src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCSTab_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_ElecCSTab_VdwCSTab_GeomW4P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_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     __m128d          dummy_mask,cutoff_mask;
 109     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 110     __m128d          one     = _mm_set1_pd(1.0);
 111     __m128d          two     = _mm_set1_pd(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm_set1_pd(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     vftab            = kernel_data->table_elec_vdw->data;
 130     vftabscale       = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
 131
 132     /* Setup water-specific parameters */
 133     inr              = nlist->iinr[0];
 134     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 135     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 136     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 137     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 138
 139     /* Avoid stupid compiler warnings */
 140     jnrA = jnrB = 0;
 141     j_coord_offsetA = 0;
 142     j_coord_offsetB = 0;
 143
 144     outeriter        = 0;
 145     inneriter        = 0;
 146
 147     /* Start outer loop over neighborlists */
 148     for(iidx=0; iidx<nri; iidx++)
 149     {
 150         /* Load shift vector for this list */
 151         i_shift_offset   = DIM*shiftidx[iidx];
 152
 153         /* Load limits for loop over neighbors */
 154         j_index_start    = jindex[iidx];
 155         j_index_end      = jindex[iidx+1];
 156
 157         /* Get outer coordinate index */
 158         inr              = iinr[iidx];
 159         i_coord_offset   = DIM*inr;
 160
 161         /* Load i particle coords and add shift vector */
 162         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 163                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 164
 165         fix0             = _mm_setzero_pd();
 166         fiy0             = _mm_setzero_pd();
 167         fiz0             = _mm_setzero_pd();
 168         fix1             = _mm_setzero_pd();
 169         fiy1             = _mm_setzero_pd();
 170         fiz1             = _mm_setzero_pd();
 171         fix2             = _mm_setzero_pd();
 172         fiy2             = _mm_setzero_pd();
 173         fiz2             = _mm_setzero_pd();
 174         fix3             = _mm_setzero_pd();
 175         fiy3             = _mm_setzero_pd();
 176         fiz3             = _mm_setzero_pd();
 177
 178         /* Reset potential sums */
 179         velecsum         = _mm_setzero_pd();
 180         vvdwsum          = _mm_setzero_pd();
 181
 182         /* Start inner kernel loop */
 183         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 184         {
 185
 186             /* Get j neighbor index, and coordinate index */
 187             jnrA             = jjnr[jidx];
 188             jnrB             = jjnr[jidx+1];
 189             j_coord_offsetA  = DIM*jnrA;
 190             j_coord_offsetB  = DIM*jnrB;
 191
 192             /* load j atom coordinates */
 193             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 194                                               &jx0,&jy0,&jz0);
 195
 196             /* Calculate displacement vector */
 197             dx00             = _mm_sub_pd(ix0,jx0);
 198             dy00             = _mm_sub_pd(iy0,jy0);
 199             dz00             = _mm_sub_pd(iz0,jz0);
 200             dx10             = _mm_sub_pd(ix1,jx0);
 201             dy10             = _mm_sub_pd(iy1,jy0);
 202             dz10             = _mm_sub_pd(iz1,jz0);
 203             dx20             = _mm_sub_pd(ix2,jx0);
 204             dy20             = _mm_sub_pd(iy2,jy0);
 205             dz20             = _mm_sub_pd(iz2,jz0);
 206             dx30             = _mm_sub_pd(ix3,jx0);
 207             dy30             = _mm_sub_pd(iy3,jy0);
 208             dz30             = _mm_sub_pd(iz3,jz0);
 209
 210             /* Calculate squared distance and things based on it */
 211             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 212             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 213             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 214             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 215
 216             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 217             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 218             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 219             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 220
 221             /* Load parameters for j particles */
 222             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 223             vdwjidx0A        = 2*vdwtype[jnrA+0];
 224             vdwjidx0B        = 2*vdwtype[jnrB+0];
 225
 226             fjx0             = _mm_setzero_pd();
 227             fjy0             = _mm_setzero_pd();
 228             fjz0             = _mm_setzero_pd();
 229
 230             /**************************
 231              * CALCULATE INTERACTIONS *
 232              **************************/
 233
 234             r00              = _mm_mul_pd(rsq00,rinv00);
 235
 236             /* Compute parameters for interactions between i and j atoms */
 237             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 238                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 239
 240             /* Calculate table index by multiplying r with table scale and truncate to integer */
 241             rt               = _mm_mul_pd(r00,vftabscale);
 242             vfitab           = _mm_cvttpd_epi32(rt);
 243 #ifdef __XOP__
 244             vfeps            = _mm_frcz_pd(rt);
 245 #else
 246             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 247 #endif
 248             twovfeps         = _mm_add_pd(vfeps,vfeps);
 249             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 250
 251             /* CUBIC SPLINE TABLE DISPERSION */
 252             vfitab           = _mm_add_epi32(vfitab,ifour);
 253             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 254             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 255             GMX_MM_TRANSPOSE2_PD(Y,F);
 256             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 257             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 258             GMX_MM_TRANSPOSE2_PD(G,H);
 259             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 260             VV               = _mm_macc_pd(vfeps,Fp,Y);
 261             vvdw6            = _mm_mul_pd(c6_00,VV);
 262             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 263             fvdw6            = _mm_mul_pd(c6_00,FF);
 264
 265             /* CUBIC SPLINE TABLE REPULSION */
 266             vfitab           = _mm_add_epi32(vfitab,ifour);
 267             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 268             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 269             GMX_MM_TRANSPOSE2_PD(Y,F);
 270             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 271             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 272             GMX_MM_TRANSPOSE2_PD(G,H);
 273             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 274             VV               = _mm_macc_pd(vfeps,Fp,Y);
 275             vvdw12           = _mm_mul_pd(c12_00,VV);
 276             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 277             fvdw12           = _mm_mul_pd(c12_00,FF);
 278             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 279             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 280
 281             /* Update potential sum for this i atom from the interaction with this j atom. */
 282             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 283
 284             fscal            = fvdw;
 285
 286             /* Update vectorial force */
 287             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 288             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 289             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 290
 291             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 292             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 293             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 294
 295             /**************************
 296              * CALCULATE INTERACTIONS *
 297              **************************/
 298
 299             r10              = _mm_mul_pd(rsq10,rinv10);
 300
 301             /* Compute parameters for interactions between i and j atoms */
 302             qq10             = _mm_mul_pd(iq1,jq0);
 303
 304             /* Calculate table index by multiplying r with table scale and truncate to integer */
 305             rt               = _mm_mul_pd(r10,vftabscale);
 306             vfitab           = _mm_cvttpd_epi32(rt);
 307 #ifdef __XOP__
 308             vfeps            = _mm_frcz_pd(rt);
 309 #else
 310             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 311 #endif
 312             twovfeps         = _mm_add_pd(vfeps,vfeps);
 313             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 314
 315             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 316             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 317             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 318             GMX_MM_TRANSPOSE2_PD(Y,F);
 319             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 320             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 321             GMX_MM_TRANSPOSE2_PD(G,H);
 322             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 323             VV               = _mm_macc_pd(vfeps,Fp,Y);
 324             velec            = _mm_mul_pd(qq10,VV);
 325             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 326             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 327
 328             /* Update potential sum for this i atom from the interaction with this j atom. */
 329             velecsum         = _mm_add_pd(velecsum,velec);
 330
 331             fscal            = felec;
 332
 333             /* Update vectorial force */
 334             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 335             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 336             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 337
 338             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 339             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 340             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 341
 342             /**************************
 343              * CALCULATE INTERACTIONS *
 344              **************************/
 345
 346             r20              = _mm_mul_pd(rsq20,rinv20);
 347
 348             /* Compute parameters for interactions between i and j atoms */
 349             qq20             = _mm_mul_pd(iq2,jq0);
 350
 351             /* Calculate table index by multiplying r with table scale and truncate to integer */
 352             rt               = _mm_mul_pd(r20,vftabscale);
 353             vfitab           = _mm_cvttpd_epi32(rt);
 354 #ifdef __XOP__
 355             vfeps            = _mm_frcz_pd(rt);
 356 #else
 357             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 358 #endif
 359             twovfeps         = _mm_add_pd(vfeps,vfeps);
 360             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 361
 362             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 363             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 364             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 365             GMX_MM_TRANSPOSE2_PD(Y,F);
 366             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 367             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 368             GMX_MM_TRANSPOSE2_PD(G,H);
 369             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 370             VV               = _mm_macc_pd(vfeps,Fp,Y);
 371             velec            = _mm_mul_pd(qq20,VV);
 372             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 373             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 374
 375             /* Update potential sum for this i atom from the interaction with this j atom. */
 376             velecsum         = _mm_add_pd(velecsum,velec);
 377
 378             fscal            = felec;
 379
 380             /* Update vectorial force */
 381             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 382             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 383             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 384
 385             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 386             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 387             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 388
 389             /**************************
 390              * CALCULATE INTERACTIONS *
 391              **************************/
 392
 393             r30              = _mm_mul_pd(rsq30,rinv30);
 394
 395             /* Compute parameters for interactions between i and j atoms */
 396             qq30             = _mm_mul_pd(iq3,jq0);
 397
 398             /* Calculate table index by multiplying r with table scale and truncate to integer */
 399             rt               = _mm_mul_pd(r30,vftabscale);
 400             vfitab           = _mm_cvttpd_epi32(rt);
 401 #ifdef __XOP__
 402             vfeps            = _mm_frcz_pd(rt);
 403 #else
 404             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 405 #endif
 406             twovfeps         = _mm_add_pd(vfeps,vfeps);
 407             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 408
 409             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 410             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 411             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 412             GMX_MM_TRANSPOSE2_PD(Y,F);
 413             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 414             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 415             GMX_MM_TRANSPOSE2_PD(G,H);
 416             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 417             VV               = _mm_macc_pd(vfeps,Fp,Y);
 418             velec            = _mm_mul_pd(qq30,VV);
 419             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 420             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 421
 422             /* Update potential sum for this i atom from the interaction with this j atom. */
 423             velecsum         = _mm_add_pd(velecsum,velec);
 424
 425             fscal            = felec;
 426
 427             /* Update vectorial force */
 428             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 429             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 430             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 431
 432             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 433             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 434             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 435
 436             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 437
 438             /* Inner loop uses 200 flops */
 439         }
 440
 441         if(jidx<j_index_end)
 442         {
 443
 444             jnrA             = jjnr[jidx];
 445             j_coord_offsetA  = DIM*jnrA;
 446
 447             /* load j atom coordinates */
 448             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 449                                               &jx0,&jy0,&jz0);
 450
 451             /* Calculate displacement vector */
 452             dx00             = _mm_sub_pd(ix0,jx0);
 453             dy00             = _mm_sub_pd(iy0,jy0);
 454             dz00             = _mm_sub_pd(iz0,jz0);
 455             dx10             = _mm_sub_pd(ix1,jx0);
 456             dy10             = _mm_sub_pd(iy1,jy0);
 457             dz10             = _mm_sub_pd(iz1,jz0);
 458             dx20             = _mm_sub_pd(ix2,jx0);
 459             dy20             = _mm_sub_pd(iy2,jy0);
 460             dz20             = _mm_sub_pd(iz2,jz0);
 461             dx30             = _mm_sub_pd(ix3,jx0);
 462             dy30             = _mm_sub_pd(iy3,jy0);
 463             dz30             = _mm_sub_pd(iz3,jz0);
 464
 465             /* Calculate squared distance and things based on it */
 466             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 467             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 468             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 469             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 470
 471             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 472             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 473             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 474             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 475
 476             /* Load parameters for j particles */
 477             jq0              = _mm_load_sd(charge+jnrA+0);
 478             vdwjidx0A        = 2*vdwtype[jnrA+0];
 479
 480             fjx0             = _mm_setzero_pd();
 481             fjy0             = _mm_setzero_pd();
 482             fjz0             = _mm_setzero_pd();
 483
 484             /**************************
 485              * CALCULATE INTERACTIONS *
 486              **************************/
 487
 488             r00              = _mm_mul_pd(rsq00,rinv00);
 489
 490             /* Compute parameters for interactions between i and j atoms */
 491             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 492
 493             /* Calculate table index by multiplying r with table scale and truncate to integer */
 494             rt               = _mm_mul_pd(r00,vftabscale);
 495             vfitab           = _mm_cvttpd_epi32(rt);
 496 #ifdef __XOP__
 497             vfeps            = _mm_frcz_pd(rt);
 498 #else
 499             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 500 #endif
 501             twovfeps         = _mm_add_pd(vfeps,vfeps);
 502             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 503
 504             /* CUBIC SPLINE TABLE DISPERSION */
 505             vfitab           = _mm_add_epi32(vfitab,ifour);
 506             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 507             F                = _mm_setzero_pd();
 508             GMX_MM_TRANSPOSE2_PD(Y,F);
 509             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 510             H                = _mm_setzero_pd();
 511             GMX_MM_TRANSPOSE2_PD(G,H);
 512             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 513             VV               = _mm_macc_pd(vfeps,Fp,Y);
 514             vvdw6            = _mm_mul_pd(c6_00,VV);
 515             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 516             fvdw6            = _mm_mul_pd(c6_00,FF);
 517
 518             /* CUBIC SPLINE TABLE REPULSION */
 519             vfitab           = _mm_add_epi32(vfitab,ifour);
 520             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 521             F                = _mm_setzero_pd();
 522             GMX_MM_TRANSPOSE2_PD(Y,F);
 523             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 524             H                = _mm_setzero_pd();
 525             GMX_MM_TRANSPOSE2_PD(G,H);
 526             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 527             VV               = _mm_macc_pd(vfeps,Fp,Y);
 528             vvdw12           = _mm_mul_pd(c12_00,VV);
 529             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 530             fvdw12           = _mm_mul_pd(c12_00,FF);
 531             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 532             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 533
 534             /* Update potential sum for this i atom from the interaction with this j atom. */
 535             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 536             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 537
 538             fscal            = fvdw;
 539
 540             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 541
 542             /* Update vectorial force */
 543             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 544             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 545             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 546
 547             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 548             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 549             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 550
 551             /**************************
 552              * CALCULATE INTERACTIONS *
 553              **************************/
 554
 555             r10              = _mm_mul_pd(rsq10,rinv10);
 556
 557             /* Compute parameters for interactions between i and j atoms */
 558             qq10             = _mm_mul_pd(iq1,jq0);
 559
 560             /* Calculate table index by multiplying r with table scale and truncate to integer */
 561             rt               = _mm_mul_pd(r10,vftabscale);
 562             vfitab           = _mm_cvttpd_epi32(rt);
 563 #ifdef __XOP__
 564             vfeps            = _mm_frcz_pd(rt);
 565 #else
 566             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 567 #endif
 568             twovfeps         = _mm_add_pd(vfeps,vfeps);
 569             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 570
 571             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 572             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 573             F                = _mm_setzero_pd();
 574             GMX_MM_TRANSPOSE2_PD(Y,F);
 575             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 576             H                = _mm_setzero_pd();
 577             GMX_MM_TRANSPOSE2_PD(G,H);
 578             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 579             VV               = _mm_macc_pd(vfeps,Fp,Y);
 580             velec            = _mm_mul_pd(qq10,VV);
 581             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 582             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 583
 584             /* Update potential sum for this i atom from the interaction with this j atom. */
 585             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 586             velecsum         = _mm_add_pd(velecsum,velec);
 587
 588             fscal            = felec;
 589
 590             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 591
 592             /* Update vectorial force */
 593             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 594             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 595             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 596
 597             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 598             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 599             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 600
 601             /**************************
 602              * CALCULATE INTERACTIONS *
 603              **************************/
 604
 605             r20              = _mm_mul_pd(rsq20,rinv20);
 606
 607             /* Compute parameters for interactions between i and j atoms */
 608             qq20             = _mm_mul_pd(iq2,jq0);
 609
 610             /* Calculate table index by multiplying r with table scale and truncate to integer */
 611             rt               = _mm_mul_pd(r20,vftabscale);
 612             vfitab           = _mm_cvttpd_epi32(rt);
 613 #ifdef __XOP__
 614             vfeps            = _mm_frcz_pd(rt);
 615 #else
 616             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 617 #endif
 618             twovfeps         = _mm_add_pd(vfeps,vfeps);
 619             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 620
 621             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 622             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 623             F                = _mm_setzero_pd();
 624             GMX_MM_TRANSPOSE2_PD(Y,F);
 625             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 626             H                = _mm_setzero_pd();
 627             GMX_MM_TRANSPOSE2_PD(G,H);
 628             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 629             VV               = _mm_macc_pd(vfeps,Fp,Y);
 630             velec            = _mm_mul_pd(qq20,VV);
 631             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 632             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 633
 634             /* Update potential sum for this i atom from the interaction with this j atom. */
 635             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 636             velecsum         = _mm_add_pd(velecsum,velec);
 637
 638             fscal            = felec;
 639
 640             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 641
 642             /* Update vectorial force */
 643             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 644             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 645             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 646
 647             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 648             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 649             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 650
 651             /**************************
 652              * CALCULATE INTERACTIONS *
 653              **************************/
 654
 655             r30              = _mm_mul_pd(rsq30,rinv30);
 656
 657             /* Compute parameters for interactions between i and j atoms */
 658             qq30             = _mm_mul_pd(iq3,jq0);
 659
 660             /* Calculate table index by multiplying r with table scale and truncate to integer */
 661             rt               = _mm_mul_pd(r30,vftabscale);
 662             vfitab           = _mm_cvttpd_epi32(rt);
 663 #ifdef __XOP__
 664             vfeps            = _mm_frcz_pd(rt);
 665 #else
 666             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 667 #endif
 668             twovfeps         = _mm_add_pd(vfeps,vfeps);
 669             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 670
 671             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 672             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 673             F                = _mm_setzero_pd();
 674             GMX_MM_TRANSPOSE2_PD(Y,F);
 675             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 676             H                = _mm_setzero_pd();
 677             GMX_MM_TRANSPOSE2_PD(G,H);
 678             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 679             VV               = _mm_macc_pd(vfeps,Fp,Y);
 680             velec            = _mm_mul_pd(qq30,VV);
 681             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 682             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 683
 684             /* Update potential sum for this i atom from the interaction with this j atom. */
 685             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 686             velecsum         = _mm_add_pd(velecsum,velec);
 687
 688             fscal            = felec;
 689
 690             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 691
 692             /* Update vectorial force */
 693             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 694             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 695             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 696
 697             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 698             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 699             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 700
 701             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 702
 703             /* Inner loop uses 200 flops */
 704         }
 705
 706         /* End of innermost loop */
 707
 708         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 709                                               f+i_coord_offset,fshift+i_shift_offset);
 710
 711         ggid                        = gid[iidx];
 712         /* Update potential energies */
 713         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 714         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 715
 716         /* Increment number of inner iterations */
 717         inneriter                  += j_index_end - j_index_start;
 718
 719         /* Outer loop uses 26 flops */
 720     }
 721
 722     /* Increment number of outer iterations */
 723     outeriter        += nri;
 724
 725     /* Update outer/inner flops */
 726
 727     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*200);
 728 }
 729 /*
 730  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_avx_128_fma_double
 731  * Electrostatics interaction: CubicSplineTable
 732  * VdW interaction:            CubicSplineTable
 733  * Geometry:                   Water4-Particle
 734  * Calculate force/pot:        Force
 735  */
 736 void
 737 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_avx_128_fma_double
 738                     (t_nblist                    * gmx_restrict       nlist,
 739                      rvec                        * gmx_restrict          xx,
 740                      rvec                        * gmx_restrict          ff,
 741                      t_forcerec                  * gmx_restrict          fr,
 742                      t_mdatoms                   * gmx_restrict     mdatoms,
 743                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 744                      t_nrnb                      * gmx_restrict        nrnb)
 745 {
 746     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 747      * just 0 for non-waters.
 748      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 749      * jnr indices corresponding to data put in the four positions in the SIMD register.
 750      */
 751     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 752     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 753     int              jnrA,jnrB;
 754     int              j_coord_offsetA,j_coord_offsetB;
 755     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 756     real             rcutoff_scalar;
 757     real             *shiftvec,*fshift,*x,*f;
 758     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 759     int              vdwioffset0;
 760     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 761     int              vdwioffset1;
 762     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 763     int              vdwioffset2;
 764     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 765     int              vdwioffset3;
 766     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 767     int              vdwjidx0A,vdwjidx0B;
 768     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 769     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 770     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 771     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 772     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 773     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 774     real             *charge;
 775     int              nvdwtype;
 776     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 777     int              *vdwtype;
 778     real             *vdwparam;
 779     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 780     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 781     __m128i          vfitab;
 782     __m128i          ifour       = _mm_set1_epi32(4);
 783     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 784     real             *vftab;
 785     __m128d          dummy_mask,cutoff_mask;
 786     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 787     __m128d          one     = _mm_set1_pd(1.0);
 788     __m128d          two     = _mm_set1_pd(2.0);
 789     x                = xx[0];
 790     f                = ff[0];
 791
 792     nri              = nlist->nri;
 793     iinr             = nlist->iinr;
 794     jindex           = nlist->jindex;
 795     jjnr             = nlist->jjnr;
 796     shiftidx         = nlist->shift;
 797     gid              = nlist->gid;
 798     shiftvec         = fr->shift_vec[0];
 799     fshift           = fr->fshift[0];
 800     facel            = _mm_set1_pd(fr->epsfac);
 801     charge           = mdatoms->chargeA;
 802     nvdwtype         = fr->ntype;
 803     vdwparam         = fr->nbfp;
 804     vdwtype          = mdatoms->typeA;
 805
 806     vftab            = kernel_data->table_elec_vdw->data;
 807     vftabscale       = _mm_set1_pd(kernel_data->table_elec_vdw->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             /* Load parameters for j particles */
 895             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 896             vdwjidx0A        = 2*vdwtype[jnrA+0];
 897             vdwjidx0B        = 2*vdwtype[jnrB+0];
 898
 899             fjx0             = _mm_setzero_pd();
 900             fjy0             = _mm_setzero_pd();
 901             fjz0             = _mm_setzero_pd();
 902
 903             /**************************
 904              * CALCULATE INTERACTIONS *
 905              **************************/
 906
 907             r00              = _mm_mul_pd(rsq00,rinv00);
 908
 909             /* Compute parameters for interactions between i and j atoms */
 910             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 911                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 912
 913             /* Calculate table index by multiplying r with table scale and truncate to integer */
 914             rt               = _mm_mul_pd(r00,vftabscale);
 915             vfitab           = _mm_cvttpd_epi32(rt);
 916 #ifdef __XOP__
 917             vfeps            = _mm_frcz_pd(rt);
 918 #else
 919             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 920 #endif
 921             twovfeps         = _mm_add_pd(vfeps,vfeps);
 922             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 923
 924             /* CUBIC SPLINE TABLE DISPERSION */
 925             vfitab           = _mm_add_epi32(vfitab,ifour);
 926             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 927             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 928             GMX_MM_TRANSPOSE2_PD(Y,F);
 929             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 930             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 931             GMX_MM_TRANSPOSE2_PD(G,H);
 932             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 933             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 934             fvdw6            = _mm_mul_pd(c6_00,FF);
 935
 936             /* CUBIC SPLINE TABLE REPULSION */
 937             vfitab           = _mm_add_epi32(vfitab,ifour);
 938             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 939             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 940             GMX_MM_TRANSPOSE2_PD(Y,F);
 941             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 942             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 943             GMX_MM_TRANSPOSE2_PD(G,H);
 944             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 945             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 946             fvdw12           = _mm_mul_pd(c12_00,FF);
 947             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 948
 949             fscal            = fvdw;
 950
 951             /* Update vectorial force */
 952             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 953             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 954             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 955
 956             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 957             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 958             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 959
 960             /**************************
 961              * CALCULATE INTERACTIONS *
 962              **************************/
 963
 964             r10              = _mm_mul_pd(rsq10,rinv10);
 965
 966             /* Compute parameters for interactions between i and j atoms */
 967             qq10             = _mm_mul_pd(iq1,jq0);
 968
 969             /* Calculate table index by multiplying r with table scale and truncate to integer */
 970             rt               = _mm_mul_pd(r10,vftabscale);
 971             vfitab           = _mm_cvttpd_epi32(rt);
 972 #ifdef __XOP__
 973             vfeps            = _mm_frcz_pd(rt);
 974 #else
 975             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 976 #endif
 977             twovfeps         = _mm_add_pd(vfeps,vfeps);
 978             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 979
 980             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 981             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 982             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 983             GMX_MM_TRANSPOSE2_PD(Y,F);
 984             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 985             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 986             GMX_MM_TRANSPOSE2_PD(G,H);
 987             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 988             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 989             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 990
 991             fscal            = felec;
 992
 993             /* Update vectorial force */
 994             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 995             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 996             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 997
 998             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 999             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1000             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1001
1002             /**************************
1003              * CALCULATE INTERACTIONS *
1004              **************************/
1005
1006             r20              = _mm_mul_pd(rsq20,rinv20);
1007
1008             /* Compute parameters for interactions between i and j atoms */
1009             qq20             = _mm_mul_pd(iq2,jq0);
1010
1011             /* Calculate table index by multiplying r with table scale and truncate to integer */
1012             rt               = _mm_mul_pd(r20,vftabscale);
1013             vfitab           = _mm_cvttpd_epi32(rt);
1014 #ifdef __XOP__
1015             vfeps            = _mm_frcz_pd(rt);
1016 #else
1017             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1018 #endif
1019             twovfeps         = _mm_add_pd(vfeps,vfeps);
1020             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1021
1022             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1023             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1024             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1025             GMX_MM_TRANSPOSE2_PD(Y,F);
1026             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1027             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1028             GMX_MM_TRANSPOSE2_PD(G,H);
1029             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1030             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1031             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1032
1033             fscal            = felec;
1034
1035             /* Update vectorial force */
1036             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1037             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1038             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1039
1040             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1041             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1042             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1043
1044             /**************************
1045              * CALCULATE INTERACTIONS *
1046              **************************/
1047
1048             r30              = _mm_mul_pd(rsq30,rinv30);
1049
1050             /* Compute parameters for interactions between i and j atoms */
1051             qq30             = _mm_mul_pd(iq3,jq0);
1052
1053             /* Calculate table index by multiplying r with table scale and truncate to integer */
1054             rt               = _mm_mul_pd(r30,vftabscale);
1055             vfitab           = _mm_cvttpd_epi32(rt);
1056 #ifdef __XOP__
1057             vfeps            = _mm_frcz_pd(rt);
1058 #else
1059             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1060 #endif
1061             twovfeps         = _mm_add_pd(vfeps,vfeps);
1062             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1063
1064             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1065             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1066             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1067             GMX_MM_TRANSPOSE2_PD(Y,F);
1068             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1069             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1070             GMX_MM_TRANSPOSE2_PD(G,H);
1071             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1072             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1073             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
1074
1075             fscal            = felec;
1076
1077             /* Update vectorial force */
1078             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1079             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1080             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1081
1082             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1083             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1084             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1085
1086             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1087
1088             /* Inner loop uses 180 flops */
1089         }
1090
1091         if(jidx<j_index_end)
1092         {
1093
1094             jnrA             = jjnr[jidx];
1095             j_coord_offsetA  = DIM*jnrA;
1096
1097             /* load j atom coordinates */
1098             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1099                                               &jx0,&jy0,&jz0);
1100
1101             /* Calculate displacement vector */
1102             dx00             = _mm_sub_pd(ix0,jx0);
1103             dy00             = _mm_sub_pd(iy0,jy0);
1104             dz00             = _mm_sub_pd(iz0,jz0);
1105             dx10             = _mm_sub_pd(ix1,jx0);
1106             dy10             = _mm_sub_pd(iy1,jy0);
1107             dz10             = _mm_sub_pd(iz1,jz0);
1108             dx20             = _mm_sub_pd(ix2,jx0);
1109             dy20             = _mm_sub_pd(iy2,jy0);
1110             dz20             = _mm_sub_pd(iz2,jz0);
1111             dx30             = _mm_sub_pd(ix3,jx0);
1112             dy30             = _mm_sub_pd(iy3,jy0);
1113             dz30             = _mm_sub_pd(iz3,jz0);
1114
1115             /* Calculate squared distance and things based on it */
1116             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1117             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1118             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1119             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1120
1121             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1122             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1123             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1124             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1125
1126             /* Load parameters for j particles */
1127             jq0              = _mm_load_sd(charge+jnrA+0);
1128             vdwjidx0A        = 2*vdwtype[jnrA+0];
1129
1130             fjx0             = _mm_setzero_pd();
1131             fjy0             = _mm_setzero_pd();
1132             fjz0             = _mm_setzero_pd();
1133
1134             /**************************
1135              * CALCULATE INTERACTIONS *
1136              **************************/
1137
1138             r00              = _mm_mul_pd(rsq00,rinv00);
1139
1140             /* Compute parameters for interactions between i and j atoms */
1141             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1142
1143             /* Calculate table index by multiplying r with table scale and truncate to integer */
1144             rt               = _mm_mul_pd(r00,vftabscale);
1145             vfitab           = _mm_cvttpd_epi32(rt);
1146 #ifdef __XOP__
1147             vfeps            = _mm_frcz_pd(rt);
1148 #else
1149             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1150 #endif
1151             twovfeps         = _mm_add_pd(vfeps,vfeps);
1152             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1153
1154             /* CUBIC SPLINE TABLE DISPERSION */
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             fvdw6            = _mm_mul_pd(c6_00,FF);
1165
1166             /* CUBIC SPLINE TABLE REPULSION */
1167             vfitab           = _mm_add_epi32(vfitab,ifour);
1168             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1169             F                = _mm_setzero_pd();
1170             GMX_MM_TRANSPOSE2_PD(Y,F);
1171             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1172             H                = _mm_setzero_pd();
1173             GMX_MM_TRANSPOSE2_PD(G,H);
1174             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1175             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1176             fvdw12           = _mm_mul_pd(c12_00,FF);
1177             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1178
1179             fscal            = fvdw;
1180
1181             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1182
1183             /* Update vectorial force */
1184             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1185             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1186             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1187
1188             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1189             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1190             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1191
1192             /**************************
1193              * CALCULATE INTERACTIONS *
1194              **************************/
1195
1196             r10              = _mm_mul_pd(rsq10,rinv10);
1197
1198             /* Compute parameters for interactions between i and j atoms */
1199             qq10             = _mm_mul_pd(iq1,jq0);
1200
1201             /* Calculate table index by multiplying r with table scale and truncate to integer */
1202             rt               = _mm_mul_pd(r10,vftabscale);
1203             vfitab           = _mm_cvttpd_epi32(rt);
1204 #ifdef __XOP__
1205             vfeps            = _mm_frcz_pd(rt);
1206 #else
1207             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1208 #endif
1209             twovfeps         = _mm_add_pd(vfeps,vfeps);
1210             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1211
1212             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1213             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1214             F                = _mm_setzero_pd();
1215             GMX_MM_TRANSPOSE2_PD(Y,F);
1216             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1217             H                = _mm_setzero_pd();
1218             GMX_MM_TRANSPOSE2_PD(G,H);
1219             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1220             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1221             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1222
1223             fscal            = felec;
1224
1225             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1226
1227             /* Update vectorial force */
1228             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1229             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1230             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1231
1232             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1233             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1234             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1235
1236             /**************************
1237              * CALCULATE INTERACTIONS *
1238              **************************/
1239
1240             r20              = _mm_mul_pd(rsq20,rinv20);
1241
1242             /* Compute parameters for interactions between i and j atoms */
1243             qq20             = _mm_mul_pd(iq2,jq0);
1244
1245             /* Calculate table index by multiplying r with table scale and truncate to integer */
1246             rt               = _mm_mul_pd(r20,vftabscale);
1247             vfitab           = _mm_cvttpd_epi32(rt);
1248 #ifdef __XOP__
1249             vfeps            = _mm_frcz_pd(rt);
1250 #else
1251             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1252 #endif
1253             twovfeps         = _mm_add_pd(vfeps,vfeps);
1254             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1255
1256             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1257             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1258             F                = _mm_setzero_pd();
1259             GMX_MM_TRANSPOSE2_PD(Y,F);
1260             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1261             H                = _mm_setzero_pd();
1262             GMX_MM_TRANSPOSE2_PD(G,H);
1263             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1264             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1265             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1266
1267             fscal            = felec;
1268
1269             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1270
1271             /* Update vectorial force */
1272             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1273             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1274             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1275
1276             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1277             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1278             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1279
1280             /**************************
1281              * CALCULATE INTERACTIONS *
1282              **************************/
1283
1284             r30              = _mm_mul_pd(rsq30,rinv30);
1285
1286             /* Compute parameters for interactions between i and j atoms */
1287             qq30             = _mm_mul_pd(iq3,jq0);
1288
1289             /* Calculate table index by multiplying r with table scale and truncate to integer */
1290             rt               = _mm_mul_pd(r30,vftabscale);
1291             vfitab           = _mm_cvttpd_epi32(rt);
1292 #ifdef __XOP__
1293             vfeps            = _mm_frcz_pd(rt);
1294 #else
1295             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1296 #endif
1297             twovfeps         = _mm_add_pd(vfeps,vfeps);
1298             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1299
1300             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1301             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1302             F                = _mm_setzero_pd();
1303             GMX_MM_TRANSPOSE2_PD(Y,F);
1304             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1305             H                = _mm_setzero_pd();
1306             GMX_MM_TRANSPOSE2_PD(G,H);
1307             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1308             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1309             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
1310
1311             fscal            = felec;
1312
1313             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1314
1315             /* Update vectorial force */
1316             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1317             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1318             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1319
1320             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1321             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1322             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1323
1324             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1325
1326             /* Inner loop uses 180 flops */
1327         }
1328
1329         /* End of innermost loop */
1330
1331         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1332                                               f+i_coord_offset,fshift+i_shift_offset);
1333
1334         /* Increment number of inner iterations */
1335         inneriter                  += j_index_end - j_index_start;
1336
1337         /* Outer loop uses 24 flops */
1338     }
1339
1340     /* Increment number of outer iterations */
1341     outeriter        += nri;
1342
1343     /* Update outer/inner flops */
1344
1345     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*180);
1346 }