src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwCSTab_GeomW4P1_avx_128_fma_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_128_fma_double.h"
  34 #include "kernelutil_x86_avx_128_fma_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_VF_avx_128_fma_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB;
  61     int              j_coord_offsetA,j_coord_offsetB;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwioffset1;
  69     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  70     int              vdwioffset2;
  71     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  72     int              vdwioffset3;
  73     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  74     int              vdwjidx0A,vdwjidx0B;
  75     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  76     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  77     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  78     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  79     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  80     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  87     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  88     __m128i          vfitab;
  89     __m128i          ifour       = _mm_set1_epi32(4);
  90     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  91     real             *vftab;
  92     __m128i          ewitab;
  93     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  94     real             *ewtab;
  95     __m128d          dummy_mask,cutoff_mask;
  96     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  97     __m128d          one     = _mm_set1_pd(1.0);
  98     __m128d          two     = _mm_set1_pd(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     facel            = _mm_set1_pd(fr->epsfac);
 111     charge           = mdatoms->chargeA;
 112     nvdwtype         = fr->ntype;
 113     vdwparam         = fr->nbfp;
 114     vdwtype          = mdatoms->typeA;
 115
 116     vftab            = kernel_data->table_vdw->data;
 117     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 118
 119     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 120     ewtab            = fr->ic->tabq_coul_FDV0;
 121     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 122     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 123
 124     /* Setup water-specific parameters */
 125     inr              = nlist->iinr[0];
 126     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 127     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 128     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 129     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 130
 131     /* Avoid stupid compiler warnings */
 132     jnrA = jnrB = 0;
 133     j_coord_offsetA = 0;
 134     j_coord_offsetB = 0;
 135
 136     outeriter        = 0;
 137     inneriter        = 0;
 138
 139     /* Start outer loop over neighborlists */
 140     for(iidx=0; iidx<nri; iidx++)
 141     {
 142         /* Load shift vector for this list */
 143         i_shift_offset   = DIM*shiftidx[iidx];
 144
 145         /* Load limits for loop over neighbors */
 146         j_index_start    = jindex[iidx];
 147         j_index_end      = jindex[iidx+1];
 148
 149         /* Get outer coordinate index */
 150         inr              = iinr[iidx];
 151         i_coord_offset   = DIM*inr;
 152
 153         /* Load i particle coords and add shift vector */
 154         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 155                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 156
 157         fix0             = _mm_setzero_pd();
 158         fiy0             = _mm_setzero_pd();
 159         fiz0             = _mm_setzero_pd();
 160         fix1             = _mm_setzero_pd();
 161         fiy1             = _mm_setzero_pd();
 162         fiz1             = _mm_setzero_pd();
 163         fix2             = _mm_setzero_pd();
 164         fiy2             = _mm_setzero_pd();
 165         fiz2             = _mm_setzero_pd();
 166         fix3             = _mm_setzero_pd();
 167         fiy3             = _mm_setzero_pd();
 168         fiz3             = _mm_setzero_pd();
 169
 170         /* Reset potential sums */
 171         velecsum         = _mm_setzero_pd();
 172         vvdwsum          = _mm_setzero_pd();
 173
 174         /* Start inner kernel loop */
 175         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 176         {
 177
 178             /* Get j neighbor index, and coordinate index */
 179             jnrA             = jjnr[jidx];
 180             jnrB             = jjnr[jidx+1];
 181             j_coord_offsetA  = DIM*jnrA;
 182             j_coord_offsetB  = DIM*jnrB;
 183
 184             /* load j atom coordinates */
 185             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 186                                               &jx0,&jy0,&jz0);
 187
 188             /* Calculate displacement vector */
 189             dx00             = _mm_sub_pd(ix0,jx0);
 190             dy00             = _mm_sub_pd(iy0,jy0);
 191             dz00             = _mm_sub_pd(iz0,jz0);
 192             dx10             = _mm_sub_pd(ix1,jx0);
 193             dy10             = _mm_sub_pd(iy1,jy0);
 194             dz10             = _mm_sub_pd(iz1,jz0);
 195             dx20             = _mm_sub_pd(ix2,jx0);
 196             dy20             = _mm_sub_pd(iy2,jy0);
 197             dz20             = _mm_sub_pd(iz2,jz0);
 198             dx30             = _mm_sub_pd(ix3,jx0);
 199             dy30             = _mm_sub_pd(iy3,jy0);
 200             dz30             = _mm_sub_pd(iz3,jz0);
 201
 202             /* Calculate squared distance and things based on it */
 203             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 204             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 205             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 206             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 207
 208             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 209             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 210             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 211             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 212
 213             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 214             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 215             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 216
 217             /* Load parameters for j particles */
 218             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 219             vdwjidx0A        = 2*vdwtype[jnrA+0];
 220             vdwjidx0B        = 2*vdwtype[jnrB+0];
 221
 222             fjx0             = _mm_setzero_pd();
 223             fjy0             = _mm_setzero_pd();
 224             fjz0             = _mm_setzero_pd();
 225
 226             /**************************
 227              * CALCULATE INTERACTIONS *
 228              **************************/
 229
 230             r00              = _mm_mul_pd(rsq00,rinv00);
 231
 232             /* Compute parameters for interactions between i and j atoms */
 233             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 234                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 235
 236             /* Calculate table index by multiplying r with table scale and truncate to integer */
 237             rt               = _mm_mul_pd(r00,vftabscale);
 238             vfitab           = _mm_cvttpd_epi32(rt);
 239 #ifdef __XOP__
 240             vfeps            = _mm_frcz_pd(rt);
 241 #else
 242             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 243 #endif
 244             twovfeps         = _mm_add_pd(vfeps,vfeps);
 245             vfitab           = _mm_slli_epi32(vfitab,3);
 246
 247             /* CUBIC SPLINE TABLE DISPERSION */
 248             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 249             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 250             GMX_MM_TRANSPOSE2_PD(Y,F);
 251             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 252             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 253             GMX_MM_TRANSPOSE2_PD(G,H);
 254             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 255             VV               = _mm_macc_pd(vfeps,Fp,Y);
 256             vvdw6            = _mm_mul_pd(c6_00,VV);
 257             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 258             fvdw6            = _mm_mul_pd(c6_00,FF);
 259
 260             /* CUBIC SPLINE TABLE REPULSION */
 261             vfitab           = _mm_add_epi32(vfitab,ifour);
 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             vvdw12           = _mm_mul_pd(c12_00,VV);
 271             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 272             fvdw12           = _mm_mul_pd(c12_00,FF);
 273             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 274             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 275
 276             /* Update potential sum for this i atom from the interaction with this j atom. */
 277             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 278
 279             fscal            = fvdw;
 280
 281             /* Update vectorial force */
 282             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 283             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 284             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 285
 286             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 287             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 288             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 289
 290             /**************************
 291              * CALCULATE INTERACTIONS *
 292              **************************/
 293
 294             r10              = _mm_mul_pd(rsq10,rinv10);
 295
 296             /* Compute parameters for interactions between i and j atoms */
 297             qq10             = _mm_mul_pd(iq1,jq0);
 298
 299             /* EWALD ELECTROSTATICS */
 300
 301             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 302             ewrt             = _mm_mul_pd(r10,ewtabscale);
 303             ewitab           = _mm_cvttpd_epi32(ewrt);
 304 #ifdef __XOP__
 305             eweps            = _mm_frcz_pd(ewrt);
 306 #else
 307             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 308 #endif
 309             twoeweps         = _mm_add_pd(eweps,eweps);
 310             ewitab           = _mm_slli_epi32(ewitab,2);
 311             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 312             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 313             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 314             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 315             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 316             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 317             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 318             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 319             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 320             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 321
 322             /* Update potential sum for this i atom from the interaction with this j atom. */
 323             velecsum         = _mm_add_pd(velecsum,velec);
 324
 325             fscal            = felec;
 326
 327             /* Update vectorial force */
 328             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 329             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 330             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 331
 332             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 333             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 334             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 335
 336             /**************************
 337              * CALCULATE INTERACTIONS *
 338              **************************/
 339
 340             r20              = _mm_mul_pd(rsq20,rinv20);
 341
 342             /* Compute parameters for interactions between i and j atoms */
 343             qq20             = _mm_mul_pd(iq2,jq0);
 344
 345             /* EWALD ELECTROSTATICS */
 346
 347             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 348             ewrt             = _mm_mul_pd(r20,ewtabscale);
 349             ewitab           = _mm_cvttpd_epi32(ewrt);
 350 #ifdef __XOP__
 351             eweps            = _mm_frcz_pd(ewrt);
 352 #else
 353             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 354 #endif
 355             twoeweps         = _mm_add_pd(eweps,eweps);
 356             ewitab           = _mm_slli_epi32(ewitab,2);
 357             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 358             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 359             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 360             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 361             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 362             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 363             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 364             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 365             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 366             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 367
 368             /* Update potential sum for this i atom from the interaction with this j atom. */
 369             velecsum         = _mm_add_pd(velecsum,velec);
 370
 371             fscal            = felec;
 372
 373             /* Update vectorial force */
 374             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 375             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 376             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 377
 378             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 379             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 380             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 381
 382             /**************************
 383              * CALCULATE INTERACTIONS *
 384              **************************/
 385
 386             r30              = _mm_mul_pd(rsq30,rinv30);
 387
 388             /* Compute parameters for interactions between i and j atoms */
 389             qq30             = _mm_mul_pd(iq3,jq0);
 390
 391             /* EWALD ELECTROSTATICS */
 392
 393             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 394             ewrt             = _mm_mul_pd(r30,ewtabscale);
 395             ewitab           = _mm_cvttpd_epi32(ewrt);
 396 #ifdef __XOP__
 397             eweps            = _mm_frcz_pd(ewrt);
 398 #else
 399             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 400 #endif
 401             twoeweps         = _mm_add_pd(eweps,eweps);
 402             ewitab           = _mm_slli_epi32(ewitab,2);
 403             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 404             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 405             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 406             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 407             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 408             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 409             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 410             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 411             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 412             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 413
 414             /* Update potential sum for this i atom from the interaction with this j atom. */
 415             velecsum         = _mm_add_pd(velecsum,velec);
 416
 417             fscal            = felec;
 418
 419             /* Update vectorial force */
 420             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 421             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 422             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 423
 424             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 425             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 426             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 427
 428             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 429
 430             /* Inner loop uses 194 flops */
 431         }
 432
 433         if(jidx<j_index_end)
 434         {
 435
 436             jnrA             = jjnr[jidx];
 437             j_coord_offsetA  = DIM*jnrA;
 438
 439             /* load j atom coordinates */
 440             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 441                                               &jx0,&jy0,&jz0);
 442
 443             /* Calculate displacement vector */
 444             dx00             = _mm_sub_pd(ix0,jx0);
 445             dy00             = _mm_sub_pd(iy0,jy0);
 446             dz00             = _mm_sub_pd(iz0,jz0);
 447             dx10             = _mm_sub_pd(ix1,jx0);
 448             dy10             = _mm_sub_pd(iy1,jy0);
 449             dz10             = _mm_sub_pd(iz1,jz0);
 450             dx20             = _mm_sub_pd(ix2,jx0);
 451             dy20             = _mm_sub_pd(iy2,jy0);
 452             dz20             = _mm_sub_pd(iz2,jz0);
 453             dx30             = _mm_sub_pd(ix3,jx0);
 454             dy30             = _mm_sub_pd(iy3,jy0);
 455             dz30             = _mm_sub_pd(iz3,jz0);
 456
 457             /* Calculate squared distance and things based on it */
 458             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 459             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 460             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 461             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 462
 463             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 464             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 465             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 466             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 467
 468             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 469             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 470             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 471
 472             /* Load parameters for j particles */
 473             jq0              = _mm_load_sd(charge+jnrA+0);
 474             vdwjidx0A        = 2*vdwtype[jnrA+0];
 475
 476             fjx0             = _mm_setzero_pd();
 477             fjy0             = _mm_setzero_pd();
 478             fjz0             = _mm_setzero_pd();
 479
 480             /**************************
 481              * CALCULATE INTERACTIONS *
 482              **************************/
 483
 484             r00              = _mm_mul_pd(rsq00,rinv00);
 485
 486             /* Compute parameters for interactions between i and j atoms */
 487             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 488
 489             /* Calculate table index by multiplying r with table scale and truncate to integer */
 490             rt               = _mm_mul_pd(r00,vftabscale);
 491             vfitab           = _mm_cvttpd_epi32(rt);
 492 #ifdef __XOP__
 493             vfeps            = _mm_frcz_pd(rt);
 494 #else
 495             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 496 #endif
 497             twovfeps         = _mm_add_pd(vfeps,vfeps);
 498             vfitab           = _mm_slli_epi32(vfitab,3);
 499
 500             /* CUBIC SPLINE TABLE DISPERSION */
 501             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 502             F                = _mm_setzero_pd();
 503             GMX_MM_TRANSPOSE2_PD(Y,F);
 504             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 505             H                = _mm_setzero_pd();
 506             GMX_MM_TRANSPOSE2_PD(G,H);
 507             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 508             VV               = _mm_macc_pd(vfeps,Fp,Y);
 509             vvdw6            = _mm_mul_pd(c6_00,VV);
 510             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 511             fvdw6            = _mm_mul_pd(c6_00,FF);
 512
 513             /* CUBIC SPLINE TABLE REPULSION */
 514             vfitab           = _mm_add_epi32(vfitab,ifour);
 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             vvdw12           = _mm_mul_pd(c12_00,VV);
 524             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 525             fvdw12           = _mm_mul_pd(c12_00,FF);
 526             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 527             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 528
 529             /* Update potential sum for this i atom from the interaction with this j atom. */
 530             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 531             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 532
 533             fscal            = fvdw;
 534
 535             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 536
 537             /* Update vectorial force */
 538             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 539             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 540             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 541
 542             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 543             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 544             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 545
 546             /**************************
 547              * CALCULATE INTERACTIONS *
 548              **************************/
 549
 550             r10              = _mm_mul_pd(rsq10,rinv10);
 551
 552             /* Compute parameters for interactions between i and j atoms */
 553             qq10             = _mm_mul_pd(iq1,jq0);
 554
 555             /* EWALD ELECTROSTATICS */
 556
 557             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 558             ewrt             = _mm_mul_pd(r10,ewtabscale);
 559             ewitab           = _mm_cvttpd_epi32(ewrt);
 560 #ifdef __XOP__
 561             eweps            = _mm_frcz_pd(ewrt);
 562 #else
 563             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 564 #endif
 565             twoeweps         = _mm_add_pd(eweps,eweps);
 566             ewitab           = _mm_slli_epi32(ewitab,2);
 567             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 568             ewtabD           = _mm_setzero_pd();
 569             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 570             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 571             ewtabFn          = _mm_setzero_pd();
 572             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 573             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 574             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 575             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 576             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 577
 578             /* Update potential sum for this i atom from the interaction with this j atom. */
 579             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 580             velecsum         = _mm_add_pd(velecsum,velec);
 581
 582             fscal            = felec;
 583
 584             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 585
 586             /* Update vectorial force */
 587             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 588             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 589             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 590
 591             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 592             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 593             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 594
 595             /**************************
 596              * CALCULATE INTERACTIONS *
 597              **************************/
 598
 599             r20              = _mm_mul_pd(rsq20,rinv20);
 600
 601             /* Compute parameters for interactions between i and j atoms */
 602             qq20             = _mm_mul_pd(iq2,jq0);
 603
 604             /* EWALD ELECTROSTATICS */
 605
 606             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 607             ewrt             = _mm_mul_pd(r20,ewtabscale);
 608             ewitab           = _mm_cvttpd_epi32(ewrt);
 609 #ifdef __XOP__
 610             eweps            = _mm_frcz_pd(ewrt);
 611 #else
 612             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 613 #endif
 614             twoeweps         = _mm_add_pd(eweps,eweps);
 615             ewitab           = _mm_slli_epi32(ewitab,2);
 616             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 617             ewtabD           = _mm_setzero_pd();
 618             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 619             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 620             ewtabFn          = _mm_setzero_pd();
 621             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 622             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 623             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 624             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 625             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 626
 627             /* Update potential sum for this i atom from the interaction with this j atom. */
 628             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 629             velecsum         = _mm_add_pd(velecsum,velec);
 630
 631             fscal            = felec;
 632
 633             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 634
 635             /* Update vectorial force */
 636             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 637             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 638             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 639
 640             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 641             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 642             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 643
 644             /**************************
 645              * CALCULATE INTERACTIONS *
 646              **************************/
 647
 648             r30              = _mm_mul_pd(rsq30,rinv30);
 649
 650             /* Compute parameters for interactions between i and j atoms */
 651             qq30             = _mm_mul_pd(iq3,jq0);
 652
 653             /* EWALD ELECTROSTATICS */
 654
 655             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 656             ewrt             = _mm_mul_pd(r30,ewtabscale);
 657             ewitab           = _mm_cvttpd_epi32(ewrt);
 658 #ifdef __XOP__
 659             eweps            = _mm_frcz_pd(ewrt);
 660 #else
 661             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 662 #endif
 663             twoeweps         = _mm_add_pd(eweps,eweps);
 664             ewitab           = _mm_slli_epi32(ewitab,2);
 665             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 666             ewtabD           = _mm_setzero_pd();
 667             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 668             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 669             ewtabFn          = _mm_setzero_pd();
 670             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 671             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 672             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 673             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 674             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 675
 676             /* Update potential sum for this i atom from the interaction with this j atom. */
 677             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 678             velecsum         = _mm_add_pd(velecsum,velec);
 679
 680             fscal            = felec;
 681
 682             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 683
 684             /* Update vectorial force */
 685             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 686             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 687             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 688
 689             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 690             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 691             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 692
 693             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 694
 695             /* Inner loop uses 194 flops */
 696         }
 697
 698         /* End of innermost loop */
 699
 700         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 701                                               f+i_coord_offset,fshift+i_shift_offset);
 702
 703         ggid                        = gid[iidx];
 704         /* Update potential energies */
 705         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 706         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 707
 708         /* Increment number of inner iterations */
 709         inneriter                  += j_index_end - j_index_start;
 710
 711         /* Outer loop uses 26 flops */
 712     }
 713
 714     /* Increment number of outer iterations */
 715     outeriter        += nri;
 716
 717     /* Update outer/inner flops */
 718
 719     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*194);
 720 }
 721 /*
 722  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_avx_128_fma_double
 723  * Electrostatics interaction: Ewald
 724  * VdW interaction:            CubicSplineTable
 725  * Geometry:                   Water4-Particle
 726  * Calculate force/pot:        Force
 727  */
 728 void
 729 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_avx_128_fma_double
 730                     (t_nblist * gmx_restrict                nlist,
 731                      rvec * gmx_restrict                    xx,
 732                      rvec * gmx_restrict                    ff,
 733                      t_forcerec * gmx_restrict              fr,
 734                      t_mdatoms * gmx_restrict               mdatoms,
 735                      nb_kernel_data_t * gmx_restrict        kernel_data,
 736                      t_nrnb * gmx_restrict                  nrnb)
 737 {
 738     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 739      * just 0 for non-waters.
 740      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 741      * jnr indices corresponding to data put in the four positions in the SIMD register.
 742      */
 743     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 744     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 745     int              jnrA,jnrB;
 746     int              j_coord_offsetA,j_coord_offsetB;
 747     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 748     real             rcutoff_scalar;
 749     real             *shiftvec,*fshift,*x,*f;
 750     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 751     int              vdwioffset0;
 752     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 753     int              vdwioffset1;
 754     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 755     int              vdwioffset2;
 756     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 757     int              vdwioffset3;
 758     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 759     int              vdwjidx0A,vdwjidx0B;
 760     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 761     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 762     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 763     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 764     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 765     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 766     real             *charge;
 767     int              nvdwtype;
 768     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 769     int              *vdwtype;
 770     real             *vdwparam;
 771     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 772     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 773     __m128i          vfitab;
 774     __m128i          ifour       = _mm_set1_epi32(4);
 775     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 776     real             *vftab;
 777     __m128i          ewitab;
 778     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 779     real             *ewtab;
 780     __m128d          dummy_mask,cutoff_mask;
 781     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 782     __m128d          one     = _mm_set1_pd(1.0);
 783     __m128d          two     = _mm_set1_pd(2.0);
 784     x                = xx[0];
 785     f                = ff[0];
 786
 787     nri              = nlist->nri;
 788     iinr             = nlist->iinr;
 789     jindex           = nlist->jindex;
 790     jjnr             = nlist->jjnr;
 791     shiftidx         = nlist->shift;
 792     gid              = nlist->gid;
 793     shiftvec         = fr->shift_vec[0];
 794     fshift           = fr->fshift[0];
 795     facel            = _mm_set1_pd(fr->epsfac);
 796     charge           = mdatoms->chargeA;
 797     nvdwtype         = fr->ntype;
 798     vdwparam         = fr->nbfp;
 799     vdwtype          = mdatoms->typeA;
 800
 801     vftab            = kernel_data->table_vdw->data;
 802     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 803
 804     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 805     ewtab            = fr->ic->tabq_coul_F;
 806     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 807     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 808
 809     /* Setup water-specific parameters */
 810     inr              = nlist->iinr[0];
 811     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 812     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 813     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 814     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 815
 816     /* Avoid stupid compiler warnings */
 817     jnrA = jnrB = 0;
 818     j_coord_offsetA = 0;
 819     j_coord_offsetB = 0;
 820
 821     outeriter        = 0;
 822     inneriter        = 0;
 823
 824     /* Start outer loop over neighborlists */
 825     for(iidx=0; iidx<nri; iidx++)
 826     {
 827         /* Load shift vector for this list */
 828         i_shift_offset   = DIM*shiftidx[iidx];
 829
 830         /* Load limits for loop over neighbors */
 831         j_index_start    = jindex[iidx];
 832         j_index_end      = jindex[iidx+1];
 833
 834         /* Get outer coordinate index */
 835         inr              = iinr[iidx];
 836         i_coord_offset   = DIM*inr;
 837
 838         /* Load i particle coords and add shift vector */
 839         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 840                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 841
 842         fix0             = _mm_setzero_pd();
 843         fiy0             = _mm_setzero_pd();
 844         fiz0             = _mm_setzero_pd();
 845         fix1             = _mm_setzero_pd();
 846         fiy1             = _mm_setzero_pd();
 847         fiz1             = _mm_setzero_pd();
 848         fix2             = _mm_setzero_pd();
 849         fiy2             = _mm_setzero_pd();
 850         fiz2             = _mm_setzero_pd();
 851         fix3             = _mm_setzero_pd();
 852         fiy3             = _mm_setzero_pd();
 853         fiz3             = _mm_setzero_pd();
 854
 855         /* Start inner kernel loop */
 856         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 857         {
 858
 859             /* Get j neighbor index, and coordinate index */
 860             jnrA             = jjnr[jidx];
 861             jnrB             = jjnr[jidx+1];
 862             j_coord_offsetA  = DIM*jnrA;
 863             j_coord_offsetB  = DIM*jnrB;
 864
 865             /* load j atom coordinates */
 866             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 867                                               &jx0,&jy0,&jz0);
 868
 869             /* Calculate displacement vector */
 870             dx00             = _mm_sub_pd(ix0,jx0);
 871             dy00             = _mm_sub_pd(iy0,jy0);
 872             dz00             = _mm_sub_pd(iz0,jz0);
 873             dx10             = _mm_sub_pd(ix1,jx0);
 874             dy10             = _mm_sub_pd(iy1,jy0);
 875             dz10             = _mm_sub_pd(iz1,jz0);
 876             dx20             = _mm_sub_pd(ix2,jx0);
 877             dy20             = _mm_sub_pd(iy2,jy0);
 878             dz20             = _mm_sub_pd(iz2,jz0);
 879             dx30             = _mm_sub_pd(ix3,jx0);
 880             dy30             = _mm_sub_pd(iy3,jy0);
 881             dz30             = _mm_sub_pd(iz3,jz0);
 882
 883             /* Calculate squared distance and things based on it */
 884             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 885             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 886             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 887             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 888
 889             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 890             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 891             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 892             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 893
 894             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 895             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 896             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 897
 898             /* Load parameters for j particles */
 899             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 900             vdwjidx0A        = 2*vdwtype[jnrA+0];
 901             vdwjidx0B        = 2*vdwtype[jnrB+0];
 902
 903             fjx0             = _mm_setzero_pd();
 904             fjy0             = _mm_setzero_pd();
 905             fjz0             = _mm_setzero_pd();
 906
 907             /**************************
 908              * CALCULATE INTERACTIONS *
 909              **************************/
 910
 911             r00              = _mm_mul_pd(rsq00,rinv00);
 912
 913             /* Compute parameters for interactions between i and j atoms */
 914             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 915                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 916
 917             /* Calculate table index by multiplying r with table scale and truncate to integer */
 918             rt               = _mm_mul_pd(r00,vftabscale);
 919             vfitab           = _mm_cvttpd_epi32(rt);
 920 #ifdef __XOP__
 921             vfeps            = _mm_frcz_pd(rt);
 922 #else
 923             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 924 #endif
 925             twovfeps         = _mm_add_pd(vfeps,vfeps);
 926             vfitab           = _mm_slli_epi32(vfitab,3);
 927
 928             /* CUBIC SPLINE TABLE DISPERSION */
 929             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 930             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 931             GMX_MM_TRANSPOSE2_PD(Y,F);
 932             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 933             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 934             GMX_MM_TRANSPOSE2_PD(G,H);
 935             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 936             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 937             fvdw6            = _mm_mul_pd(c6_00,FF);
 938
 939             /* CUBIC SPLINE TABLE REPULSION */
 940             vfitab           = _mm_add_epi32(vfitab,ifour);
 941             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 942             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 943             GMX_MM_TRANSPOSE2_PD(Y,F);
 944             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 945             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 946             GMX_MM_TRANSPOSE2_PD(G,H);
 947             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 948             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 949             fvdw12           = _mm_mul_pd(c12_00,FF);
 950             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 951
 952             fscal            = fvdw;
 953
 954             /* Update vectorial force */
 955             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 956             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 957             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 958
 959             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 960             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 961             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 962
 963             /**************************
 964              * CALCULATE INTERACTIONS *
 965              **************************/
 966
 967             r10              = _mm_mul_pd(rsq10,rinv10);
 968
 969             /* Compute parameters for interactions between i and j atoms */
 970             qq10             = _mm_mul_pd(iq1,jq0);
 971
 972             /* EWALD ELECTROSTATICS */
 973
 974             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 975             ewrt             = _mm_mul_pd(r10,ewtabscale);
 976             ewitab           = _mm_cvttpd_epi32(ewrt);
 977 #ifdef __XOP__
 978             eweps            = _mm_frcz_pd(ewrt);
 979 #else
 980             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 981 #endif
 982             twoeweps         = _mm_add_pd(eweps,eweps);
 983             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 984                                          &ewtabF,&ewtabFn);
 985             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 986             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 987
 988             fscal            = felec;
 989
 990             /* Update vectorial force */
 991             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 992             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 993             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 994
 995             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 996             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 997             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 998
 999             /**************************
1000              * CALCULATE INTERACTIONS *
1001              **************************/
1002
1003             r20              = _mm_mul_pd(rsq20,rinv20);
1004
1005             /* Compute parameters for interactions between i and j atoms */
1006             qq20             = _mm_mul_pd(iq2,jq0);
1007
1008             /* EWALD ELECTROSTATICS */
1009
1010             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1011             ewrt             = _mm_mul_pd(r20,ewtabscale);
1012             ewitab           = _mm_cvttpd_epi32(ewrt);
1013 #ifdef __XOP__
1014             eweps            = _mm_frcz_pd(ewrt);
1015 #else
1016             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1017 #endif
1018             twoeweps         = _mm_add_pd(eweps,eweps);
1019             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1020                                          &ewtabF,&ewtabFn);
1021             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1022             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1023
1024             fscal            = felec;
1025
1026             /* Update vectorial force */
1027             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1028             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1029             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1030
1031             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1032             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1033             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1034
1035             /**************************
1036              * CALCULATE INTERACTIONS *
1037              **************************/
1038
1039             r30              = _mm_mul_pd(rsq30,rinv30);
1040
1041             /* Compute parameters for interactions between i and j atoms */
1042             qq30             = _mm_mul_pd(iq3,jq0);
1043
1044             /* EWALD ELECTROSTATICS */
1045
1046             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1047             ewrt             = _mm_mul_pd(r30,ewtabscale);
1048             ewitab           = _mm_cvttpd_epi32(ewrt);
1049 #ifdef __XOP__
1050             eweps            = _mm_frcz_pd(ewrt);
1051 #else
1052             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1053 #endif
1054             twoeweps         = _mm_add_pd(eweps,eweps);
1055             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1056                                          &ewtabF,&ewtabFn);
1057             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1058             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1059
1060             fscal            = felec;
1061
1062             /* Update vectorial force */
1063             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1064             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1065             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1066
1067             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1068             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1069             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1070
1071             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1072
1073             /* Inner loop uses 171 flops */
1074         }
1075
1076         if(jidx<j_index_end)
1077         {
1078
1079             jnrA             = jjnr[jidx];
1080             j_coord_offsetA  = DIM*jnrA;
1081
1082             /* load j atom coordinates */
1083             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1084                                               &jx0,&jy0,&jz0);
1085
1086             /* Calculate displacement vector */
1087             dx00             = _mm_sub_pd(ix0,jx0);
1088             dy00             = _mm_sub_pd(iy0,jy0);
1089             dz00             = _mm_sub_pd(iz0,jz0);
1090             dx10             = _mm_sub_pd(ix1,jx0);
1091             dy10             = _mm_sub_pd(iy1,jy0);
1092             dz10             = _mm_sub_pd(iz1,jz0);
1093             dx20             = _mm_sub_pd(ix2,jx0);
1094             dy20             = _mm_sub_pd(iy2,jy0);
1095             dz20             = _mm_sub_pd(iz2,jz0);
1096             dx30             = _mm_sub_pd(ix3,jx0);
1097             dy30             = _mm_sub_pd(iy3,jy0);
1098             dz30             = _mm_sub_pd(iz3,jz0);
1099
1100             /* Calculate squared distance and things based on it */
1101             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1102             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1103             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1104             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1105
1106             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1107             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1108             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1109             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1110
1111             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1112             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1113             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1114
1115             /* Load parameters for j particles */
1116             jq0              = _mm_load_sd(charge+jnrA+0);
1117             vdwjidx0A        = 2*vdwtype[jnrA+0];
1118
1119             fjx0             = _mm_setzero_pd();
1120             fjy0             = _mm_setzero_pd();
1121             fjz0             = _mm_setzero_pd();
1122
1123             /**************************
1124              * CALCULATE INTERACTIONS *
1125              **************************/
1126
1127             r00              = _mm_mul_pd(rsq00,rinv00);
1128
1129             /* Compute parameters for interactions between i and j atoms */
1130             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1131
1132             /* Calculate table index by multiplying r with table scale and truncate to integer */
1133             rt               = _mm_mul_pd(r00,vftabscale);
1134             vfitab           = _mm_cvttpd_epi32(rt);
1135 #ifdef __XOP__
1136             vfeps            = _mm_frcz_pd(rt);
1137 #else
1138             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1139 #endif
1140             twovfeps         = _mm_add_pd(vfeps,vfeps);
1141             vfitab           = _mm_slli_epi32(vfitab,3);
1142
1143             /* CUBIC SPLINE TABLE DISPERSION */
1144             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1145             F                = _mm_setzero_pd();
1146             GMX_MM_TRANSPOSE2_PD(Y,F);
1147             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1148             H                = _mm_setzero_pd();
1149             GMX_MM_TRANSPOSE2_PD(G,H);
1150             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1151             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1152             fvdw6            = _mm_mul_pd(c6_00,FF);
1153
1154             /* CUBIC SPLINE TABLE REPULSION */
1155             vfitab           = _mm_add_epi32(vfitab,ifour);
1156             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1157             F                = _mm_setzero_pd();
1158             GMX_MM_TRANSPOSE2_PD(Y,F);
1159             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1160             H                = _mm_setzero_pd();
1161             GMX_MM_TRANSPOSE2_PD(G,H);
1162             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1163             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1164             fvdw12           = _mm_mul_pd(c12_00,FF);
1165             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1166
1167             fscal            = fvdw;
1168
1169             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1170
1171             /* Update vectorial force */
1172             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1173             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1174             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1175
1176             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1177             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1178             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1179
1180             /**************************
1181              * CALCULATE INTERACTIONS *
1182              **************************/
1183
1184             r10              = _mm_mul_pd(rsq10,rinv10);
1185
1186             /* Compute parameters for interactions between i and j atoms */
1187             qq10             = _mm_mul_pd(iq1,jq0);
1188
1189             /* EWALD ELECTROSTATICS */
1190
1191             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1192             ewrt             = _mm_mul_pd(r10,ewtabscale);
1193             ewitab           = _mm_cvttpd_epi32(ewrt);
1194 #ifdef __XOP__
1195             eweps            = _mm_frcz_pd(ewrt);
1196 #else
1197             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1198 #endif
1199             twoeweps         = _mm_add_pd(eweps,eweps);
1200             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1201             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1202             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1203
1204             fscal            = felec;
1205
1206             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1207
1208             /* Update vectorial force */
1209             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1210             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1211             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1212
1213             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1214             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1215             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1216
1217             /**************************
1218              * CALCULATE INTERACTIONS *
1219              **************************/
1220
1221             r20              = _mm_mul_pd(rsq20,rinv20);
1222
1223             /* Compute parameters for interactions between i and j atoms */
1224             qq20             = _mm_mul_pd(iq2,jq0);
1225
1226             /* EWALD ELECTROSTATICS */
1227
1228             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1229             ewrt             = _mm_mul_pd(r20,ewtabscale);
1230             ewitab           = _mm_cvttpd_epi32(ewrt);
1231 #ifdef __XOP__
1232             eweps            = _mm_frcz_pd(ewrt);
1233 #else
1234             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1235 #endif
1236             twoeweps         = _mm_add_pd(eweps,eweps);
1237             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1238             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1239             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1240
1241             fscal            = felec;
1242
1243             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1244
1245             /* Update vectorial force */
1246             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1247             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1248             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1249
1250             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1251             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1252             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1253
1254             /**************************
1255              * CALCULATE INTERACTIONS *
1256              **************************/
1257
1258             r30              = _mm_mul_pd(rsq30,rinv30);
1259
1260             /* Compute parameters for interactions between i and j atoms */
1261             qq30             = _mm_mul_pd(iq3,jq0);
1262
1263             /* EWALD ELECTROSTATICS */
1264
1265             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1266             ewrt             = _mm_mul_pd(r30,ewtabscale);
1267             ewitab           = _mm_cvttpd_epi32(ewrt);
1268 #ifdef __XOP__
1269             eweps            = _mm_frcz_pd(ewrt);
1270 #else
1271             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1272 #endif
1273             twoeweps         = _mm_add_pd(eweps,eweps);
1274             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1275             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1276             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1277
1278             fscal            = felec;
1279
1280             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1281
1282             /* Update vectorial force */
1283             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1284             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1285             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1286
1287             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1288             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1289             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1290
1291             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1292
1293             /* Inner loop uses 171 flops */
1294         }
1295
1296         /* End of innermost loop */
1297
1298         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1299                                               f+i_coord_offset,fshift+i_shift_offset);
1300
1301         /* Increment number of inner iterations */
1302         inneriter                  += j_index_end - j_index_start;
1303
1304         /* Outer loop uses 24 flops */
1305     }
1306
1307     /* Increment number of outer iterations */
1308     outeriter        += nri;
1309
1310     /* Update outer/inner flops */
1311
1312     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*171);
1313 }