _mm_setzero_pd();
 223             fjy0             = _mm_setzero_pd();
 224             fjz0             = _mm_setzero_pd();
 225
 226             /**************************
 227              * CALCULATE INTERACTIONS *
 228              **************************/
 229
 230             r00              = _mm_mul_pd(rsq00,rinv00);
 231
 232             /* Compute parameters for interactions between i and j atoms */
 233             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 234                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 235
 236             /* Calculate table index by multiplying r with table scale and truncate to integer */
 237             rt               = _mm_mul_pd(r00,vftabscale);
 238             vfitab           = _mm_cvttpd_epi32(rt);
 239             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 240             vfitab           = _mm_slli_epi32(vfitab,3);
 241
 242             /* CUBIC SPLINE TABLE DISPERSION */
 243             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 244             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 245             GMX_MM_TRANSPOSE2_PD(Y,F);
 246             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 247             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 248             GMX_MM_TRANSPOSE2_PD(G,H);
 249             Heps             = _mm_mul_pd(vfeps,H);
 250             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 251             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 252             vvdw6            = _mm_mul_pd(c6_00,VV);
 253             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 254             fvdw6            = _mm_mul_pd(c6_00,FF);
 255
 256             /* CUBIC SPLINE TABLE REPULSION */
 257             vfitab           = _mm_add_epi32(vfitab,ifour);
 258             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 259             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 260             GMX_MM_TRANSPOSE2_PD(Y,F);
 261             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 262             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 263             GMX_MM_TRANSPOSE2_PD(G,H);
 264             Heps             = _mm_mul_pd(vfeps,H);
 265             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 266             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 267             vvdw12           = _mm_mul_pd(c12_00,VV);
 268             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 269             fvdw12           = _mm_mul_pd(c12_00,FF);
 270             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 271             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 272
 273             /* Update potential sum for this i atom from the interaction with this j atom. */
 274             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 275
 276             fscal            = fvdw;
 277
 278             /* Calculate temporary vectorial force */
 279             tx               = _mm_mul_pd(fscal,dx00);
 280             ty               = _mm_mul_pd(fscal,dy00);
 281             tz               = _mm_mul_pd(fscal,dz00);
 282
 283             /* Update vectorial force */
 284             fix0             = _mm_add_pd(fix0,tx);
 285             fiy0             = _mm_add_pd(fiy0,ty);
 286             fiz0             = _mm_add_pd(fiz0,tz);
 287
 288             fjx0             = _mm_add_pd(fjx0,tx);
 289             fjy0             = _mm_add_pd(fjy0,ty);
 290             fjz0             = _mm_add_pd(fjz0,tz);
 291
 292             /**************************
 293              * CALCULATE INTERACTIONS *
 294              **************************/
 295
 296             r10              = _mm_mul_pd(rsq10,rinv10);
 297
 298             /* Compute parameters for interactions between i and j atoms */
 299             qq10             = _mm_mul_pd(iq1,jq0);
 300
 301             /* EWALD ELECTROSTATICS */
 302
 303             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 304             ewrt             = _mm_mul_pd(r10,ewtabscale);
 305             ewitab           = _mm_cvttpd_epi32(ewrt);
 306             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 307             ewitab           = _mm_slli_epi32(ewitab,2);
 308             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 309             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 310             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 311             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 312             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 313             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 314             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 315             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 316             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 317             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 318
 319             /* Update potential sum for this i atom from the interaction with this j atom. */
 320             velecsum         = _mm_add_pd(velecsum,velec);
 321
 322             fscal            = felec;
 323
 324             /* Calculate temporary vectorial force */
 325             tx               = _mm_mul_pd(fscal,dx10);
 326             ty               = _mm_mul_pd(fscal,dy10);
 327             tz               = _mm_mul_pd(fscal,dz10);
 328
 329             /* Update vectorial force */
 330             fix1             = _mm_add_pd(fix1,tx);
 331             fiy1             = _mm_add_pd(fiy1,ty);
 332             fiz1             = _mm_add_pd(fiz1,tz);
 333
 334             fjx0             = _mm_add_pd(fjx0,tx);
 335             fjy0             = _mm_add_pd(fjy0,ty);
 336             fjz0             = _mm_add_pd(fjz0,tz);
 337
 338             /**************************
 339              * CALCULATE INTERACTIONS *
 340              **************************/
 341
 342             r20              = _mm_mul_pd(rsq20,rinv20);
 343
 344             /* Compute parameters for interactions between i and j atoms */
 345             qq20             = _mm_mul_pd(iq2,jq0);
 346
 347             /* EWALD ELECTROSTATICS */
 348
 349             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 350             ewrt             = _mm_mul_pd(r20,ewtabscale);
 351             ewitab           = _mm_cvttpd_epi32(ewrt);
 352             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 353             ewitab           = _mm_slli_epi32(ewitab,2);
 354             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 355             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 356             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 357             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 358             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 359             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 360             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 361             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 362             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 363             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 364
 365             /* Update potential sum for this i atom from the interaction with this j atom. */
 366             velecsum         = _mm_add_pd(velecsum,velec);
 367
 368             fscal            = felec;
 369
 370             /* Calculate temporary vectorial force */
 371             tx               = _mm_mul_pd(fscal,dx20);
 372             ty               = _mm_mul_pd(fscal,dy20);
 373             tz               = _mm_mul_pd(fscal,dz20);
 374
 375             /* Update vectorial force */
 376             fix2             = _mm_add_pd(fix2,tx);
 377             fiy2             = _mm_add_pd(fiy2,ty);
 378             fiz2             = _mm_add_pd(fiz2,tz);
 379
 380             fjx0             = _mm_add_pd(fjx0,tx);
 381             fjy0             = _mm_add_pd(fjy0,ty);
 382             fjz0             = _mm_add_pd(fjz0,tz);
 383
 384             /**************************
 385              * CALCULATE INTERACTIONS *
 386              **************************/
 387
 388             r30              = _mm_mul_pd(rsq30,rinv30);
 389
 390             /* Compute parameters for interactions between i and j atoms */
 391             qq30             = _mm_mul_pd(iq3,jq0);
 392
 393             /* EWALD ELECTROSTATICS */
 394
 395             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 396             ewrt             = _mm_mul_pd(r30,ewtabscale);
 397             ewitab           = _mm_cvttpd_epi32(ewrt);
 398             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 399             ewitab           = _mm_slli_epi32(ewitab,2);
 400             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 401             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 402             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 403             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 404             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 405             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 406             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 407             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 408             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 409             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 410
 411             /* Update potential sum for this i atom from the interaction with this j atom. */
 412             velecsum         = _mm_add_pd(velecsum,velec);
 413
 414             fscal            = felec;
 415
 416             /* Calculate temporary vectorial force */
 417             tx               = _mm_mul_pd(fscal,dx30);
 418             ty               = _mm_mul_pd(fscal,dy30);
 419             tz               = _mm_mul_pd(fscal,dz30);
 420
 421             /* Update vectorial force */
 422             fix3             = _mm_add_pd(fix3,tx);
 423             fiy3             = _mm_add_pd(fiy3,ty);
 424             fiz3             = _mm_add_pd(fiz3,tz);
 425
 426             fjx0             = _mm_add_pd(fjx0,tx);
 427             fjy0             = _mm_add_pd(fjy0,ty);
 428             fjz0             = _mm_add_pd(fjz0,tz);
 429
 430             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 431
 432             /* Inner loop uses 182 flops */
 433         }
 434
 435         if(jidx<j_index_end)
 436         {
 437
 438             jnrA             = jjnr[jidx];
 439             j_coord_offsetA  = DIM*jnrA;
 440
 441             /* load j atom coordinates */
 442             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 443                                               &jx0,&jy0,&jz0);
 444
 445             /* Calculate displacement vector */
 446             dx00             = _mm_sub_pd(ix0,jx0);
 447             dy00             = _mm_sub_pd(iy0,jy0);
 448             dz00             = _mm_sub_pd(iz0,jz0);
 449             dx10             = _mm_sub_pd(ix1,jx0);
 450             dy10             = _mm_sub_pd(iy1,jy0);
 451             dz10             = _mm_sub_pd(iz1,jz0);
 452             dx20             = _mm_sub_pd(ix2,jx0);
 453             dy20             = _mm_sub_pd(iy2,jy0);
 454             dz20             = _mm_sub_pd(iz2,jz0);
 455             dx30             = _mm_sub_pd(ix3,jx0);
 456             dy30             = _mm_sub_pd(iy3,jy0);
 457             dz30             = _mm_sub_pd(iz3,jz0);
 458
 459             /* Calculate squared distance and things based on it */
 460             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 461             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 462             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 463             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 464
 465             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 466             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 467             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 468             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 469
 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             /* Calculate table index by multiplying r with table scale and truncate to integer */
 492             rt               = _mm_mul_pd(r00,vftabscale);
 493             vfitab           = _mm_cvttpd_epi32(rt);
 494             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 495             vfitab           = _mm_slli_epi32(vfitab,3);
 496
 497             /* CUBIC SPLINE TABLE DISPERSION */
 498             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 499             F                = _mm_setzero_pd();
 500             GMX_MM_TRANSPOSE2_PD(Y,F);
 501             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 502             H                = _mm_setzero_pd();
 503             GMX_MM_TRANSPOSE2_PD(G,H);
 504             Heps             = _mm_mul_pd(vfeps,H);
 505             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 506             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 507             vvdw6            = _mm_mul_pd(c6_00,VV);
 508             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 509             fvdw6            = _mm_mul_pd(c6_00,FF);
 510
 511             /* CUBIC SPLINE TABLE REPULSION */
 512             vfitab           = _mm_add_epi32(vfitab,ifour);
 513             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 514             F                = _mm_setzero_pd();
 515             GMX_MM_TRANSPOSE2_PD(Y,F);
 516             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 517             H                = _mm_setzero_pd();
 518             GMX_MM_TRANSPOSE2_PD(G,H);
 519             Heps             = _mm_mul_pd(vfeps,H);
 520             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 521             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 522             vvdw12           = _mm_mul_pd(c12_00,VV);
 523             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 524             fvdw12           = _mm_mul_pd(c12_00,FF);
 525             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 526             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 527
 528             /* Update potential sum for this i atom from the interaction with this j atom. */
 529             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 530             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 531
 532             fscal            = fvdw;
 533
 534             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 535
 536             /* Calculate temporary vectorial force */
 537             tx               = _mm_mul_pd(fscal,dx00);
 538             ty               = _mm_mul_pd(fscal,dy00);
 539             tz               = _mm_mul_pd(fscal,dz00);
 540
 541             /* Update vectorial force */
 542             fix0             = _mm_add_pd(fix0,tx);
 543             fiy0             = _mm_add_pd(fiy0,ty);
 544             fiz0             = _mm_add_pd(fiz0,tz);
 545
 546             fjx0             = _mm_add_pd(fjx0,tx);
 547             fjy0             = _mm_add_pd(fjy0,ty);
 548             fjz0             = _mm_add_pd(fjz0,tz);
 549
 550             /**************************
 551              * CALCULATE INTERACTIONS *
 552              **************************/
 553
 554             r10              = _mm_mul_pd(rsq10,rinv10);
 555
 556             /* Compute parameters for interactions between i and j atoms */
 557             qq10             = _mm_mul_pd(iq1,jq0);
 558
 559             /* EWALD ELECTROSTATICS */
 560
 561             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 562             ewrt             = _mm_mul_pd(r10,ewtabscale);
 563             ewitab           = _mm_cvttpd_epi32(ewrt);
 564             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 565             ewitab           = _mm_slli_epi32(ewitab,2);
 566             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 567             ewtabD           = _mm_setzero_pd();
 568             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 569             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 570             ewtabFn          = _mm_setzero_pd();
 571             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 572             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 573             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 574             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 575             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 576
 577             /* Update potential sum for this i atom from the interaction with this j atom. */
 578             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 579             velecsum         = _mm_add_pd(velecsum,velec);
 580
 581             fscal            = felec;
 582
 583             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 584
 585             /* Calculate temporary vectorial force */
 586             tx               = _mm_mul_pd(fscal,dx10);
 587             ty               = _mm_mul_pd(fscal,dy10);
 588             tz               = _mm_mul_pd(fscal,dz10);
 589
 590             /* Update vectorial force */
 591             fix1             = _mm_add_pd(fix1,tx);
 592             fiy1             = _mm_add_pd(fiy1,ty);
 593             fiz1             = _mm_add_pd(fiz1,tz);
 594
 595             fjx0             = _mm_add_pd(fjx0,tx);
 596             fjy0             = _mm_add_pd(fjy0,ty);
 597             fjz0             = _mm_add_pd(fjz0,tz);
 598
 599             /**************************
 600              * CALCULATE INTERACTIONS *
 601              **************************/
 602
 603             r20              = _mm_mul_pd(rsq20,rinv20);
 604
 605             /* Compute parameters for interactions between i and j atoms */
 606             qq20             = _mm_mul_pd(iq2,jq0);
 607
 608             /* EWALD ELECTROSTATICS */
 609
 610             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 611             ewrt             = _mm_mul_pd(r20,ewtabscale);
 612             ewitab           = _mm_cvttpd_epi32(ewrt);
 613             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 614             ewitab           = _mm_slli_epi32(ewitab,2);
 615             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 616             ewtabD           = _mm_setzero_pd();
 617             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 618             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 619             ewtabFn          = _mm_setzero_pd();
 620             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 621             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 622             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 623             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 624             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 625
 626             /* Update potential sum for this i atom from the interaction with this j atom. */
 627             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 628             velecsum         = _mm_add_pd(velecsum,velec);
 629
 630             fscal            = felec;
 631
 632             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 633
 634             /* Calculate temporary vectorial force */
 635             tx               = _mm_mul_pd(fscal,dx20);
 636             ty               = _mm_mul_pd(fscal,dy20);
 637             tz               = _mm_mul_pd(fscal,dz20);
 638
 639             /* Update vectorial force */
 640             fix2             = _mm_add_pd(fix2,tx);
 641             fiy2             = _mm_add_pd(fiy2,ty);
 642             fiz2             = _mm_add_pd(fiz2,tz);
 643
 644             fjx0             = _mm_add_pd(fjx0,tx);
 645             fjy0             = _mm_add_pd(fjy0,ty);
 646             fjz0             = _mm_add_pd(fjz0,tz);
 647
 648             /**************************
 649              * CALCULATE INTERACTIONS *
 650              **************************/
 651
 652             r30              = _mm_mul_pd(rsq30,rinv30);
 653
 654             /* Compute parameters for interactions between i and j atoms */
 655             qq30             = _mm_mul_pd(iq3,jq0);
 656
 657             /* EWALD ELECTROSTATICS */
 658
 659             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 660             ewrt             = _mm_mul_pd(r30,ewtabscale);
 661             ewitab           = _mm_cvttpd_epi32(ewrt);
 662             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 663             ewitab           = _mm_slli_epi32(ewitab,2);
 664             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 665             ewtabD           = _mm_setzero_pd();
 666             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 667             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 668             ewtabFn          = _mm_setzero_pd();
 669             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 670             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 671             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 672             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 673             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 674
 675             /* Update potential sum for this i atom from the interaction with this j atom. */
 676             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 677             velecsum         = _mm_add_pd(velecsum,velec);
 678
 679             fscal            = felec;
 680
 681             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 682
 683             /* Calculate temporary vectorial force */
 684             tx               = _mm_mul_pd(fscal,dx30);
 685             ty               = _mm_mul_pd(fscal,dy30);
 686             tz               = _mm_mul_pd(fscal,dz30);
 687
 688             /* Update vectorial force */
 689             fix3             = _mm_add_pd(fix3,tx);
 690             fiy3             = _mm_add_pd(fiy3,ty);
 691             fiz3             = _mm_add_pd(fiz3,tz);
 692
 693             fjx0             = _mm_add_pd(fjx0,tx);
 694             fjy0             = _mm_add_pd(fjy0,ty);
 695             fjz0             = _mm_add_pd(fjz0,tz);
 696
 697             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 698
 699             /* Inner loop uses 182 flops */
 700         }
 701
 702         /* End of innermost loop */
 703
 704         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 705                                               f+i_coord_offset,fshift+i_shift_offset);
 706
 707         ggid                        = gid[iidx];
 708         /* Update potential energies */
 709         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 710         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 711
 712         /* Increment number of inner iterations */
 713         inneriter                  += j_index_end - j_index_start;
 714
 715         /* Outer loop uses 26 flops */
 716     }
 717
 718     /* Increment number of outer iterations */
 719     outeriter        += nri;
 720
 721     /* Update outer/inner flops */
 722
 723     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*182);
 724 }
 725 /*
 726  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_sse4_1_double
 727  * Electrostatics interaction: Ewald
 728  * VdW interaction:            CubicSplineTable
 729  * Geometry:                   Water4-Particle
 730  * Calculate force/pot:        Force
 731  */
 732 void
 733 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_sse4_1_double
 734                     (t_nblist * gmx_restrict                nlist,
 735                      rvec * gmx_restrict                    xx,
 736                      rvec * gmx_restrict                    ff,
 737                      t_forcerec * gmx_restrict              fr,
 738                      t_mdatoms * gmx_restrict               mdatoms,
 739                      nb_kernel_data_t * gmx_restrict        kernel_data,
 740                      t_nrnb * gmx_restrict                  nrnb)
 741 {
 742     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 743      * just 0 for non-waters.
 744      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 745      * jnr indices corresponding to data put in the four positions in the SIMD register.
 746      */
 747     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 748     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 749     int              jnrA,jnrB;
 750     int              j_coord_offsetA,j_coord_offsetB;
 751     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 752     real             rcutoff_scalar;
 753     real             *shiftvec,*fshift,*x,*f;
 754     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 755     int              vdwioffset0;
 756     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 757     int              vdwioffset1;
 758     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 759     int              vdwioffset2;
 760     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 761     int              vdwioffset3;
 762     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 763     int              vdwjidx0A,vdwjidx0B;
 764     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 765     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 766     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 767     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 768     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 769     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 770     real             *charge;
 771     int              nvdwtype;
 772     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 773     int              *vdwtype;
 774     real             *vdwparam;
 775     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 776     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 777     __m128i          vfitab;
 778     __m128i          ifour       = _mm_set1_epi32(4);
 779     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 780     real             *vftab;
 781     __m128i          ewitab;
 782     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 783     real             *ewtab;
 784     __m128d          dummy_mask,cutoff_mask;
 785     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 786     __m128d          one     = _mm_set1_pd(1.0);
 787     __m128d          two     = _mm_set1_pd(2.0);
 788     x                = xx[0];
 789     f                = ff[0];
 790
 791     nri              = nlist->nri;
 792     iinr             = nlist->iinr;
 793     jindex           = nlist->jindex;
 794     jjnr             = nlist->jjnr;
 795     shiftidx         = nlist->shift;
 796     gid              = nlist->gid;
 797     shiftvec         = fr->shift_vec[0];
 798     fshift           = fr->fshift[0];
 799     facel            = _mm_set1_pd(fr->epsfac);
 800     charge           = mdatoms->chargeA;
 801     nvdwtype         = fr->ntype;
 802     vdwparam         = fr->nbfp;
 803     vdwtype          = mdatoms->typeA;
 804
 805     vftab            = kernel_data->table_vdw->data;
 806     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 807
 808     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 809     ewtab            = fr->ic->tabq_coul_F;
 810     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 811     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 812
 813     /* Setup water-specific parameters */
 814     inr              = nlist->iinr[0];
 815     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 816     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 817     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 818     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 819
 820     /* Avoid stupid compiler warnings */
 821     jnrA = jnrB = 0;
 822     j_coord_offsetA = 0;
 823     j_coord_offsetB = 0;
 824
 825     outeriter        = 0;
 826     inneriter        = 0;
 827
 828     /* Start outer loop over neighborlists */
 829     for(iidx=0; iidx<nri; iidx++)
 830     {
 831         /* Load shift vector for this list */
 832         i_shift_offset   = DIM*shiftidx[iidx];
 833
 834         /* Load limits for loop over neighbors */
 835         j_index_start    = jindex[iidx];
 836         j_index_end      = jindex[iidx+1];
 837
 838         /* Get outer coordinate index */
 839         inr              = iinr[iidx];
 840         i_coord_offset   = DIM*inr;
 841
 842         /* Load i particle coords and add shift vector */
 843         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 844                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 845
 846         fix0             = _mm_setzero_pd();
 847         fiy0             = _mm_setzero_pd();
 848         fiz0             = _mm_setzero_pd();
 849         fix1             = _mm_setzero_pd();
 850         fiy1             = _mm_setzero_pd();
 851         fiz1             = _mm_setzero_pd();
 852         fix2             = _mm_setzero_pd();
 853         fiy2             = _mm_setzero_pd();
 854         fiz2             = _mm_setzero_pd();
 855         fix3             = _mm_setzero_pd();
 856         fiy3             = _mm_setzero_pd();
 857         fiz3             = _mm_setzero_pd();
 858
 859         /* Start inner kernel loop */
 860         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 861         {
 862
 863             /* Get j neighbor index, and coordinate index */
 864             jnrA             = jjnr[jidx];
 865             jnrB             = jjnr[jidx+1];
 866             j_coord_offsetA  = DIM*jnrA;
 867             j_coord_offsetB  = DIM*jnrB;
 868
 869             /* load j atom coordinates */
 870             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 871                                               &jx0,&jy0,&jz0);
 872
 873             /* Calculate displacement vector */
 874             dx00             = _mm_sub_pd(ix0,jx0);
 875             dy00             = _mm_sub_pd(iy0,jy0);
 876             dz00             = _mm_sub_pd(iz0,jz0);
 877             dx10             = _mm_sub_pd(ix1,jx0);
 878             dy10             = _mm_sub_pd(iy1,jy0);
 879             dz10             = _mm_sub_pd(iz1,jz0);
 880             dx20             = _mm_sub_pd(ix2,jx0);
 881             dy20             = _mm_sub_pd(iy2,jy0);
 882             dz20             = _mm_sub_pd(iz2,jz0);
 883             dx30             = _mm_sub_pd(ix3,jx0);
 884             dy30             = _mm_sub_pd(iy3,jy0);
 885             dz30             = _mm_sub_pd(iz3,jz0);
 886
 887             /* Calculate squared distance and things based on it */
 888             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 889             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 890             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 891             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 892
 893             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 894             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 895             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 896             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 897
 898             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 899             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 900             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 901
 902             /* Load parameters for j particles */
 903             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 904             vdwjidx0A        = 2*vdwtype[jnrA+0];
 905             vdwjidx0B        = 2*vdwtype[jnrB+0];
 906
 907             fjx0             = _mm_setzero_pd();
 908             fjy0             = _mm_setzero_pd();
 909             fjz0             = _mm_setzero_pd();
 910
 911             /**************************
 912              * CALCULATE INTERACTIONS *
 913              **************************/
 914
 915             r00              = _mm_mul_pd(rsq00,rinv00);
 916
 917             /* Compute parameters for interactions between i and j atoms */
 918             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 919                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 920
 921             /* Calculate table index by multiplying r with table scale and truncate to integer */
 922             rt               = _mm_mul_pd(r00,vftabscale);
 923             vfitab           = _mm_cvttpd_epi32(rt);
 924             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 925             vfitab           = _mm_slli_epi32(vfitab,3);
 926
 927             /* CUBIC SPLINE TABLE DISPERSION */
 928             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 929             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 930             GMX_MM_TRANSPOSE2_PD(Y,F);
 931             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 932             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 933             GMX_MM_TRANSPOSE2_PD(G,H);
 934             Heps             = _mm_mul_pd(vfeps,H);
 935             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 936             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 937             fvdw6            = _mm_mul_pd(c6_00,FF);
 938
 939             /* CUBIC SPLINE TABLE REPULSION */
 940             vfitab           = _mm_add_epi32(vfitab,ifour);
 941             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 942             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 943             GMX_MM_TRANSPOSE2_PD(Y,F);
 944             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 945             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 946             GMX_MM_TRANSPOSE2_PD(G,H);
 947             Heps             = _mm_mul_pd(vfeps,H);
 948             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 949             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 950             fvdw12           = _mm_mul_pd(c12_00,FF);
 951             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 952
 953             fscal            = fvdw;
 954
 955             /* Calculate temporary vectorial force */
 956             tx               = _mm_mul_pd(fscal,dx00);
 957             ty               = _mm_mul_pd(fscal,dy00);
 958             tz               = _mm_mul_pd(fscal,dz00);
 959
 960             /* Update vectorial force */
 961             fix0             = _mm_add_pd(fix0,tx);
 962             fiy0             = _mm_add_pd(fiy0,ty);
 963             fiz0             = _mm_add_pd(fiz0,tz);
 964
 965             fjx0             = _mm_add_pd(fjx0,tx);
 966             fjy0             = _mm_add_pd(fjy0,ty);
 967             fjz0             = _mm_add_pd(fjz0,tz);
 968
 969             /**************************
 970              * CALCULATE INTERACTIONS *
 971              **************************/
 972
 973             r10              = _mm_mul_pd(rsq10,rinv10);
 974
 975             /* Compute parameters for interactions between i and j atoms */
 976             qq10             = _mm_mul_pd(iq1,jq0);
 977
 978             /* EWALD ELECTROSTATICS */
 979
 980             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 981             ewrt             = _mm_mul_pd(r10,ewtabscale);
 982             ewitab           = _mm_cvttpd_epi32(ewrt);
 983             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 984             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 985                                          &ewtabF,&ewtabFn);
 986             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 987             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 988
 989             fscal            = felec;
 990
 991             /* Calculate temporary vectorial force */
 992             tx               = _mm_mul_pd(fscal,dx10);
 993             ty               = _mm_mul_pd(fscal,dy10);
 994             tz               = _mm_mul_pd(fscal,dz10);
 995
 996             /* Update vectorial force */
 997             fix1             = _mm_add_pd(fix1,tx);
 998             fiy1             = _mm_add_pd(fiy1,ty);
 999             fiz1             = _mm_add_pd(fiz1,tz);
1000
1001             fjx0             = _mm_add_pd(fjx0,tx);
1002             fjy0             = _mm_add_pd(fjy0,ty);
1003             fjz0             = _mm_add_pd(fjz0,tz);
1004
1005             /**************************
1006              * CALCULATE INTERACTIONS *
1007              **************************/
1008
1009             r20              = _mm_mul_pd(rsq20,rinv20);
1010
1011             /* Compute parameters for interactions between i and j atoms */
1012             qq20             = _mm_mul_pd(iq2,jq0);
1013
1014             /* EWALD ELECTROSTATICS */
1015
1016             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1017             ewrt             = _mm_mul_pd(r20,ewtabscale);
1018             ewitab           = _mm_cvttpd_epi32(ewrt);
1019             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1020             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1021                                          &ewtabF,&ewtabFn);
1022             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1023             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1024
1025             fscal            = felec;
1026
1027             /* Calculate temporary vectorial force */
1028             tx               = _mm_mul_pd(fscal,dx20);
1029             ty               = _mm_mul_pd(fscal,dy20);
1030             tz               = _mm_mul_pd(fscal,dz20);
1031
1032             /* Update vectorial force */
1033             fix2             = _mm_add_pd(fix2,tx);
1034             fiy2             = _mm_add_pd(fiy2,ty);
1035             fiz2             = _mm_add_pd(fiz2,tz);
1036
1037             fjx0             = _mm_add_pd(fjx0,tx);
1038             fjy0             = _mm_add_pd(fjy0,ty);
1039             fjz0             = _mm_add_pd(fjz0,tz);
1040
1041             /**************************
1042              * CALCULATE INTERACTIONS *
1043              **************************/
1044
1045             r30              = _mm_mul_pd(rsq30,rinv30);
1046
1047             /* Compute parameters for interactions between i and j atoms */
1048             qq30             = _mm_mul_pd(iq3,jq0);
1049
1050             /* EWALD ELECTROSTATICS */
1051
1052             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1053             ewrt             = _mm_mul_pd(r30,ewtabscale);
1054             ewitab           = _mm_cvttpd_epi32(ewrt);
1055             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1056             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1057                                          &ewtabF,&ewtabFn);
1058             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1059             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1060
1061             fscal            = felec;
1062
1063             /* Calculate temporary vectorial force */
1064             tx               = _mm_mul_pd(fscal,dx30);
1065             ty               = _mm_mul_pd(fscal,dy30);
1066             tz               = _mm_mul_pd(fscal,dz30);
1067
1068             /* Update vectorial force */
1069             fix3             = _mm_add_pd(fix3,tx);
1070             fiy3             = _mm_add_pd(fiy3,ty);
1071             fiz3             = _mm_add_pd(fiz3,tz);
1072
1073             fjx0             = _mm_add_pd(fjx0,tx);
1074             fjy0             = _mm_add_pd(fjy0,ty);
1075             fjz0             = _mm_add_pd(fjz0,tz);
1076
1077             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1078
1079             /* Inner loop uses 159 flops */
1080         }
1081
1082         if(jidx<j_index_end)
1083         {
1084
1085             jnrA             = jjnr[jidx];
1086             j_coord_offsetA  = DIM*jnrA;
1087
1088             /* load j atom coordinates */
1089             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1090                                               &jx0,&jy0,&jz0);
1091
1092             /* Calculate displacement vector */
1093             dx00             = _mm_sub_pd(ix0,jx0);
1094             dy00             = _mm_sub_pd(iy0,jy0);
1095             dz00             = _mm_sub_pd(iz0,jz0);
1096             dx10             = _mm_sub_pd(ix1,jx0);
1097             dy10             = _mm_sub_pd(iy1,jy0);
1098             dz10             = _mm_sub_pd(iz1,jz0);
1099             dx20             = _mm_sub_pd(ix2,jx0);
1100             dy20             = _mm_sub_pd(iy2,jy0);
1101             dz20             = _mm_sub_pd(iz2,jz0);
1102             dx30             = _mm_sub_pd(ix3,jx0);
1103             dy30             = _mm_sub_pd(iy3,jy0);
1104             dz30             = _mm_sub_pd(iz3,jz0);
1105
1106             /* Calculate squared distance and things based on it */
1107             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1108             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1109             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1110             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1111
1112             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1113             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1114             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1115             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1116
1117             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1118             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1119             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1120
1121             /* Load parameters for j particles */
1122             jq0              = _mm_load_sd(charge+jnrA+0);
1123             vdwjidx0A        = 2*vdwtype[jnrA+0];
1124
1125             fjx0             = _mm_setzero_pd();
1126             fjy0             = _mm_setzero_pd();
1127             fjz0             = _mm_setzero_pd();
1128
1129             /**************************
1130              * CALCULATE INTERACTIONS *
1131              **************************/
1132
1133             r00              = _mm_mul_pd(rsq00,rinv00);
1134
1135             /* Compute parameters for interactions between i and j atoms */
1136             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1137
1138             /* Calculate table index by multiplying r with table scale and truncate to integer */
1139             rt               = _mm_mul_pd(r00,vftabscale);
1140             vfitab           = _mm_cvttpd_epi32(rt);
1141             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1142             vfitab           = _mm_slli_epi32(vfitab,3);
1143
1144             /* CUBIC SPLINE TABLE DISPERSION */
1145             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1146             F                = _mm_setzero_pd();
1147             GMX_MM_TRANSPOSE2_PD(Y,F);
1148             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1149             H                = _mm_setzero_pd();
1150             GMX_MM_TRANSPOSE2_PD(G,H);
1151             Heps             = _mm_mul_pd(vfeps,H);
1152             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1153             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1154             fvdw6            = _mm_mul_pd(c6_00,FF);
1155
1156             /* CUBIC SPLINE TABLE REPULSION */
1157             vfitab           = _mm_add_epi32(vfitab,ifour);
1158             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1159             F                = _mm_setzero_pd();
1160             GMX_MM_TRANSPOSE2_PD(Y,F);
1161             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1162             H                = _mm_setzero_pd();
1163             GMX_MM_TRANSPOSE2_PD(G,H);
1164             Heps             = _mm_mul_pd(vfeps,H);
1165             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1166             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1167             fvdw12           = _mm_mul_pd(c12_00,FF);
1168             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1169
1170             fscal            = fvdw;
1171
1172             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1173
1174             /* Calculate temporary vectorial force */
1175             tx               = _mm_mul_pd(fscal,dx00);
1176             ty               = _mm_mul_pd(fscal,dy00);
1177             tz               = _mm_mul_pd(fscal,dz00);
1178
1179             /* Update vectorial force */
1180             fix0             = _mm_add_pd(fix0,tx);
1181             fiy0             = _mm_add_pd(fiy0,ty);
1182             fiz0             = _mm_add_pd(fiz0,tz);
1183
1184             fjx0             = _mm_add_pd(fjx0,tx);
1185             fjy0             = _mm_add_pd(fjy0,ty);
1186             fjz0             = _mm_add_pd(fjz0,tz);
1187
1188             /**************************
1189              * CALCULATE INTERACTIONS *
1190              **************************/
1191
1192             r10              = _mm_mul_pd(rsq10,rinv10);
1193
1194             /* Compute parameters for interactions between i and j atoms */
1195             qq10             = _mm_mul_pd(iq1,jq0);
1196
1197             /* EWALD ELECTROSTATICS */
1198
1199             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1200             ewrt             = _mm_mul_pd(r10,ewtabscale);
1201             ewitab           = _mm_cvttpd_epi32(ewrt);
1202             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1203             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1204             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1205             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1206
1207             fscal            = felec;
1208
1209             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1210
1211             /* Calculate temporary vectorial force */
1212             tx               = _mm_mul_pd(fscal,dx10);
1213             ty               = _mm_mul_pd(fscal,dy10);
1214             tz               = _mm_mul_pd(fscal,dz10);
1215
1216             /* Update vectorial force */
1217             fix1             = _mm_add_pd(fix1,tx);
1218             fiy1             = _mm_add_pd(fiy1,ty);
1219             fiz1             = _mm_add_pd(fiz1,tz);
1220
1221             fjx0             = _mm_add_pd(fjx0,tx);
1222             fjy0             = _mm_add_pd(fjy0,ty);
1223             fjz0             = _mm_add_pd(fjz0,tz);
1224
1225             /**************************
1226              * CALCULATE INTERACTIONS *
1227              **************************/
1228
1229             r20              = _mm_mul_pd(rsq20,rinv20);
1230
1231             /* Compute parameters for interactions between i and j atoms */
1232             qq20             = _mm_mul_pd(iq2,jq0);
1233
1234             /* EWALD ELECTROSTATICS */
1235
1236             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1237             ewrt             = _mm_mul_pd(r20,ewtabscale);
1238             ewitab           = _mm_cvttpd_epi32(ewrt);
1239             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1240             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1241             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1242             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1243
1244             fscal            = felec;
1245
1246             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1247
1248             /* Calculate temporary vectorial force */
1249             tx               = _mm_mul_pd(fscal,dx20);
1250             ty               = _mm_mul_pd(fscal,dy20);
1251             tz               = _mm_mul_pd(fscal,dz20);
1252
1253             /* Update vectorial force */
1254             fix2             = _mm_add_pd(fix2,tx);
1255             fiy2             = _mm_add_pd(fiy2,ty);
1256             fiz2             = _mm_add_pd(fiz2,tz);
1257
1258             fjx0             = _mm_add_pd(fjx0,tx);
1259             fjy0             = _mm_add_pd(fjy0,ty);
1260             fjz0             = _mm_add_pd(fjz0,tz);
1261
1262             /**************************
1263              * CALCULATE INTERACTIONS *
1264              **************************/
1265
1266             r30              = _mm_mul_pd(rsq30,rinv30);
1267
1268             /* Compute parameters for interactions between i and j atoms */
1269             qq30             = _mm_mul_pd(iq3,jq0);
1270
1271             /* EWALD ELECTROSTATICS */
1272
1273             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1274             ewrt             = _mm_mul_pd(r30,ewtabscale);
1275             ewitab           = _mm_cvttpd_epi32(ewrt);
1276             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1277             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1278             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1279             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1280
1281             fscal            = felec;
1282
1283             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1284
1285             /* Calculate temporary vectorial force */
1286             tx               = _mm_mul_pd(fscal,dx30);
1287             ty               = _mm_mul_pd(fscal,dy30);
1288             tz               = _mm_mul_pd(fscal,dz30);
1289
1290             /* Update vectorial force */
1291             fix3             = _mm_add_pd(fix3,tx);
1292             fiy3             = _mm_add_pd(fiy3,ty);
1293             fiz3             = _mm_add_pd(fiz3,tz);
1294
1295             fjx0             = _mm_add_pd(fjx0,tx);
1296             fjy0             = _mm_add_pd(fjy0,ty);
1297             fjz0             = _mm_add_pd(fjz0,tz);
1298
1299             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1300
1301             /* Inner loop uses 159 flops */
1302         }
1303
1304         /* End of innermost loop */
1305
1306         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1307                                               f+i_coord_offset,fshift+i_shift_offset);
1308
1309         /* Increment number of inner iterations */
1310         inneriter                  += j_index_end - j_index_start;
1311
1312         /* Outer loop uses 24 flops */
1313     }
1314
1315     /* Increment number of outer iterations */
1316     outeriter        += nri;
1317
1318     /* Update outer/inner flops */
1319
1320     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*159);
1321 }