src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwCSTab_GeomW4P1_avx_128_fma_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 avx_128_fma_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_avx_128_fma_double.h"
  48 #include "kernelutil_x86_avx_128_fma_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4P1_VF_avx_128_fma_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_VF_avx_128_fma_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     __m128i          vfitab;
 103     __m128i          ifour       = _mm_set1_epi32(4);
 104     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 105     real             *vftab;
 106     __m128i          ewitab;
 107     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 108     real             *ewtab;
 109     __m128d          dummy_mask,cutoff_mask;
 110     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 111     __m128d          one     = _mm_set1_pd(1.0);
 112     __m128d          two     = _mm_set1_pd(2.0);
 113     x                = xx[0];
 114     f                = ff[0];
 115
 116     nri              = nlist->nri;
 117     iinr             = nlist->iinr;
 118     jindex           = nlist->jindex;
 119     jjnr             = nlist->jjnr;
 120     shiftidx         = nlist->shift;
 121     gid              = nlist->gid;
 122     shiftvec         = fr->shift_vec[0];
 123     fshift           = fr->fshift[0];
 124     facel            = _mm_set1_pd(fr->epsfac);
 125     charge           = mdatoms->chargeA;
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     vftab            = kernel_data->table_vdw->data;
 131     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 132
 133     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 134     ewtab            = fr->ic->tabq_coul_FDV0;
 135     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 136     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 137
 138     /* Setup water-specific parameters */
 139     inr              = nlist->iinr[0];
 140     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 141     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 142     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 143     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 144
 145     /* Avoid stupid compiler warnings */
 146     jnrA = jnrB = 0;
 147     j_coord_offsetA = 0;
 148     j_coord_offsetB = 0;
 149
 150     outeriter        = 0;
 151     inneriter        = 0;
 152
 153     /* Start outer loop over neighborlists */
 154     for(iidx=0; iidx<nri; iidx++)
 155     {
 156         /* Load shift vector for this list */
 157         i_shift_offset   = DIM*shiftidx[iidx];
 158
 159         /* Load limits for loop over neighbors */
 160         j_index_start    = jindex[iidx];
 161         j_index_end      = jindex[iidx+1];
 162
 163         /* Get outer coordinate index */
 164         inr              = iinr[iidx];
 165         i_coord_offset   = DIM*inr;
 166
 167         /* Load i particle coords and add shift vector */
 168         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 169                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 170
 171         fix0             = _mm_setzero_pd();
 172         fiy0             = _mm_setzero_pd();
 173         fiz0             = _mm_setzero_pd();
 174         fix1             = _mm_setzero_pd();
 175         fiy1             = _mm_setzero_pd();
 176         fiz1             = _mm_setzero_pd();
 177         fix2             = _mm_setzero_pd();
 178         fiy2             = _mm_setzero_pd();
 179         fiz2             = _mm_setzero_pd();
 180         fix3             = _mm_setzero_pd();
 181         fiy3             = _mm_setzero_pd();
 182         fiz3             = _mm_setzero_pd();
 183
 184         /* Reset potential sums */
 185         velecsum         = _mm_setzero_pd();
 186         vvdwsum          = _mm_setzero_pd();
 187
 188         /* Start inner kernel loop */
 189         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 190         {
 191
 192             /* Get j neighbor index, and coordinate index */
 193             jnrA             = jjnr[jidx];
 194             jnrB             = jjnr[jidx+1];
 195             j_coord_offsetA  = DIM*jnrA;
 196             j_coord_offsetB  = DIM*jnrB;
 197
 198             /* load j atom coordinates */
 199             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 200                                               &jx0,&jy0,&jz0);
 201
 202             /* Calculate displacement vector */
 203             dx00             = _mm_sub_pd(ix0,jx0);
 204             dy00             = _mm_sub_pd(iy0,jy0);
 205             dz00             = _mm_sub_pd(iz0,jz0);
 206             dx10             = _mm_sub_pd(ix1,jx0);
 207             dy10             = _mm_sub_pd(iy1,jy0);
 208             dz10             = _mm_sub_pd(iz1,jz0);
 209             dx20             = _mm_sub_pd(ix2,jx0);
 210             dy20             = _mm_sub_pd(iy2,jy0);
 211             dz20             = _mm_sub_pd(iz2,jz0);
 212             dx30             = _mm_sub_pd(ix3,jx0);
 213             dy30             = _mm_sub_pd(iy3,jy0);
 214             dz30             = _mm_sub_pd(iz3,jz0);
 215
 216             /* Calculate squared distance and things based on it */
 217             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 218             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 219             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 220             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 221
 222             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 223             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 224             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 225             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 226
 227             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 228             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 229             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 230
 231             /* Load parameters for j particles */
 232             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 233             vdwjidx0A        = 2*vdwtype[jnrA+0];
 234             vdwjidx0B        = 2*vdwtype[jnrB+0];
 235
 236             fjx0             = _mm_setzero_pd();
 237             fjy0             = _mm_setzero_pd();
 238             fjz0             = _mm_setzero_pd();
 239
 240             /**************************
 241              * CALCULATE INTERACTIONS *
 242              **************************/
 243
 244             r00              = _mm_mul_pd(rsq00,rinv00);
 245
 246             /* Compute parameters for interactions between i and j atoms */
 247             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 248                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 249
 250             /* Calculate table index by multiplying r with table scale and truncate to integer */
 251             rt               = _mm_mul_pd(r00,vftabscale);
 252             vfitab           = _mm_cvttpd_epi32(rt);
 253 #ifdef __XOP__
 254             vfeps            = _mm_frcz_pd(rt);
 255 #else
 256             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 257 #endif
 258             twovfeps         = _mm_add_pd(vfeps,vfeps);
 259             vfitab           = _mm_slli_epi32(vfitab,3);
 260
 261             /* CUBIC SPLINE TABLE DISPERSION */
 262             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 263             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 264             GMX_MM_TRANSPOSE2_PD(Y,F);
 265             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 266             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 267             GMX_MM_TRANSPOSE2_PD(G,H);
 268             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 269             VV               = _mm_macc_pd(vfeps,Fp,Y);
 270             vvdw6            = _mm_mul_pd(c6_00,VV);
 271             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 272             fvdw6            = _mm_mul_pd(c6_00,FF);
 273
 274             /* CUBIC SPLINE TABLE REPULSION */
 275             vfitab           = _mm_add_epi32(vfitab,ifour);
 276             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 277             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 278             GMX_MM_TRANSPOSE2_PD(Y,F);
 279             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 280             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 281             GMX_MM_TRANSPOSE2_PD(G,H);
 282             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 283             VV               = _mm_macc_pd(vfeps,Fp,Y);
 284             vvdw12           = _mm_mul_pd(c12_00,VV);
 285             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 286             fvdw12           = _mm_mul_pd(c12_00,FF);
 287             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 288             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 289
 290             /* Update potential sum for this i atom from the interaction with this j atom. */
 291             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 292
 293             fscal            = fvdw;
 294
 295             /* Update vectorial force */
 296             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 297             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 298             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 299
 300             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 301             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 302             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 303
 304             /**************************
 305              * CALCULATE INTERACTIONS *
 306              **************************/
 307
 308             r10              = _mm_mul_pd(rsq10,rinv10);
 309
 310             /* Compute parameters for interactions between i and j atoms */
 311             qq10             = _mm_mul_pd(iq1,jq0);
 312
 313             /* EWALD ELECTROSTATICS */
 314
 315             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 316             ewrt             = _mm_mul_pd(r10,ewtabscale);
 317             ewitab           = _mm_cvttpd_epi32(ewrt);
 318 #ifdef __XOP__
 319             eweps            = _mm_frcz_pd(ewrt);
 320 #else
 321             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 322 #endif
 323             twoeweps         = _mm_add_pd(eweps,eweps);
 324             ewitab           = _mm_slli_epi32(ewitab,2);
 325             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 326             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 327             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 328             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 329             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 330             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 331             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 332             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 333             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 334             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 335
 336             /* Update potential sum for this i atom from the interaction with this j atom. */
 337             velecsum         = _mm_add_pd(velecsum,velec);
 338
 339             fscal            = felec;
 340
 341             /* Update vectorial force */
 342             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 343             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 344             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 345
 346             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 347             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 348             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 349
 350             /**************************
 351              * CALCULATE INTERACTIONS *
 352              **************************/
 353
 354             r20              = _mm_mul_pd(rsq20,rinv20);
 355
 356             /* Compute parameters for interactions between i and j atoms */
 357             qq20             = _mm_mul_pd(iq2,jq0);
 358
 359             /* EWALD ELECTROSTATICS */
 360
 361             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 362             ewrt             = _mm_mul_pd(r20,ewtabscale);
 363             ewitab           = _mm_cvttpd_epi32(ewrt);
 364 #ifdef __XOP__
 365             eweps            = _mm_frcz_pd(ewrt);
 366 #else
 367             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 368 #endif
 369             twoeweps         = _mm_add_pd(eweps,eweps);
 370             ewitab           = _mm_slli_epi32(ewitab,2);
 371             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 372             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 373             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 374             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 375             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 376             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 377             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 378             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 379             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 380             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 381
 382             /* Update potential sum for this i atom from the interaction with this j atom. */
 383             velecsum         = _mm_add_pd(velecsum,velec);
 384
 385             fscal            = felec;
 386
 387             /* Update vectorial force */
 388             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 389             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 390             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 391
 392             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 393             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 394             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 395
 396             /**************************
 397              * CALCULATE INTERACTIONS *
 398              **************************/
 399
 400             r30              = _mm_mul_pd(rsq30,rinv30);
 401
 402             /* Compute parameters for interactions between i and j atoms */
 403             qq30             = _mm_mul_pd(iq3,jq0);
 404
 405             /* EWALD ELECTROSTATICS */
 406
 407             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 408             ewrt             = _mm_mul_pd(r30,ewtabscale);
 409             ewitab           = _mm_cvttpd_epi32(ewrt);
 410 #ifdef __XOP__
 411             eweps            = _mm_frcz_pd(ewrt);
 412 #else
 413             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 414 #endif
 415             twoeweps         = _mm_add_pd(eweps,eweps);
 416             ewitab           = _mm_slli_epi32(ewitab,2);
 417             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 418             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 419             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 420             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 421             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 422             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 423             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 424             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 425             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 426             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 427
 428             /* Update potential sum for this i atom from the interaction with this j atom. */
 429             velecsum         = _mm_add_pd(velecsum,velec);
 430
 431             fscal            = felec;
 432
 433             /* Update vectorial force */
 434             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 435             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 436             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 437
 438             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 439             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 440             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 441
 442             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 443
 444             /* Inner loop uses 194 flops */
 445         }
 446
 447         if(jidx<j_index_end)
 448         {
 449
 450             jnrA             = jjnr[jidx];
 451             j_coord_offsetA  = DIM*jnrA;
 452
 453             /* load j atom coordinates */
 454             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 455                                               &jx0,&jy0,&jz0);
 456
 457             /* Calculate displacement vector */
 458             dx00             = _mm_sub_pd(ix0,jx0);
 459             dy00             = _mm_sub_pd(iy0,jy0);
 460             dz00             = _mm_sub_pd(iz0,jz0);
 461             dx10             = _mm_sub_pd(ix1,jx0);
 462             dy10             = _mm_sub_pd(iy1,jy0);
 463             dz10             = _mm_sub_pd(iz1,jz0);
 464             dx20             = _mm_sub_pd(ix2,jx0);
 465             dy20             = _mm_sub_pd(iy2,jy0);
 466             dz20             = _mm_sub_pd(iz2,jz0);
 467             dx30             = _mm_sub_pd(ix3,jx0);
 468             dy30             = _mm_sub_pd(iy3,jy0);
 469             dz30             = _mm_sub_pd(iz3,jz0);
 470
 471             /* Calculate squared distance and things based on it */
 472             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 473             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 474             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 475             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 476
 477             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 478             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 479             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 480             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 481
 482             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 483             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 484             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 485
 486             /* Load parameters for j particles */
 487             jq0              = _mm_load_sd(charge+jnrA+0);
 488             vdwjidx0A        = 2*vdwtype[jnrA+0];
 489
 490             fjx0             = _mm_setzero_pd();
 491             fjy0             = _mm_setzero_pd();
 492             fjz0             = _mm_setzero_pd();
 493
 494             /**************************
 495              * CALCULATE INTERACTIONS *
 496              **************************/
 497
 498             r00              = _mm_mul_pd(rsq00,rinv00);
 499
 500             /* Compute parameters for interactions between i and j atoms */
 501             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 502
 503             /* Calculate table index by multiplying r with table scale and truncate to integer */
 504             rt               = _mm_mul_pd(r00,vftabscale);
 505             vfitab           = _mm_cvttpd_epi32(rt);
 506 #ifdef __XOP__
 507             vfeps            = _mm_frcz_pd(rt);
 508 #else
 509             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 510 #endif
 511             twovfeps         = _mm_add_pd(vfeps,vfeps);
 512             vfitab           = _mm_slli_epi32(vfitab,3);
 513
 514             /* CUBIC SPLINE TABLE DISPERSION */
 515             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 516             F                = _mm_setzero_pd();
 517             GMX_MM_TRANSPOSE2_PD(Y,F);
 518             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 519             H                = _mm_setzero_pd();
 520             GMX_MM_TRANSPOSE2_PD(G,H);
 521             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 522             VV               = _mm_macc_pd(vfeps,Fp,Y);
 523             vvdw6            = _mm_mul_pd(c6_00,VV);
 524             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 525             fvdw6            = _mm_mul_pd(c6_00,FF);
 526
 527             /* CUBIC SPLINE TABLE REPULSION */
 528             vfitab           = _mm_add_epi32(vfitab,ifour);
 529             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 530             F                = _mm_setzero_pd();
 531             GMX_MM_TRANSPOSE2_PD(Y,F);
 532             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 533             H                = _mm_setzero_pd();
 534             GMX_MM_TRANSPOSE2_PD(G,H);
 535             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 536             VV               = _mm_macc_pd(vfeps,Fp,Y);
 537             vvdw12           = _mm_mul_pd(c12_00,VV);
 538             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 539             fvdw12           = _mm_mul_pd(c12_00,FF);
 540             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 541             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 542
 543             /* Update potential sum for this i atom from the interaction with this j atom. */
 544             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 545             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 546
 547             fscal            = fvdw;
 548
 549             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 550
 551             /* Update vectorial force */
 552             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 553             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 554             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 555
 556             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 557             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 558             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 559
 560             /**************************
 561              * CALCULATE INTERACTIONS *
 562              **************************/
 563
 564             r10              = _mm_mul_pd(rsq10,rinv10);
 565
 566             /* Compute parameters for interactions between i and j atoms */
 567             qq10             = _mm_mul_pd(iq1,jq0);
 568
 569             /* EWALD ELECTROSTATICS */
 570
 571             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 572             ewrt             = _mm_mul_pd(r10,ewtabscale);
 573             ewitab           = _mm_cvttpd_epi32(ewrt);
 574 #ifdef __XOP__
 575             eweps            = _mm_frcz_pd(ewrt);
 576 #else
 577             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 578 #endif
 579             twoeweps         = _mm_add_pd(eweps,eweps);
 580             ewitab           = _mm_slli_epi32(ewitab,2);
 581             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 582             ewtabD           = _mm_setzero_pd();
 583             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 584             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 585             ewtabFn          = _mm_setzero_pd();
 586             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 587             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 588             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 589             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 590             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 591
 592             /* Update potential sum for this i atom from the interaction with this j atom. */
 593             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 594             velecsum         = _mm_add_pd(velecsum,velec);
 595
 596             fscal            = felec;
 597
 598             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 599
 600             /* Update vectorial force */
 601             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 602             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 603             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 604
 605             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 606             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 607             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 608
 609             /**************************
 610              * CALCULATE INTERACTIONS *
 611              **************************/
 612
 613             r20              = _mm_mul_pd(rsq20,rinv20);
 614
 615             /* Compute parameters for interactions between i and j atoms */
 616             qq20             = _mm_mul_pd(iq2,jq0);
 617
 618             /* EWALD ELECTROSTATICS */
 619
 620             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 621             ewrt             = _mm_mul_pd(r20,ewtabscale);
 622             ewitab           = _mm_cvttpd_epi32(ewrt);
 623 #ifdef __XOP__
 624             eweps            = _mm_frcz_pd(ewrt);
 625 #else
 626             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 627 #endif
 628             twoeweps         = _mm_add_pd(eweps,eweps);
 629             ewitab           = _mm_slli_epi32(ewitab,2);
 630             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 631             ewtabD           = _mm_setzero_pd();
 632             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 633             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 634             ewtabFn          = _mm_setzero_pd();
 635             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 636             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 637             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 638             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 639             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 640
 641             /* Update potential sum for this i atom from the interaction with this j atom. */
 642             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 643             velecsum         = _mm_add_pd(velecsum,velec);
 644
 645             fscal            = felec;
 646
 647             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 648
 649             /* Update vectorial force */
 650             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 651             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 652             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 653
 654             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 655             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 656             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 657
 658             /**************************
 659              * CALCULATE INTERACTIONS *
 660              **************************/
 661
 662             r30              = _mm_mul_pd(rsq30,rinv30);
 663
 664             /* Compute parameters for interactions between i and j atoms */
 665             qq30             = _mm_mul_pd(iq3,jq0);
 666
 667             /* EWALD ELECTROSTATICS */
 668
 669             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 670             ewrt             = _mm_mul_pd(r30,ewtabscale);
 671             ewitab           = _mm_cvttpd_epi32(ewrt);
 672 #ifdef __XOP__
 673             eweps            = _mm_frcz_pd(ewrt);
 674 #else
 675             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 676 #endif
 677             twoeweps         = _mm_add_pd(eweps,eweps);
 678             ewitab           = _mm_slli_epi32(ewitab,2);
 679             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 680             ewtabD           = _mm_setzero_pd();
 681             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 682             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 683             ewtabFn          = _mm_setzero_pd();
 684             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 685             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 686             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 687             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 688             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 689
 690             /* Update potential sum for this i atom from the interaction with this j atom. */
 691             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 692             velecsum         = _mm_add_pd(velecsum,velec);
 693
 694             fscal            = felec;
 695
 696             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 697
 698             /* Update vectorial force */
 699             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 700             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 701             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 702
 703             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 704             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 705             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 706
 707             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 708
 709             /* Inner loop uses 194 flops */
 710         }
 711
 712         /* End of innermost loop */
 713
 714         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 715                                               f+i_coord_offset,fshift+i_shift_offset);
 716
 717         ggid                        = gid[iidx];
 718         /* Update potential energies */
 719         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 720         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 721
 722         /* Increment number of inner iterations */
 723         inneriter                  += j_index_end - j_index_start;
 724
 725         /* Outer loop uses 26 flops */
 726     }
 727
 728     /* Increment number of outer iterations */
 729     outeriter        += nri;
 730
 731     /* Update outer/inner flops */
 732
 733     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*194);
 734 }
 735 /*
 736  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_avx_128_fma_double
 737  * Electrostatics interaction: Ewald
 738  * VdW interaction:            CubicSplineTable
 739  * Geometry:                   Water4-Particle
 740  * Calculate force/pot:        Force
 741  */
 742 void
 743 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_avx_128_fma_double
 744                     (t_nblist                    * gmx_restrict       nlist,
 745                      rvec                        * gmx_restrict          xx,
 746                      rvec                        * gmx_restrict          ff,
 747                      t_forcerec                  * gmx_restrict          fr,
 748                      t_mdatoms                   * gmx_restrict     mdatoms,
 749                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 750                      t_nrnb                      * gmx_restrict        nrnb)
 751 {
 752     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 753      * just 0 for non-waters.
 754      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 755      * jnr indices corresponding to data put in the four positions in the SIMD register.
 756      */
 757     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 758     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 759     int              jnrA,jnrB;
 760     int              j_coord_offsetA,j_coord_offsetB;
 761     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 762     real             rcutoff_scalar;
 763     real             *shiftvec,*fshift,*x,*f;
 764     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 765     int              vdwioffset0;
 766     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 767     int              vdwioffset1;
 768     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 769     int              vdwioffset2;
 770     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 771     int              vdwioffset3;
 772     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 773     int              vdwjidx0A,vdwjidx0B;
 774     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 775     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 776     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 777     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 778     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 779     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 780     real             *charge;
 781     int              nvdwtype;
 782     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 783     int              *vdwtype;
 784     real             *vdwparam;
 785     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 786     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 787     __m128i          vfitab;
 788     __m128i          ifour       = _mm_set1_epi32(4);
 789     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 790     real             *vftab;
 791     __m128i          ewitab;
 792     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 793     real             *ewtab;
 794     __m128d          dummy_mask,cutoff_mask;
 795     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 796     __m128d          one     = _mm_set1_pd(1.0);
 797     __m128d          two     = _mm_set1_pd(2.0);
 798     x                = xx[0];
 799     f                = ff[0];
 800
 801     nri              = nlist->nri;
 802     iinr             = nlist->iinr;
 803     jindex           = nlist->jindex;
 804     jjnr             = nlist->jjnr;
 805     shiftidx         = nlist->shift;
 806     gid              = nlist->gid;
 807     shiftvec         = fr->shift_vec[0];
 808     fshift           = fr->fshift[0];
 809     facel            = _mm_set1_pd(fr->epsfac);
 810     charge           = mdatoms->chargeA;
 811     nvdwtype         = fr->ntype;
 812     vdwparam         = fr->nbfp;
 813     vdwtype          = mdatoms->typeA;
 814
 815     vftab            = kernel_data->table_vdw->data;
 816     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 817
 818     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 819     ewtab            = fr->ic->tabq_coul_F;
 820     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 821     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 822
 823     /* Setup water-specific parameters */
 824     inr              = nlist->iinr[0];
 825     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 826     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 827     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 828     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 829
 830     /* Avoid stupid compiler warnings */
 831     jnrA = jnrB = 0;
 832     j_coord_offsetA = 0;
 833     j_coord_offsetB = 0;
 834
 835     outeriter        = 0;
 836     inneriter        = 0;
 837
 838     /* Start outer loop over neighborlists */
 839     for(iidx=0; iidx<nri; iidx++)
 840     {
 841         /* Load shift vector for this list */
 842         i_shift_offset   = DIM*shiftidx[iidx];
 843
 844         /* Load limits for loop over neighbors */
 845         j_index_start    = jindex[iidx];
 846         j_index_end      = jindex[iidx+1];
 847
 848         /* Get outer coordinate index */
 849         inr              = iinr[iidx];
 850         i_coord_offset   = DIM*inr;
 851
 852         /* Load i particle coords and add shift vector */
 853         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 854                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 855
 856         fix0             = _mm_setzero_pd();
 857         fiy0             = _mm_setzero_pd();
 858         fiz0             = _mm_setzero_pd();
 859         fix1             = _mm_setzero_pd();
 860         fiy1             = _mm_setzero_pd();
 861         fiz1             = _mm_setzero_pd();
 862         fix2             = _mm_setzero_pd();
 863         fiy2             = _mm_setzero_pd();
 864         fiz2             = _mm_setzero_pd();
 865         fix3             = _mm_setzero_pd();
 866         fiy3             = _mm_setzero_pd();
 867         fiz3             = _mm_setzero_pd();
 868
 869         /* Start inner kernel loop */
 870         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 871         {
 872
 873             /* Get j neighbor index, and coordinate index */
 874             jnrA             = jjnr[jidx];
 875             jnrB             = jjnr[jidx+1];
 876             j_coord_offsetA  = DIM*jnrA;
 877             j_coord_offsetB  = DIM*jnrB;
 878
 879             /* load j atom coordinates */
 880             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 881                                               &jx0,&jy0,&jz0);
 882
 883             /* Calculate displacement vector */
 884             dx00             = _mm_sub_pd(ix0,jx0);
 885             dy00             = _mm_sub_pd(iy0,jy0);
 886             dz00             = _mm_sub_pd(iz0,jz0);
 887             dx10             = _mm_sub_pd(ix1,jx0);
 888             dy10             = _mm_sub_pd(iy1,jy0);
 889             dz10             = _mm_sub_pd(iz1,jz0);
 890             dx20             = _mm_sub_pd(ix2,jx0);
 891             dy20             = _mm_sub_pd(iy2,jy0);
 892             dz20             = _mm_sub_pd(iz2,jz0);
 893             dx30             = _mm_sub_pd(ix3,jx0);
 894             dy30             = _mm_sub_pd(iy3,jy0);
 895             dz30             = _mm_sub_pd(iz3,jz0);
 896
 897             /* Calculate squared distance and things based on it */
 898             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 899             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 900             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 901             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 902
 903             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 904             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 905             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 906             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 907
 908             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 909             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 910             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 911
 912             /* Load parameters for j particles */
 913             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 914             vdwjidx0A        = 2*vdwtype[jnrA+0];
 915             vdwjidx0B        = 2*vdwtype[jnrB+0];
 916
 917             fjx0             = _mm_setzero_pd();
 918             fjy0             = _mm_setzero_pd();
 919             fjz0             = _mm_setzero_pd();
 920
 921             /**************************
 922              * CALCULATE INTERACTIONS *
 923              **************************/
 924
 925             r00              = _mm_mul_pd(rsq00,rinv00);
 926
 927             /* Compute parameters for interactions between i and j atoms */
 928             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 929                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 930
 931             /* Calculate table index by multiplying r with table scale and truncate to integer */
 932             rt               = _mm_mul_pd(r00,vftabscale);
 933             vfitab           = _mm_cvttpd_epi32(rt);
 934 #ifdef __XOP__
 935             vfeps            = _mm_frcz_pd(rt);
 936 #else
 937             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 938 #endif
 939             twovfeps         = _mm_add_pd(vfeps,vfeps);
 940             vfitab           = _mm_slli_epi32(vfitab,3);
 941
 942             /* CUBIC SPLINE TABLE DISPERSION */
 943             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 944             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 945             GMX_MM_TRANSPOSE2_PD(Y,F);
 946             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 947             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 948             GMX_MM_TRANSPOSE2_PD(G,H);
 949             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 950             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 951             fvdw6            = _mm_mul_pd(c6_00,FF);
 952
 953             /* CUBIC SPLINE TABLE REPULSION */
 954             vfitab           = _mm_add_epi32(vfitab,ifour);
 955             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 956             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 957             GMX_MM_TRANSPOSE2_PD(Y,F);
 958             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 959             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 960             GMX_MM_TRANSPOSE2_PD(G,H);
 961             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 962             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 963             fvdw12           = _mm_mul_pd(c12_00,FF);
 964             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 965
 966             fscal            = fvdw;
 967
 968             /* Update vectorial force */
 969             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 970             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 971             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 972
 973             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 974             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 975             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 976
 977             /**************************
 978              * CALCULATE INTERACTIONS *
 979              **************************/
 980
 981             r10              = _mm_mul_pd(rsq10,rinv10);
 982
 983             /* Compute parameters for interactions between i and j atoms */
 984             qq10             = _mm_mul_pd(iq1,jq0);
 985
 986             /* EWALD ELECTROSTATICS */
 987
 988             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 989             ewrt             = _mm_mul_pd(r10,ewtabscale);
 990             ewitab           = _mm_cvttpd_epi32(ewrt);
 991 #ifdef __XOP__
 992             eweps            = _mm_frcz_pd(ewrt);
 993 #else
 994             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 995 #endif
 996             twoeweps         = _mm_add_pd(eweps,eweps);
 997             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 998                                          &ewtabF,&ewtabFn);
 999             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1000             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1001
1002             fscal            = felec;
1003
1004             /* Update vectorial force */
1005             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1006             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1007             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1008
1009             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1010             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1011             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1012
1013             /**************************
1014              * CALCULATE INTERACTIONS *
1015              **************************/
1016
1017             r20              = _mm_mul_pd(rsq20,rinv20);
1018
1019             /* Compute parameters for interactions between i and j atoms */
1020             qq20             = _mm_mul_pd(iq2,jq0);
1021
1022             /* EWALD ELECTROSTATICS */
1023
1024             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1025             ewrt             = _mm_mul_pd(r20,ewtabscale);
1026             ewitab           = _mm_cvttpd_epi32(ewrt);
1027 #ifdef __XOP__
1028             eweps            = _mm_frcz_pd(ewrt);
1029 #else
1030             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1031 #endif
1032             twoeweps         = _mm_add_pd(eweps,eweps);
1033             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1034                                          &ewtabF,&ewtabFn);
1035             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1036             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1037
1038             fscal            = felec;
1039
1040             /* Update vectorial force */
1041             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1042             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1043             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1044
1045             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1046             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1047             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1048
1049             /**************************
1050              * CALCULATE INTERACTIONS *
1051              **************************/
1052
1053             r30              = _mm_mul_pd(rsq30,rinv30);
1054
1055             /* Compute parameters for interactions between i and j atoms */
1056             qq30             = _mm_mul_pd(iq3,jq0);
1057
1058             /* EWALD ELECTROSTATICS */
1059
1060             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1061             ewrt             = _mm_mul_pd(r30,ewtabscale);
1062             ewitab           = _mm_cvttpd_epi32(ewrt);
1063 #ifdef __XOP__
1064             eweps            = _mm_frcz_pd(ewrt);
1065 #else
1066             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1067 #endif
1068             twoeweps         = _mm_add_pd(eweps,eweps);
1069             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1070                                          &ewtabF,&ewtabFn);
1071             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1072             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1073
1074             fscal            = felec;
1075
1076             /* Update vectorial force */
1077             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1078             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1079             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1080
1081             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1082             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1083             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1084
1085             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1086
1087             /* Inner loop uses 171 flops */
1088         }
1089
1090         if(jidx<j_index_end)
1091         {
1092
1093             jnrA             = jjnr[jidx];
1094             j_coord_offsetA  = DIM*jnrA;
1095
1096             /* load j atom coordinates */
1097             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1098                                               &jx0,&jy0,&jz0);
1099
1100             /* Calculate displacement vector */
1101             dx00             = _mm_sub_pd(ix0,jx0);
1102             dy00             = _mm_sub_pd(iy0,jy0);
1103             dz00             = _mm_sub_pd(iz0,jz0);
1104             dx10             = _mm_sub_pd(ix1,jx0);
1105             dy10             = _mm_sub_pd(iy1,jy0);
1106             dz10             = _mm_sub_pd(iz1,jz0);
1107             dx20             = _mm_sub_pd(ix2,jx0);
1108             dy20             = _mm_sub_pd(iy2,jy0);
1109             dz20             = _mm_sub_pd(iz2,jz0);
1110             dx30             = _mm_sub_pd(ix3,jx0);
1111             dy30             = _mm_sub_pd(iy3,jy0);
1112             dz30             = _mm_sub_pd(iz3,jz0);
1113
1114             /* Calculate squared distance and things based on it */
1115             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1116             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1117             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1118             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1119
1120             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1121             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1122             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1123             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1124
1125             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1126             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1127             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1128
1129             /* Load parameters for j particles */
1130             jq0              = _mm_load_sd(charge+jnrA+0);
1131             vdwjidx0A        = 2*vdwtype[jnrA+0];
1132
1133             fjx0             = _mm_setzero_pd();
1134             fjy0             = _mm_setzero_pd();
1135             fjz0             = _mm_setzero_pd();
1136
1137             /**************************
1138              * CALCULATE INTERACTIONS *
1139              **************************/
1140
1141             r00              = _mm_mul_pd(rsq00,rinv00);
1142
1143             /* Compute parameters for interactions between i and j atoms */
1144             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1145
1146             /* Calculate table index by multiplying r with table scale and truncate to integer */
1147             rt               = _mm_mul_pd(r00,vftabscale);
1148             vfitab           = _mm_cvttpd_epi32(rt);
1149 #ifdef __XOP__
1150             vfeps            = _mm_frcz_pd(rt);
1151 #else
1152             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1153 #endif
1154             twovfeps         = _mm_add_pd(vfeps,vfeps);
1155             vfitab           = _mm_slli_epi32(vfitab,3);
1156
1157             /* CUBIC SPLINE TABLE DISPERSION */
1158             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1159             F                = _mm_setzero_pd();
1160             GMX_MM_TRANSPOSE2_PD(Y,F);
1161             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1162             H                = _mm_setzero_pd();
1163             GMX_MM_TRANSPOSE2_PD(G,H);
1164             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1165             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1166             fvdw6            = _mm_mul_pd(c6_00,FF);
1167
1168             /* CUBIC SPLINE TABLE REPULSION */
1169             vfitab           = _mm_add_epi32(vfitab,ifour);
1170             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1171             F                = _mm_setzero_pd();
1172             GMX_MM_TRANSPOSE2_PD(Y,F);
1173             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1174             H                = _mm_setzero_pd();
1175             GMX_MM_TRANSPOSE2_PD(G,H);
1176             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1177             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1178             fvdw12           = _mm_mul_pd(c12_00,FF);
1179             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1180
1181             fscal            = fvdw;
1182
1183             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1184
1185             /* Update vectorial force */
1186             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1187             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1188             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1189
1190             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1191             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1192             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1193
1194             /**************************
1195              * CALCULATE INTERACTIONS *
1196              **************************/
1197
1198             r10              = _mm_mul_pd(rsq10,rinv10);
1199
1200             /* Compute parameters for interactions between i and j atoms */
1201             qq10             = _mm_mul_pd(iq1,jq0);
1202
1203             /* EWALD ELECTROSTATICS */
1204
1205             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1206             ewrt             = _mm_mul_pd(r10,ewtabscale);
1207             ewitab           = _mm_cvttpd_epi32(ewrt);
1208 #ifdef __XOP__
1209             eweps            = _mm_frcz_pd(ewrt);
1210 #else
1211             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1212 #endif
1213             twoeweps         = _mm_add_pd(eweps,eweps);
1214             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1215             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1216             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1217
1218             fscal            = felec;
1219
1220             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1221
1222             /* Update vectorial force */
1223             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1224             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1225             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1226
1227             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1228             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1229             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1230
1231             /**************************
1232              * CALCULATE INTERACTIONS *
1233              **************************/
1234
1235             r20              = _mm_mul_pd(rsq20,rinv20);
1236
1237             /* Compute parameters for interactions between i and j atoms */
1238             qq20             = _mm_mul_pd(iq2,jq0);
1239
1240             /* EWALD ELECTROSTATICS */
1241
1242             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1243             ewrt             = _mm_mul_pd(r20,ewtabscale);
1244             ewitab           = _mm_cvttpd_epi32(ewrt);
1245 #ifdef __XOP__
1246             eweps            = _mm_frcz_pd(ewrt);
1247 #else
1248             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1249 #endif
1250             twoeweps         = _mm_add_pd(eweps,eweps);
1251             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1252             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1253             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1254
1255             fscal            = felec;
1256
1257             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1258
1259             /* Update vectorial force */
1260             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1261             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1262             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1263
1264             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1265             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1266             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1267
1268             /**************************
1269              * CALCULATE INTERACTIONS *
1270              **************************/
1271
1272             r30              = _mm_mul_pd(rsq30,rinv30);
1273
1274             /* Compute parameters for interactions between i and j atoms */
1275             qq30             = _mm_mul_pd(iq3,jq0);
1276
1277             /* EWALD ELECTROSTATICS */
1278
1279             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1280             ewrt             = _mm_mul_pd(r30,ewtabscale);
1281             ewitab           = _mm_cvttpd_epi32(ewrt);
1282 #ifdef __XOP__
1283             eweps            = _mm_frcz_pd(ewrt);
1284 #else
1285             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1286 #endif
1287             twoeweps         = _mm_add_pd(eweps,eweps);
1288             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1289             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1290             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1291
1292             fscal            = felec;
1293
1294             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1295
1296             /* Update vectorial force */
1297             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1298             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1299             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1300
1301             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1302             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1303             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1304
1305             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1306
1307             /* Inner loop uses 171 flops */
1308         }
1309
1310         /* End of innermost loop */
1311
1312         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1313                                               f+i_coord_offset,fshift+i_shift_offset);
1314
1315         /* Increment number of inner iterations */
1316         inneriter                  += j_index_end - j_index_start;
1317
1318         /* Outer loop uses 24 flops */
1319     }
1320
1321     /* Increment number of outer iterations */
1322     outeriter        += nri;
1323
1324     /* Update outer/inner flops */
1325
1326     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*171);
1327 }