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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_VF_avx_128_fma_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwioffset3;
  89     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B;
  91     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 103     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 104     __m128i          vfitab;
 105     __m128i          ifour       = _mm_set1_epi32(4);
 106     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 107     real             *vftab;
 108     __m128i          ewitab;
 109     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 110     real             *ewtab;
 111     __m128d          dummy_mask,cutoff_mask;
 112     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 113     __m128d          one     = _mm_set1_pd(1.0);
 114     __m128d          two     = _mm_set1_pd(2.0);
 115     x                = xx[0];
 116     f                = ff[0];
 117
 118     nri              = nlist->nri;
 119     iinr             = nlist->iinr;
 120     jindex           = nlist->jindex;
 121     jjnr             = nlist->jjnr;
 122     shiftidx         = nlist->shift;
 123     gid              = nlist->gid;
 124     shiftvec         = fr->shift_vec[0];
 125     fshift           = fr->fshift[0];
 126     facel            = _mm_set1_pd(fr->epsfac);
 127     charge           = mdatoms->chargeA;
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131
 132     vftab            = kernel_data->table_vdw->data;
 133     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 134
 135     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 136     ewtab            = fr->ic->tabq_coul_FDV0;
 137     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 138     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 139
 140     /* Setup water-specific parameters */
 141     inr              = nlist->iinr[0];
 142     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 143     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 144     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 145     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 146
 147     /* Avoid stupid compiler warnings */
 148     jnrA = jnrB = 0;
 149     j_coord_offsetA = 0;
 150     j_coord_offsetB = 0;
 151
 152     outeriter        = 0;
 153     inneriter        = 0;
 154
 155     /* Start outer loop over neighborlists */
 156     for(iidx=0; iidx<nri; iidx++)
 157     {
 158         /* Load shift vector for this list */
 159         i_shift_offset   = DIM*shiftidx[iidx];
 160
 161         /* Load limits for loop over neighbors */
 162         j_index_start    = jindex[iidx];
 163         j_index_end      = jindex[iidx+1];
 164
 165         /* Get outer coordinate index */
 166         inr              = iinr[iidx];
 167         i_coord_offset   = DIM*inr;
 168
 169         /* Load i particle coords and add shift vector */
 170         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 171                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 172
 173         fix0             = _mm_setzero_pd();
 174         fiy0             = _mm_setzero_pd();
 175         fiz0             = _mm_setzero_pd();
 176         fix1             = _mm_setzero_pd();
 177         fiy1             = _mm_setzero_pd();
 178         fiz1             = _mm_setzero_pd();
 179         fix2             = _mm_setzero_pd();
 180         fiy2             = _mm_setzero_pd();
 181         fiz2             = _mm_setzero_pd();
 182         fix3             = _mm_setzero_pd();
 183         fiy3             = _mm_setzero_pd();
 184         fiz3             = _mm_setzero_pd();
 185
 186         /* Reset potential sums */
 187         velecsum         = _mm_setzero_pd();
 188         vvdwsum          = _mm_setzero_pd();
 189
 190         /* Start inner kernel loop */
 191         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 192         {
 193
 194             /* Get j neighbor index, and coordinate index */
 195             jnrA             = jjnr[jidx];
 196             jnrB             = jjnr[jidx+1];
 197             j_coord_offsetA  = DIM*jnrA;
 198             j_coord_offsetB  = DIM*jnrB;
 199
 200             /* load j atom coordinates */
 201             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 202                                               &jx0,&jy0,&jz0);
 203
 204             /* Calculate displacement vector */
 205             dx00             = _mm_sub_pd(ix0,jx0);
 206             dy00             = _mm_sub_pd(iy0,jy0);
 207             dz00             = _mm_sub_pd(iz0,jz0);
 208             dx10             = _mm_sub_pd(ix1,jx0);
 209             dy10             = _mm_sub_pd(iy1,jy0);
 210             dz10             = _mm_sub_pd(iz1,jz0);
 211             dx20             = _mm_sub_pd(ix2,jx0);
 212             dy20             = _mm_sub_pd(iy2,jy0);
 213             dz20             = _mm_sub_pd(iz2,jz0);
 214             dx30             = _mm_sub_pd(ix3,jx0);
 215             dy30             = _mm_sub_pd(iy3,jy0);
 216             dz30             = _mm_sub_pd(iz3,jz0);
 217
 218             /* Calculate squared distance and things based on it */
 219             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 220             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 221             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 222             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 223
 224             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 225             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 226             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 227             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 228
 229             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 230             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 231             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 232
 233             /* Load parameters for j particles */
 234             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 235             vdwjidx0A        = 2*vdwtype[jnrA+0];
 236             vdwjidx0B        = 2*vdwtype[jnrB+0];
 237
 238             fjx0             = _mm_setzero_pd();
 239             fjy0             = _mm_setzero_pd();
 240             fjz0             = _mm_setzero_pd();
 241
 242             /**************************
 243              * CALCULATE INTERACTIONS *
 244              **************************/
 245
 246             r00              = _mm_mul_pd(rsq00,rinv00);
 247
 248             /* Compute parameters for interactions between i and j atoms */
 249             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 250                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 251
 252             /* Calculate table index by multiplying r with table scale and truncate to integer */
 253             rt               = _mm_mul_pd(r00,vftabscale);
 254             vfitab           = _mm_cvttpd_epi32(rt);
 255 #ifdef __XOP__
 256             vfeps            = _mm_frcz_pd(rt);
 257 #else
 258             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 259 #endif
 260             twovfeps         = _mm_add_pd(vfeps,vfeps);
 261             vfitab           = _mm_slli_epi32(vfitab,3);
 262
 263             /* CUBIC SPLINE TABLE DISPERSION */
 264             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 265             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 266             GMX_MM_TRANSPOSE2_PD(Y,F);
 267             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 268             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 269             GMX_MM_TRANSPOSE2_PD(G,H);
 270             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 271             VV               = _mm_macc_pd(vfeps,Fp,Y);
 272             vvdw6            = _mm_mul_pd(c6_00,VV);
 273             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 274             fvdw6            = _mm_mul_pd(c6_00,FF);
 275
 276             /* CUBIC SPLINE TABLE REPULSION */
 277             vfitab           = _mm_add_epi32(vfitab,ifour);
 278             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 279             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 280             GMX_MM_TRANSPOSE2_PD(Y,F);
 281             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 282             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 283             GMX_MM_TRANSPOSE2_PD(G,H);
 284             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 285             VV               = _mm_macc_pd(vfeps,Fp,Y);
 286             vvdw12           = _mm_mul_pd(c12_00,VV);
 287             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 288             fvdw12           = _mm_mul_pd(c12_00,FF);
 289             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 290             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 291
 292             /* Update potential sum for this i atom from the interaction with this j atom. */
 293             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 294
 295             fscal            = fvdw;
 296
 297             /* Update vectorial force */
 298             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 299             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 300             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 301
 302             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 303             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 304             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 305
 306             /**************************
 307              * CALCULATE INTERACTIONS *
 308              **************************/
 309
 310             r10              = _mm_mul_pd(rsq10,rinv10);
 311
 312             /* Compute parameters for interactions between i and j atoms */
 313             qq10             = _mm_mul_pd(iq1,jq0);
 314
 315             /* EWALD ELECTROSTATICS */
 316
 317             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 318             ewrt             = _mm_mul_pd(r10,ewtabscale);
 319             ewitab           = _mm_cvttpd_epi32(ewrt);
 320 #ifdef __XOP__
 321             eweps            = _mm_frcz_pd(ewrt);
 322 #else
 323             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 324 #endif
 325             twoeweps         = _mm_add_pd(eweps,eweps);
 326             ewitab           = _mm_slli_epi32(ewitab,2);
 327             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 328             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 329             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 330             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 331             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 332             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 333             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 334             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 335             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 336             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 337
 338             /* Update potential sum for this i atom from the interaction with this j atom. */
 339             velecsum         = _mm_add_pd(velecsum,velec);
 340
 341             fscal            = felec;
 342
 343             /* Update vectorial force */
 344             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 345             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 346             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 347
 348             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 349             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 350             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 351
 352             /**************************
 353              * CALCULATE INTERACTIONS *
 354              **************************/
 355
 356             r20              = _mm_mul_pd(rsq20,rinv20);
 357
 358             /* Compute parameters for interactions between i and j atoms */
 359             qq20             = _mm_mul_pd(iq2,jq0);
 360
 361             /* EWALD ELECTROSTATICS */
 362
 363             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 364             ewrt             = _mm_mul_pd(r20,ewtabscale);
 365             ewitab           = _mm_cvttpd_epi32(ewrt);
 366 #ifdef __XOP__
 367             eweps            = _mm_frcz_pd(ewrt);
 368 #else
 369             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 370 #endif
 371             twoeweps         = _mm_add_pd(eweps,eweps);
 372             ewitab           = _mm_slli_epi32(ewitab,2);
 373             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 374             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 375             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 376             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 377             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 378             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 379             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 380             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 381             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 382             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 383
 384             /* Update potential sum for this i atom from the interaction with this j atom. */
 385             velecsum         = _mm_add_pd(velecsum,velec);
 386
 387             fscal            = felec;
 388
 389             /* Update vectorial force */
 390             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 391             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 392             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 393
 394             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 395             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 396             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 397
 398             /**************************
 399              * CALCULATE INTERACTIONS *
 400              **************************/
 401
 402             r30              = _mm_mul_pd(rsq30,rinv30);
 403
 404             /* Compute parameters for interactions between i and j atoms */
 405             qq30             = _mm_mul_pd(iq3,jq0);
 406
 407             /* EWALD ELECTROSTATICS */
 408
 409             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 410             ewrt             = _mm_mul_pd(r30,ewtabscale);
 411             ewitab           = _mm_cvttpd_epi32(ewrt);
 412 #ifdef __XOP__
 413             eweps            = _mm_frcz_pd(ewrt);
 414 #else
 415             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 416 #endif
 417             twoeweps         = _mm_add_pd(eweps,eweps);
 418             ewitab           = _mm_slli_epi32(ewitab,2);
 419             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 420             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 421             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 422             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 423             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 424             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 425             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 426             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 427             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 428             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 429
 430             /* Update potential sum for this i atom from the interaction with this j atom. */
 431             velecsum         = _mm_add_pd(velecsum,velec);
 432
 433             fscal            = felec;
 434
 435             /* Update vectorial force */
 436             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 437             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 438             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 439
 440             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 441             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 442             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 443
 444             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 445
 446             /* Inner loop uses 194 flops */
 447         }
 448
 449         if(jidx<j_index_end)
 450         {
 451
 452             jnrA             = jjnr[jidx];
 453             j_coord_offsetA  = DIM*jnrA;
 454
 455             /* load j atom coordinates */
 456             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 457                                               &jx0,&jy0,&jz0);
 458
 459             /* Calculate displacement vector */
 460             dx00             = _mm_sub_pd(ix0,jx0);
 461             dy00             = _mm_sub_pd(iy0,jy0);
 462             dz00             = _mm_sub_pd(iz0,jz0);
 463             dx10             = _mm_sub_pd(ix1,jx0);
 464             dy10             = _mm_sub_pd(iy1,jy0);
 465             dz10             = _mm_sub_pd(iz1,jz0);
 466             dx20             = _mm_sub_pd(ix2,jx0);
 467             dy20             = _mm_sub_pd(iy2,jy0);
 468             dz20             = _mm_sub_pd(iz2,jz0);
 469             dx30             = _mm_sub_pd(ix3,jx0);
 470             dy30             = _mm_sub_pd(iy3,jy0);
 471             dz30             = _mm_sub_pd(iz3,jz0);
 472
 473             /* Calculate squared distance and things based on it */
 474             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 475             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 476             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 477             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 478
 479             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 480             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 481             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 482             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 483
 484             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 485             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 486             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 487
 488             /* Load parameters for j particles */
 489             jq0              = _mm_load_sd(charge+jnrA+0);
 490             vdwjidx0A        = 2*vdwtype[jnrA+0];
 491
 492             fjx0             = _mm_setzero_pd();
 493             fjy0             = _mm_setzero_pd();
 494             fjz0             = _mm_setzero_pd();
 495
 496             /**************************
 497              * CALCULATE INTERACTIONS *
 498              **************************/
 499
 500             r00              = _mm_mul_pd(rsq00,rinv00);
 501
 502             /* Compute parameters for interactions between i and j atoms */
 503             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 504
 505             /* Calculate table index by multiplying r with table scale and truncate to integer */
 506             rt               = _mm_mul_pd(r00,vftabscale);
 507             vfitab           = _mm_cvttpd_epi32(rt);
 508 #ifdef __XOP__
 509             vfeps            = _mm_frcz_pd(rt);
 510 #else
 511             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 512 #endif
 513             twovfeps         = _mm_add_pd(vfeps,vfeps);
 514             vfitab           = _mm_slli_epi32(vfitab,3);
 515
 516             /* CUBIC SPLINE TABLE DISPERSION */
 517             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 518             F                = _mm_setzero_pd();
 519             GMX_MM_TRANSPOSE2_PD(Y,F);
 520             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 521             H                = _mm_setzero_pd();
 522             GMX_MM_TRANSPOSE2_PD(G,H);
 523             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 524             VV               = _mm_macc_pd(vfeps,Fp,Y);
 525             vvdw6            = _mm_mul_pd(c6_00,VV);
 526             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 527             fvdw6            = _mm_mul_pd(c6_00,FF);
 528
 529             /* CUBIC SPLINE TABLE REPULSION */
 530             vfitab           = _mm_add_epi32(vfitab,ifour);
 531             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 532             F                = _mm_setzero_pd();
 533             GMX_MM_TRANSPOSE2_PD(Y,F);
 534             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 535             H                = _mm_setzero_pd();
 536             GMX_MM_TRANSPOSE2_PD(G,H);
 537             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 538             VV               = _mm_macc_pd(vfeps,Fp,Y);
 539             vvdw12           = _mm_mul_pd(c12_00,VV);
 540             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 541             fvdw12           = _mm_mul_pd(c12_00,FF);
 542             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 543             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 544
 545             /* Update potential sum for this i atom from the interaction with this j atom. */
 546             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 547             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 548
 549             fscal            = fvdw;
 550
 551             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 552
 553             /* Update vectorial force */
 554             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 555             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 556             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 557
 558             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 559             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 560             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 561
 562             /**************************
 563              * CALCULATE INTERACTIONS *
 564              **************************/
 565
 566             r10              = _mm_mul_pd(rsq10,rinv10);
 567
 568             /* Compute parameters for interactions between i and j atoms */
 569             qq10             = _mm_mul_pd(iq1,jq0);
 570
 571             /* EWALD ELECTROSTATICS */
 572
 573             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 574             ewrt             = _mm_mul_pd(r10,ewtabscale);
 575             ewitab           = _mm_cvttpd_epi32(ewrt);
 576 #ifdef __XOP__
 577             eweps            = _mm_frcz_pd(ewrt);
 578 #else
 579             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 580 #endif
 581             twoeweps         = _mm_add_pd(eweps,eweps);
 582             ewitab           = _mm_slli_epi32(ewitab,2);
 583             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 584             ewtabD           = _mm_setzero_pd();
 585             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 586             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 587             ewtabFn          = _mm_setzero_pd();
 588             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 589             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 590             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 591             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 592             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 593
 594             /* Update potential sum for this i atom from the interaction with this j atom. */
 595             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 596             velecsum         = _mm_add_pd(velecsum,velec);
 597
 598             fscal            = felec;
 599
 600             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 601
 602             /* Update vectorial force */
 603             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 604             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 605             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 606
 607             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 608             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 609             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 610
 611             /**************************
 612              * CALCULATE INTERACTIONS *
 613              **************************/
 614
 615             r20              = _mm_mul_pd(rsq20,rinv20);
 616
 617             /* Compute parameters for interactions between i and j atoms */
 618             qq20             = _mm_mul_pd(iq2,jq0);
 619
 620             /* EWALD ELECTROSTATICS */
 621
 622             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 623             ewrt             = _mm_mul_pd(r20,ewtabscale);
 624             ewitab           = _mm_cvttpd_epi32(ewrt);
 625 #ifdef __XOP__
 626             eweps            = _mm_frcz_pd(ewrt);
 627 #else
 628             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 629 #endif
 630             twoeweps         = _mm_add_pd(eweps,eweps);
 631             ewitab           = _mm_slli_epi32(ewitab,2);
 632             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 633             ewtabD           = _mm_setzero_pd();
 634             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 635             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 636             ewtabFn          = _mm_setzero_pd();
 637             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 638             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 639             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 640             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 641             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 642
 643             /* Update potential sum for this i atom from the interaction with this j atom. */
 644             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 645             velecsum         = _mm_add_pd(velecsum,velec);
 646
 647             fscal            = felec;
 648
 649             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 650
 651             /* Update vectorial force */
 652             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 653             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 654             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 655
 656             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 657             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 658             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 659
 660             /**************************
 661              * CALCULATE INTERACTIONS *
 662              **************************/
 663
 664             r30              = _mm_mul_pd(rsq30,rinv30);
 665
 666             /* Compute parameters for interactions between i and j atoms */
 667             qq30             = _mm_mul_pd(iq3,jq0);
 668
 669             /* EWALD ELECTROSTATICS */
 670
 671             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 672             ewrt             = _mm_mul_pd(r30,ewtabscale);
 673             ewitab           = _mm_cvttpd_epi32(ewrt);
 674 #ifdef __XOP__
 675             eweps            = _mm_frcz_pd(ewrt);
 676 #else
 677             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 678 #endif
 679             twoeweps         = _mm_add_pd(eweps,eweps);
 680             ewitab           = _mm_slli_epi32(ewitab,2);
 681             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 682             ewtabD           = _mm_setzero_pd();
 683             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 684             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 685             ewtabFn          = _mm_setzero_pd();
 686             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 687             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 688             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 689             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 690             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 691
 692             /* Update potential sum for this i atom from the interaction with this j atom. */
 693             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 694             velecsum         = _mm_add_pd(velecsum,velec);
 695
 696             fscal            = felec;
 697
 698             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 699
 700             /* Update vectorial force */
 701             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 702             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 703             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 704
 705             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 706             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 707             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 708
 709             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 710
 711             /* Inner loop uses 194 flops */
 712         }
 713
 714         /* End of innermost loop */
 715
 716         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 717                                               f+i_coord_offset,fshift+i_shift_offset);
 718
 719         ggid                        = gid[iidx];
 720         /* Update potential energies */
 721         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 722         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 723
 724         /* Increment number of inner iterations */
 725         inneriter                  += j_index_end - j_index_start;
 726
 727         /* Outer loop uses 26 flops */
 728     }
 729
 730     /* Increment number of outer iterations */
 731     outeriter        += nri;
 732
 733     /* Update outer/inner flops */
 734
 735     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*194);
 736 }
 737 /*
 738  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_avx_128_fma_double
 739  * Electrostatics interaction: Ewald
 740  * VdW interaction:            CubicSplineTable
 741  * Geometry:                   Water4-Particle
 742  * Calculate force/pot:        Force
 743  */
 744 void
 745 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_avx_128_fma_double
 746                     (t_nblist                    * gmx_restrict       nlist,
 747                      rvec                        * gmx_restrict          xx,
 748                      rvec                        * gmx_restrict          ff,
 749                      t_forcerec                  * gmx_restrict          fr,
 750                      t_mdatoms                   * gmx_restrict     mdatoms,
 751                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 752                      t_nrnb                      * gmx_restrict        nrnb)
 753 {
 754     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 755      * just 0 for non-waters.
 756      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 757      * jnr indices corresponding to data put in the four positions in the SIMD register.
 758      */
 759     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 760     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 761     int              jnrA,jnrB;
 762     int              j_coord_offsetA,j_coord_offsetB;
 763     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 764     real             rcutoff_scalar;
 765     real             *shiftvec,*fshift,*x,*f;
 766     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 767     int              vdwioffset0;
 768     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 769     int              vdwioffset1;
 770     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 771     int              vdwioffset2;
 772     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 773     int              vdwioffset3;
 774     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 775     int              vdwjidx0A,vdwjidx0B;
 776     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 777     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 778     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 779     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 780     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 781     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 782     real             *charge;
 783     int              nvdwtype;
 784     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 785     int              *vdwtype;
 786     real             *vdwparam;
 787     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 788     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 789     __m128i          vfitab;
 790     __m128i          ifour       = _mm_set1_epi32(4);
 791     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 792     real             *vftab;
 793     __m128i          ewitab;
 794     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 795     real             *ewtab;
 796     __m128d          dummy_mask,cutoff_mask;
 797     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 798     __m128d          one     = _mm_set1_pd(1.0);
 799     __m128d          two     = _mm_set1_pd(2.0);
 800     x                = xx[0];
 801     f                = ff[0];
 802
 803     nri              = nlist->nri;
 804     iinr             = nlist->iinr;
 805     jindex           = nlist->jindex;
 806     jjnr             = nlist->jjnr;
 807     shiftidx         = nlist->shift;
 808     gid              = nlist->gid;
 809     shiftvec         = fr->shift_vec[0];
 810     fshift           = fr->fshift[0];
 811     facel            = _mm_set1_pd(fr->epsfac);
 812     charge           = mdatoms->chargeA;
 813     nvdwtype         = fr->ntype;
 814     vdwparam         = fr->nbfp;
 815     vdwtype          = mdatoms->typeA;
 816
 817     vftab            = kernel_data->table_vdw->data;
 818     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 819
 820     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 821     ewtab            = fr->ic->tabq_coul_F;
 822     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 823     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 824
 825     /* Setup water-specific parameters */
 826     inr              = nlist->iinr[0];
 827     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 828     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 829     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 830     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 831
 832     /* Avoid stupid compiler warnings */
 833     jnrA = jnrB = 0;
 834     j_coord_offsetA = 0;
 835     j_coord_offsetB = 0;
 836
 837     outeriter        = 0;
 838     inneriter        = 0;
 839
 840     /* Start outer loop over neighborlists */
 841     for(iidx=0; iidx<nri; iidx++)
 842     {
 843         /* Load shift vector for this list */
 844         i_shift_offset   = DIM*shiftidx[iidx];
 845
 846         /* Load limits for loop over neighbors */
 847         j_index_start    = jindex[iidx];
 848         j_index_end      = jindex[iidx+1];
 849
 850         /* Get outer coordinate index */
 851         inr              = iinr[iidx];
 852         i_coord_offset   = DIM*inr;
 853
 854         /* Load i particle coords and add shift vector */
 855         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 856                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 857
 858         fix0             = _mm_setzero_pd();
 859         fiy0             = _mm_setzero_pd();
 860         fiz0             = _mm_setzero_pd();
 861         fix1             = _mm_setzero_pd();
 862         fiy1             = _mm_setzero_pd();
 863         fiz1             = _mm_setzero_pd();
 864         fix2             = _mm_setzero_pd();
 865         fiy2             = _mm_setzero_pd();
 866         fiz2             = _mm_setzero_pd();
 867         fix3             = _mm_setzero_pd();
 868         fiy3             = _mm_setzero_pd();
 869         fiz3             = _mm_setzero_pd();
 870
 871         /* Start inner kernel loop */
 872         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 873         {
 874
 875             /* Get j neighbor index, and coordinate index */
 876             jnrA             = jjnr[jidx];
 877             jnrB             = jjnr[jidx+1];
 878             j_coord_offsetA  = DIM*jnrA;
 879             j_coord_offsetB  = DIM*jnrB;
 880
 881             /* load j atom coordinates */
 882             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 883                                               &jx0,&jy0,&jz0);
 884
 885             /* Calculate displacement vector */
 886             dx00             = _mm_sub_pd(ix0,jx0);
 887             dy00             = _mm_sub_pd(iy0,jy0);
 888             dz00             = _mm_sub_pd(iz0,jz0);
 889             dx10             = _mm_sub_pd(ix1,jx0);
 890             dy10             = _mm_sub_pd(iy1,jy0);
 891             dz10             = _mm_sub_pd(iz1,jz0);
 892             dx20             = _mm_sub_pd(ix2,jx0);
 893             dy20             = _mm_sub_pd(iy2,jy0);
 894             dz20             = _mm_sub_pd(iz2,jz0);
 895             dx30             = _mm_sub_pd(ix3,jx0);
 896             dy30             = _mm_sub_pd(iy3,jy0);
 897             dz30             = _mm_sub_pd(iz3,jz0);
 898
 899             /* Calculate squared distance and things based on it */
 900             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 901             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 902             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 903             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 904
 905             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 906             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 907             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 908             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 909
 910             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 911             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 912             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 913
 914             /* Load parameters for j particles */
 915             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 916             vdwjidx0A        = 2*vdwtype[jnrA+0];
 917             vdwjidx0B        = 2*vdwtype[jnrB+0];
 918
 919             fjx0             = _mm_setzero_pd();
 920             fjy0             = _mm_setzero_pd();
 921             fjz0             = _mm_setzero_pd();
 922
 923             /**************************
 924              * CALCULATE INTERACTIONS *
 925              **************************/
 926
 927             r00              = _mm_mul_pd(rsq00,rinv00);
 928
 929             /* Compute parameters for interactions between i and j atoms */
 930             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 931                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 932
 933             /* Calculate table index by multiplying r with table scale and truncate to integer */
 934             rt               = _mm_mul_pd(r00,vftabscale);
 935             vfitab           = _mm_cvttpd_epi32(rt);
 936 #ifdef __XOP__
 937             vfeps            = _mm_frcz_pd(rt);
 938 #else
 939             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 940 #endif
 941             twovfeps         = _mm_add_pd(vfeps,vfeps);
 942             vfitab           = _mm_slli_epi32(vfitab,3);
 943
 944             /* CUBIC SPLINE TABLE DISPERSION */
 945             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 946             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 947             GMX_MM_TRANSPOSE2_PD(Y,F);
 948             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 949             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 950             GMX_MM_TRANSPOSE2_PD(G,H);
 951             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 952             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 953             fvdw6            = _mm_mul_pd(c6_00,FF);
 954
 955             /* CUBIC SPLINE TABLE REPULSION */
 956             vfitab           = _mm_add_epi32(vfitab,ifour);
 957             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 958             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 959             GMX_MM_TRANSPOSE2_PD(Y,F);
 960             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 961             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 962             GMX_MM_TRANSPOSE2_PD(G,H);
 963             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 964             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 965             fvdw12           = _mm_mul_pd(c12_00,FF);
 966             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 967
 968             fscal            = fvdw;
 969
 970             /* Update vectorial force */
 971             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 972             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 973             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 974
 975             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 976             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 977             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 978
 979             /**************************
 980              * CALCULATE INTERACTIONS *
 981              **************************/
 982
 983             r10              = _mm_mul_pd(rsq10,rinv10);
 984
 985             /* Compute parameters for interactions between i and j atoms */
 986             qq10             = _mm_mul_pd(iq1,jq0);
 987
 988             /* EWALD ELECTROSTATICS */
 989
 990             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 991             ewrt             = _mm_mul_pd(r10,ewtabscale);
 992             ewitab           = _mm_cvttpd_epi32(ewrt);
 993 #ifdef __XOP__
 994             eweps            = _mm_frcz_pd(ewrt);
 995 #else
 996             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 997 #endif
 998             twoeweps         = _mm_add_pd(eweps,eweps);
 999             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1000                                          &ewtabF,&ewtabFn);
1001             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1002             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1003
1004             fscal            = felec;
1005
1006             /* Update vectorial force */
1007             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1008             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1009             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1010
1011             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1012             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1013             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1014
1015             /**************************
1016              * CALCULATE INTERACTIONS *
1017              **************************/
1018
1019             r20              = _mm_mul_pd(rsq20,rinv20);
1020
1021             /* Compute parameters for interactions between i and j atoms */
1022             qq20             = _mm_mul_pd(iq2,jq0);
1023
1024             /* EWALD ELECTROSTATICS */
1025
1026             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1027             ewrt             = _mm_mul_pd(r20,ewtabscale);
1028             ewitab           = _mm_cvttpd_epi32(ewrt);
1029 #ifdef __XOP__
1030             eweps            = _mm_frcz_pd(ewrt);
1031 #else
1032             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1033 #endif
1034             twoeweps         = _mm_add_pd(eweps,eweps);
1035             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1036                                          &ewtabF,&ewtabFn);
1037             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1038             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1039
1040             fscal            = felec;
1041
1042             /* Update vectorial force */
1043             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1044             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1045             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1046
1047             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1048             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1049             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1050
1051             /**************************
1052              * CALCULATE INTERACTIONS *
1053              **************************/
1054
1055             r30              = _mm_mul_pd(rsq30,rinv30);
1056
1057             /* Compute parameters for interactions between i and j atoms */
1058             qq30             = _mm_mul_pd(iq3,jq0);
1059
1060             /* EWALD ELECTROSTATICS */
1061
1062             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1063             ewrt             = _mm_mul_pd(r30,ewtabscale);
1064             ewitab           = _mm_cvttpd_epi32(ewrt);
1065 #ifdef __XOP__
1066             eweps            = _mm_frcz_pd(ewrt);
1067 #else
1068             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1069 #endif
1070             twoeweps         = _mm_add_pd(eweps,eweps);
1071             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1072                                          &ewtabF,&ewtabFn);
1073             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1074             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1075
1076             fscal            = felec;
1077
1078             /* Update vectorial force */
1079             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1080             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1081             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1082
1083             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1084             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1085             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1086
1087             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1088
1089             /* Inner loop uses 171 flops */
1090         }
1091
1092         if(jidx<j_index_end)
1093         {
1094
1095             jnrA             = jjnr[jidx];
1096             j_coord_offsetA  = DIM*jnrA;
1097
1098             /* load j atom coordinates */
1099             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1100                                               &jx0,&jy0,&jz0);
1101
1102             /* Calculate displacement vector */
1103             dx00             = _mm_sub_pd(ix0,jx0);
1104             dy00             = _mm_sub_pd(iy0,jy0);
1105             dz00             = _mm_sub_pd(iz0,jz0);
1106             dx10             = _mm_sub_pd(ix1,jx0);
1107             dy10             = _mm_sub_pd(iy1,jy0);
1108             dz10             = _mm_sub_pd(iz1,jz0);
1109             dx20             = _mm_sub_pd(ix2,jx0);
1110             dy20             = _mm_sub_pd(iy2,jy0);
1111             dz20             = _mm_sub_pd(iz2,jz0);
1112             dx30             = _mm_sub_pd(ix3,jx0);
1113             dy30             = _mm_sub_pd(iy3,jy0);
1114             dz30             = _mm_sub_pd(iz3,jz0);
1115
1116             /* Calculate squared distance and things based on it */
1117             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1118             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1119             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1120             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1121
1122             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1123             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1124             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1125             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1126
1127             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1128             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1129             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1130
1131             /* Load parameters for j particles */
1132             jq0              = _mm_load_sd(charge+jnrA+0);
1133             vdwjidx0A        = 2*vdwtype[jnrA+0];
1134
1135             fjx0             = _mm_setzero_pd();
1136             fjy0             = _mm_setzero_pd();
1137             fjz0             = _mm_setzero_pd();
1138
1139             /**************************
1140              * CALCULATE INTERACTIONS *
1141              **************************/
1142
1143             r00              = _mm_mul_pd(rsq00,rinv00);
1144
1145             /* Compute parameters for interactions between i and j atoms */
1146             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1147
1148             /* Calculate table index by multiplying r with table scale and truncate to integer */
1149             rt               = _mm_mul_pd(r00,vftabscale);
1150             vfitab           = _mm_cvttpd_epi32(rt);
1151 #ifdef __XOP__
1152             vfeps            = _mm_frcz_pd(rt);
1153 #else
1154             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1155 #endif
1156             twovfeps         = _mm_add_pd(vfeps,vfeps);
1157             vfitab           = _mm_slli_epi32(vfitab,3);
1158
1159             /* CUBIC SPLINE TABLE DISPERSION */
1160             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1161             F                = _mm_setzero_pd();
1162             GMX_MM_TRANSPOSE2_PD(Y,F);
1163             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1164             H                = _mm_setzero_pd();
1165             GMX_MM_TRANSPOSE2_PD(G,H);
1166             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1167             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1168             fvdw6            = _mm_mul_pd(c6_00,FF);
1169
1170             /* CUBIC SPLINE TABLE REPULSION */
1171             vfitab           = _mm_add_epi32(vfitab,ifour);
1172             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1173             F                = _mm_setzero_pd();
1174             GMX_MM_TRANSPOSE2_PD(Y,F);
1175             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1176             H                = _mm_setzero_pd();
1177             GMX_MM_TRANSPOSE2_PD(G,H);
1178             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1179             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1180             fvdw12           = _mm_mul_pd(c12_00,FF);
1181             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1182
1183             fscal            = fvdw;
1184
1185             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1186
1187             /* Update vectorial force */
1188             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1189             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1190             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1191
1192             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1193             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1194             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1195
1196             /**************************
1197              * CALCULATE INTERACTIONS *
1198              **************************/
1199
1200             r10              = _mm_mul_pd(rsq10,rinv10);
1201
1202             /* Compute parameters for interactions between i and j atoms */
1203             qq10             = _mm_mul_pd(iq1,jq0);
1204
1205             /* EWALD ELECTROSTATICS */
1206
1207             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1208             ewrt             = _mm_mul_pd(r10,ewtabscale);
1209             ewitab           = _mm_cvttpd_epi32(ewrt);
1210 #ifdef __XOP__
1211             eweps            = _mm_frcz_pd(ewrt);
1212 #else
1213             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1214 #endif
1215             twoeweps         = _mm_add_pd(eweps,eweps);
1216             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1217             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1218             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1219
1220             fscal            = felec;
1221
1222             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1223
1224             /* Update vectorial force */
1225             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1226             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1227             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1228
1229             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1230             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1231             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1232
1233             /**************************
1234              * CALCULATE INTERACTIONS *
1235              **************************/
1236
1237             r20              = _mm_mul_pd(rsq20,rinv20);
1238
1239             /* Compute parameters for interactions between i and j atoms */
1240             qq20             = _mm_mul_pd(iq2,jq0);
1241
1242             /* EWALD ELECTROSTATICS */
1243
1244             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1245             ewrt             = _mm_mul_pd(r20,ewtabscale);
1246             ewitab           = _mm_cvttpd_epi32(ewrt);
1247 #ifdef __XOP__
1248             eweps            = _mm_frcz_pd(ewrt);
1249 #else
1250             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1251 #endif
1252             twoeweps         = _mm_add_pd(eweps,eweps);
1253             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1254             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1255             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1256
1257             fscal            = felec;
1258
1259             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1260
1261             /* Update vectorial force */
1262             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1263             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1264             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1265
1266             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1267             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1268             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1269
1270             /**************************
1271              * CALCULATE INTERACTIONS *
1272              **************************/
1273
1274             r30              = _mm_mul_pd(rsq30,rinv30);
1275
1276             /* Compute parameters for interactions between i and j atoms */
1277             qq30             = _mm_mul_pd(iq3,jq0);
1278
1279             /* EWALD ELECTROSTATICS */
1280
1281             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1282             ewrt             = _mm_mul_pd(r30,ewtabscale);
1283             ewitab           = _mm_cvttpd_epi32(ewrt);
1284 #ifdef __XOP__
1285             eweps            = _mm_frcz_pd(ewrt);
1286 #else
1287             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1288 #endif
1289             twoeweps         = _mm_add_pd(eweps,eweps);
1290             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1291             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1292             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1293
1294             fscal            = felec;
1295
1296             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1297
1298             /* Update vectorial force */
1299             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1300             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1301             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1302
1303             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1304             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1305             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1306
1307             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1308
1309             /* Inner loop uses 171 flops */
1310         }
1311
1312         /* End of innermost loop */
1313
1314         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1315                                               f+i_coord_offset,fshift+i_shift_offset);
1316
1317         /* Increment number of inner iterations */
1318         inneriter                  += j_index_end - j_index_start;
1319
1320         /* Outer loop uses 24 flops */
1321     }
1322
1323     /* Increment number of outer iterations */
1324     outeriter        += nri;
1325
1326     /* Update outer/inner flops */
1327
1328     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*171);
1329 }