src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRF_VdwCSTab_GeomW3P1_avx_128_fma_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_128_fma_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_avx_128_fma_double
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_avx_128_fma_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwioffset1;
  82     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  83     int              vdwioffset2;
  84     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  85     int              vdwjidx0A,vdwjidx0B;
  86     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  89     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  90     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     int              nvdwtype;
  93     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  97     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  98     __m128i          vfitab;
  99     __m128i          ifour       = _mm_set1_epi32(4);
 100     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 101     real             *vftab;
 102     __m128d          dummy_mask,cutoff_mask;
 103     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 104     __m128d          one     = _mm_set1_pd(1.0);
 105     __m128d          two     = _mm_set1_pd(2.0);
 106     x                = xx[0];
 107     f                = ff[0];
 108
 109     nri              = nlist->nri;
 110     iinr             = nlist->iinr;
 111     jindex           = nlist->jindex;
 112     jjnr             = nlist->jjnr;
 113     shiftidx         = nlist->shift;
 114     gid              = nlist->gid;
 115     shiftvec         = fr->shift_vec[0];
 116     fshift           = fr->fshift[0];
 117     facel            = _mm_set1_pd(fr->ic->epsfac);
 118     charge           = mdatoms->chargeA;
 119     krf              = _mm_set1_pd(fr->ic->k_rf);
 120     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 121     crf              = _mm_set1_pd(fr->ic->c_rf);
 122     nvdwtype         = fr->ntype;
 123     vdwparam         = fr->nbfp;
 124     vdwtype          = mdatoms->typeA;
 125
 126     vftab            = kernel_data->table_vdw->data;
 127     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 128
 129     /* Setup water-specific parameters */
 130     inr              = nlist->iinr[0];
 131     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 132     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 133     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 134     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 135
 136     /* Avoid stupid compiler warnings */
 137     jnrA = jnrB = 0;
 138     j_coord_offsetA = 0;
 139     j_coord_offsetB = 0;
 140
 141     outeriter        = 0;
 142     inneriter        = 0;
 143
 144     /* Start outer loop over neighborlists */
 145     for(iidx=0; iidx<nri; iidx++)
 146     {
 147         /* Load shift vector for this list */
 148         i_shift_offset   = DIM*shiftidx[iidx];
 149
 150         /* Load limits for loop over neighbors */
 151         j_index_start    = jindex[iidx];
 152         j_index_end      = jindex[iidx+1];
 153
 154         /* Get outer coordinate index */
 155         inr              = iinr[iidx];
 156         i_coord_offset   = DIM*inr;
 157
 158         /* Load i particle coords and add shift vector */
 159         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 160                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 161
 162         fix0             = _mm_setzero_pd();
 163         fiy0             = _mm_setzero_pd();
 164         fiz0             = _mm_setzero_pd();
 165         fix1             = _mm_setzero_pd();
 166         fiy1             = _mm_setzero_pd();
 167         fiz1             = _mm_setzero_pd();
 168         fix2             = _mm_setzero_pd();
 169         fiy2             = _mm_setzero_pd();
 170         fiz2             = _mm_setzero_pd();
 171
 172         /* Reset potential sums */
 173         velecsum         = _mm_setzero_pd();
 174         vvdwsum          = _mm_setzero_pd();
 175
 176         /* Start inner kernel loop */
 177         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 178         {
 179
 180             /* Get j neighbor index, and coordinate index */
 181             jnrA             = jjnr[jidx];
 182             jnrB             = jjnr[jidx+1];
 183             j_coord_offsetA  = DIM*jnrA;
 184             j_coord_offsetB  = DIM*jnrB;
 185
 186             /* load j atom coordinates */
 187             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 188                                               &jx0,&jy0,&jz0);
 189
 190             /* Calculate displacement vector */
 191             dx00             = _mm_sub_pd(ix0,jx0);
 192             dy00             = _mm_sub_pd(iy0,jy0);
 193             dz00             = _mm_sub_pd(iz0,jz0);
 194             dx10             = _mm_sub_pd(ix1,jx0);
 195             dy10             = _mm_sub_pd(iy1,jy0);
 196             dz10             = _mm_sub_pd(iz1,jz0);
 197             dx20             = _mm_sub_pd(ix2,jx0);
 198             dy20             = _mm_sub_pd(iy2,jy0);
 199             dz20             = _mm_sub_pd(iz2,jz0);
 200
 201             /* Calculate squared distance and things based on it */
 202             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 203             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 204             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 205
 206             rinv00           = avx128fma_invsqrt_d(rsq00);
 207             rinv10           = avx128fma_invsqrt_d(rsq10);
 208             rinv20           = avx128fma_invsqrt_d(rsq20);
 209
 210             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 211             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 212             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 213
 214             /* Load parameters for j particles */
 215             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 216             vdwjidx0A        = 2*vdwtype[jnrA+0];
 217             vdwjidx0B        = 2*vdwtype[jnrB+0];
 218
 219             fjx0             = _mm_setzero_pd();
 220             fjy0             = _mm_setzero_pd();
 221             fjz0             = _mm_setzero_pd();
 222
 223             /**************************
 224              * CALCULATE INTERACTIONS *
 225              **************************/
 226
 227             r00              = _mm_mul_pd(rsq00,rinv00);
 228
 229             /* Compute parameters for interactions between i and j atoms */
 230             qq00             = _mm_mul_pd(iq0,jq0);
 231             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 232                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 233
 234             /* Calculate table index by multiplying r with table scale and truncate to integer */
 235             rt               = _mm_mul_pd(r00,vftabscale);
 236             vfitab           = _mm_cvttpd_epi32(rt);
 237 #ifdef __XOP__
 238             vfeps            = _mm_frcz_pd(rt);
 239 #else
 240             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 241 #endif
 242             twovfeps         = _mm_add_pd(vfeps,vfeps);
 243             vfitab           = _mm_slli_epi32(vfitab,3);
 244
 245             /* REACTION-FIELD ELECTROSTATICS */
 246             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 247             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 248
 249             /* CUBIC SPLINE TABLE DISPERSION */
 250             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 251             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 252             GMX_MM_TRANSPOSE2_PD(Y,F);
 253             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 254             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 255             GMX_MM_TRANSPOSE2_PD(G,H);
 256             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 257             VV               = _mm_macc_pd(vfeps,Fp,Y);
 258             vvdw6            = _mm_mul_pd(c6_00,VV);
 259             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 260             fvdw6            = _mm_mul_pd(c6_00,FF);
 261
 262             /* CUBIC SPLINE TABLE REPULSION */
 263             vfitab           = _mm_add_epi32(vfitab,ifour);
 264             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 265             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 266             GMX_MM_TRANSPOSE2_PD(Y,F);
 267             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 268             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 269             GMX_MM_TRANSPOSE2_PD(G,H);
 270             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 271             VV               = _mm_macc_pd(vfeps,Fp,Y);
 272             vvdw12           = _mm_mul_pd(c12_00,VV);
 273             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 274             fvdw12           = _mm_mul_pd(c12_00,FF);
 275             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 276             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 277
 278             /* Update potential sum for this i atom from the interaction with this j atom. */
 279             velecsum         = _mm_add_pd(velecsum,velec);
 280             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 281
 282             fscal            = _mm_add_pd(felec,fvdw);
 283
 284             /* Update vectorial force */
 285             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 286             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 287             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 288
 289             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 290             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 291             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 292
 293             /**************************
 294              * CALCULATE INTERACTIONS *
 295              **************************/
 296
 297             /* Compute parameters for interactions between i and j atoms */
 298             qq10             = _mm_mul_pd(iq1,jq0);
 299
 300             /* REACTION-FIELD ELECTROSTATICS */
 301             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 302             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 303
 304             /* Update potential sum for this i atom from the interaction with this j atom. */
 305             velecsum         = _mm_add_pd(velecsum,velec);
 306
 307             fscal            = felec;
 308
 309             /* Update vectorial force */
 310             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 311             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 312             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 313
 314             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 315             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 316             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 317
 318             /**************************
 319              * CALCULATE INTERACTIONS *
 320              **************************/
 321
 322             /* Compute parameters for interactions between i and j atoms */
 323             qq20             = _mm_mul_pd(iq2,jq0);
 324
 325             /* REACTION-FIELD ELECTROSTATICS */
 326             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 327             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 328
 329             /* Update potential sum for this i atom from the interaction with this j atom. */
 330             velecsum         = _mm_add_pd(velecsum,velec);
 331
 332             fscal            = felec;
 333
 334             /* Update vectorial force */
 335             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 336             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 337             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 338
 339             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 340             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 341             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 342
 343             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 344
 345             /* Inner loop uses 143 flops */
 346         }
 347
 348         if(jidx<j_index_end)
 349         {
 350
 351             jnrA             = jjnr[jidx];
 352             j_coord_offsetA  = DIM*jnrA;
 353
 354             /* load j atom coordinates */
 355             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 356                                               &jx0,&jy0,&jz0);
 357
 358             /* Calculate displacement vector */
 359             dx00             = _mm_sub_pd(ix0,jx0);
 360             dy00             = _mm_sub_pd(iy0,jy0);
 361             dz00             = _mm_sub_pd(iz0,jz0);
 362             dx10             = _mm_sub_pd(ix1,jx0);
 363             dy10             = _mm_sub_pd(iy1,jy0);
 364             dz10             = _mm_sub_pd(iz1,jz0);
 365             dx20             = _mm_sub_pd(ix2,jx0);
 366             dy20             = _mm_sub_pd(iy2,jy0);
 367             dz20             = _mm_sub_pd(iz2,jz0);
 368
 369             /* Calculate squared distance and things based on it */
 370             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 371             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 372             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 373
 374             rinv00           = avx128fma_invsqrt_d(rsq00);
 375             rinv10           = avx128fma_invsqrt_d(rsq10);
 376             rinv20           = avx128fma_invsqrt_d(rsq20);
 377
 378             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 379             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 380             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 381
 382             /* Load parameters for j particles */
 383             jq0              = _mm_load_sd(charge+jnrA+0);
 384             vdwjidx0A        = 2*vdwtype[jnrA+0];
 385
 386             fjx0             = _mm_setzero_pd();
 387             fjy0             = _mm_setzero_pd();
 388             fjz0             = _mm_setzero_pd();
 389
 390             /**************************
 391              * CALCULATE INTERACTIONS *
 392              **************************/
 393
 394             r00              = _mm_mul_pd(rsq00,rinv00);
 395
 396             /* Compute parameters for interactions between i and j atoms */
 397             qq00             = _mm_mul_pd(iq0,jq0);
 398             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 399
 400             /* Calculate table index by multiplying r with table scale and truncate to integer */
 401             rt               = _mm_mul_pd(r00,vftabscale);
 402             vfitab           = _mm_cvttpd_epi32(rt);
 403 #ifdef __XOP__
 404             vfeps            = _mm_frcz_pd(rt);
 405 #else
 406             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 407 #endif
 408             twovfeps         = _mm_add_pd(vfeps,vfeps);
 409             vfitab           = _mm_slli_epi32(vfitab,3);
 410
 411             /* REACTION-FIELD ELECTROSTATICS */
 412             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 413             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 414
 415             /* CUBIC SPLINE TABLE DISPERSION */
 416             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 417             F                = _mm_setzero_pd();
 418             GMX_MM_TRANSPOSE2_PD(Y,F);
 419             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 420             H                = _mm_setzero_pd();
 421             GMX_MM_TRANSPOSE2_PD(G,H);
 422             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 423             VV               = _mm_macc_pd(vfeps,Fp,Y);
 424             vvdw6            = _mm_mul_pd(c6_00,VV);
 425             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 426             fvdw6            = _mm_mul_pd(c6_00,FF);
 427
 428             /* CUBIC SPLINE TABLE REPULSION */
 429             vfitab           = _mm_add_epi32(vfitab,ifour);
 430             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 431             F                = _mm_setzero_pd();
 432             GMX_MM_TRANSPOSE2_PD(Y,F);
 433             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 434             H                = _mm_setzero_pd();
 435             GMX_MM_TRANSPOSE2_PD(G,H);
 436             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 437             VV               = _mm_macc_pd(vfeps,Fp,Y);
 438             vvdw12           = _mm_mul_pd(c12_00,VV);
 439             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 440             fvdw12           = _mm_mul_pd(c12_00,FF);
 441             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 442             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 443
 444             /* Update potential sum for this i atom from the interaction with this j atom. */
 445             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 446             velecsum         = _mm_add_pd(velecsum,velec);
 447             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 448             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 449
 450             fscal            = _mm_add_pd(felec,fvdw);
 451
 452             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 453
 454             /* Update vectorial force */
 455             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 456             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 457             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 458
 459             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 460             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 461             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 462
 463             /**************************
 464              * CALCULATE INTERACTIONS *
 465              **************************/
 466
 467             /* Compute parameters for interactions between i and j atoms */
 468             qq10             = _mm_mul_pd(iq1,jq0);
 469
 470             /* REACTION-FIELD ELECTROSTATICS */
 471             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 472             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 473
 474             /* Update potential sum for this i atom from the interaction with this j atom. */
 475             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 476             velecsum         = _mm_add_pd(velecsum,velec);
 477
 478             fscal            = felec;
 479
 480             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 481
 482             /* Update vectorial force */
 483             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 484             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 485             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 486
 487             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 488             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 489             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 490
 491             /**************************
 492              * CALCULATE INTERACTIONS *
 493              **************************/
 494
 495             /* Compute parameters for interactions between i and j atoms */
 496             qq20             = _mm_mul_pd(iq2,jq0);
 497
 498             /* REACTION-FIELD ELECTROSTATICS */
 499             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 500             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 501
 502             /* Update potential sum for this i atom from the interaction with this j atom. */
 503             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 504             velecsum         = _mm_add_pd(velecsum,velec);
 505
 506             fscal            = felec;
 507
 508             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 509
 510             /* Update vectorial force */
 511             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 512             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 513             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 514
 515             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 516             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 517             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 518
 519             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 520
 521             /* Inner loop uses 143 flops */
 522         }
 523
 524         /* End of innermost loop */
 525
 526         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 527                                               f+i_coord_offset,fshift+i_shift_offset);
 528
 529         ggid                        = gid[iidx];
 530         /* Update potential energies */
 531         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 532         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 533
 534         /* Increment number of inner iterations */
 535         inneriter                  += j_index_end - j_index_start;
 536
 537         /* Outer loop uses 20 flops */
 538     }
 539
 540     /* Increment number of outer iterations */
 541     outeriter        += nri;
 542
 543     /* Update outer/inner flops */
 544
 545     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*143);
 546 }
 547 /*
 548  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_avx_128_fma_double
 549  * Electrostatics interaction: ReactionField
 550  * VdW interaction:            CubicSplineTable
 551  * Geometry:                   Water3-Particle
 552  * Calculate force/pot:        Force
 553  */
 554 void
 555 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_avx_128_fma_double
 556                     (t_nblist                    * gmx_restrict       nlist,
 557                      rvec                        * gmx_restrict          xx,
 558                      rvec                        * gmx_restrict          ff,
 559                      struct t_forcerec           * gmx_restrict          fr,
 560                      t_mdatoms                   * gmx_restrict     mdatoms,
 561                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 562                      t_nrnb                      * gmx_restrict        nrnb)
 563 {
 564     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 565      * just 0 for non-waters.
 566      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 567      * jnr indices corresponding to data put in the four positions in the SIMD register.
 568      */
 569     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 570     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 571     int              jnrA,jnrB;
 572     int              j_coord_offsetA,j_coord_offsetB;
 573     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 574     real             rcutoff_scalar;
 575     real             *shiftvec,*fshift,*x,*f;
 576     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 577     int              vdwioffset0;
 578     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 579     int              vdwioffset1;
 580     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 581     int              vdwioffset2;
 582     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 583     int              vdwjidx0A,vdwjidx0B;
 584     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 585     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 586     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 587     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 588     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 589     real             *charge;
 590     int              nvdwtype;
 591     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 592     int              *vdwtype;
 593     real             *vdwparam;
 594     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 595     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 596     __m128i          vfitab;
 597     __m128i          ifour       = _mm_set1_epi32(4);
 598     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 599     real             *vftab;
 600     __m128d          dummy_mask,cutoff_mask;
 601     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 602     __m128d          one     = _mm_set1_pd(1.0);
 603     __m128d          two     = _mm_set1_pd(2.0);
 604     x                = xx[0];
 605     f                = ff[0];
 606
 607     nri              = nlist->nri;
 608     iinr             = nlist->iinr;
 609     jindex           = nlist->jindex;
 610     jjnr             = nlist->jjnr;
 611     shiftidx         = nlist->shift;
 612     gid              = nlist->gid;
 613     shiftvec         = fr->shift_vec[0];
 614     fshift           = fr->fshift[0];
 615     facel            = _mm_set1_pd(fr->ic->epsfac);
 616     charge           = mdatoms->chargeA;
 617     krf              = _mm_set1_pd(fr->ic->k_rf);
 618     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 619     crf              = _mm_set1_pd(fr->ic->c_rf);
 620     nvdwtype         = fr->ntype;
 621     vdwparam         = fr->nbfp;
 622     vdwtype          = mdatoms->typeA;
 623
 624     vftab            = kernel_data->table_vdw->data;
 625     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 626
 627     /* Setup water-specific parameters */
 628     inr              = nlist->iinr[0];
 629     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 630     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 631     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 632     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 633
 634     /* Avoid stupid compiler warnings */
 635     jnrA = jnrB = 0;
 636     j_coord_offsetA = 0;
 637     j_coord_offsetB = 0;
 638
 639     outeriter        = 0;
 640     inneriter        = 0;
 641
 642     /* Start outer loop over neighborlists */
 643     for(iidx=0; iidx<nri; iidx++)
 644     {
 645         /* Load shift vector for this list */
 646         i_shift_offset   = DIM*shiftidx[iidx];
 647
 648         /* Load limits for loop over neighbors */
 649         j_index_start    = jindex[iidx];
 650         j_index_end      = jindex[iidx+1];
 651
 652         /* Get outer coordinate index */
 653         inr              = iinr[iidx];
 654         i_coord_offset   = DIM*inr;
 655
 656         /* Load i particle coords and add shift vector */
 657         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 658                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 659
 660         fix0             = _mm_setzero_pd();
 661         fiy0             = _mm_setzero_pd();
 662         fiz0             = _mm_setzero_pd();
 663         fix1             = _mm_setzero_pd();
 664         fiy1             = _mm_setzero_pd();
 665         fiz1             = _mm_setzero_pd();
 666         fix2             = _mm_setzero_pd();
 667         fiy2             = _mm_setzero_pd();
 668         fiz2             = _mm_setzero_pd();
 669
 670         /* Start inner kernel loop */
 671         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 672         {
 673
 674             /* Get j neighbor index, and coordinate index */
 675             jnrA             = jjnr[jidx];
 676             jnrB             = jjnr[jidx+1];
 677             j_coord_offsetA  = DIM*jnrA;
 678             j_coord_offsetB  = DIM*jnrB;
 679
 680             /* load j atom coordinates */
 681             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 682                                               &jx0,&jy0,&jz0);
 683
 684             /* Calculate displacement vector */
 685             dx00             = _mm_sub_pd(ix0,jx0);
 686             dy00             = _mm_sub_pd(iy0,jy0);
 687             dz00             = _mm_sub_pd(iz0,jz0);
 688             dx10             = _mm_sub_pd(ix1,jx0);
 689             dy10             = _mm_sub_pd(iy1,jy0);
 690             dz10             = _mm_sub_pd(iz1,jz0);
 691             dx20             = _mm_sub_pd(ix2,jx0);
 692             dy20             = _mm_sub_pd(iy2,jy0);
 693             dz20             = _mm_sub_pd(iz2,jz0);
 694
 695             /* Calculate squared distance and things based on it */
 696             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 697             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 698             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 699
 700             rinv00           = avx128fma_invsqrt_d(rsq00);
 701             rinv10           = avx128fma_invsqrt_d(rsq10);
 702             rinv20           = avx128fma_invsqrt_d(rsq20);
 703
 704             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 705             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 706             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 707
 708             /* Load parameters for j particles */
 709             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 710             vdwjidx0A        = 2*vdwtype[jnrA+0];
 711             vdwjidx0B        = 2*vdwtype[jnrB+0];
 712
 713             fjx0             = _mm_setzero_pd();
 714             fjy0             = _mm_setzero_pd();
 715             fjz0             = _mm_setzero_pd();
 716
 717             /**************************
 718              * CALCULATE INTERACTIONS *
 719              **************************/
 720
 721             r00              = _mm_mul_pd(rsq00,rinv00);
 722
 723             /* Compute parameters for interactions between i and j atoms */
 724             qq00             = _mm_mul_pd(iq0,jq0);
 725             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 726                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 727
 728             /* Calculate table index by multiplying r with table scale and truncate to integer */
 729             rt               = _mm_mul_pd(r00,vftabscale);
 730             vfitab           = _mm_cvttpd_epi32(rt);
 731 #ifdef __XOP__
 732             vfeps            = _mm_frcz_pd(rt);
 733 #else
 734             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 735 #endif
 736             twovfeps         = _mm_add_pd(vfeps,vfeps);
 737             vfitab           = _mm_slli_epi32(vfitab,3);
 738
 739             /* REACTION-FIELD ELECTROSTATICS */
 740             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 741
 742             /* CUBIC SPLINE TABLE DISPERSION */
 743             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 744             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 745             GMX_MM_TRANSPOSE2_PD(Y,F);
 746             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 747             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 748             GMX_MM_TRANSPOSE2_PD(G,H);
 749             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 750             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 751             fvdw6            = _mm_mul_pd(c6_00,FF);
 752
 753             /* CUBIC SPLINE TABLE REPULSION */
 754             vfitab           = _mm_add_epi32(vfitab,ifour);
 755             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 756             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 757             GMX_MM_TRANSPOSE2_PD(Y,F);
 758             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 759             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 760             GMX_MM_TRANSPOSE2_PD(G,H);
 761             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 762             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 763             fvdw12           = _mm_mul_pd(c12_00,FF);
 764             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 765
 766             fscal            = _mm_add_pd(felec,fvdw);
 767
 768             /* Update vectorial force */
 769             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 770             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 771             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 772
 773             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 774             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 775             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 776
 777             /**************************
 778              * CALCULATE INTERACTIONS *
 779              **************************/
 780
 781             /* Compute parameters for interactions between i and j atoms */
 782             qq10             = _mm_mul_pd(iq1,jq0);
 783
 784             /* REACTION-FIELD ELECTROSTATICS */
 785             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 786
 787             fscal            = felec;
 788
 789             /* Update vectorial force */
 790             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 791             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 792             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 793
 794             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 795             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 796             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 797
 798             /**************************
 799              * CALCULATE INTERACTIONS *
 800              **************************/
 801
 802             /* Compute parameters for interactions between i and j atoms */
 803             qq20             = _mm_mul_pd(iq2,jq0);
 804
 805             /* REACTION-FIELD ELECTROSTATICS */
 806             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 807
 808             fscal            = felec;
 809
 810             /* Update vectorial force */
 811             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 812             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 813             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 814
 815             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 816             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 817             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 818
 819             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 820
 821             /* Inner loop uses 120 flops */
 822         }
 823
 824         if(jidx<j_index_end)
 825         {
 826
 827             jnrA             = jjnr[jidx];
 828             j_coord_offsetA  = DIM*jnrA;
 829
 830             /* load j atom coordinates */
 831             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 832                                               &jx0,&jy0,&jz0);
 833
 834             /* Calculate displacement vector */
 835             dx00             = _mm_sub_pd(ix0,jx0);
 836             dy00             = _mm_sub_pd(iy0,jy0);
 837             dz00             = _mm_sub_pd(iz0,jz0);
 838             dx10             = _mm_sub_pd(ix1,jx0);
 839             dy10             = _mm_sub_pd(iy1,jy0);
 840             dz10             = _mm_sub_pd(iz1,jz0);
 841             dx20             = _mm_sub_pd(ix2,jx0);
 842             dy20             = _mm_sub_pd(iy2,jy0);
 843             dz20             = _mm_sub_pd(iz2,jz0);
 844
 845             /* Calculate squared distance and things based on it */
 846             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 847             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 848             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 849
 850             rinv00           = avx128fma_invsqrt_d(rsq00);
 851             rinv10           = avx128fma_invsqrt_d(rsq10);
 852             rinv20           = avx128fma_invsqrt_d(rsq20);
 853
 854             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 855             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 856             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 857
 858             /* Load parameters for j particles */
 859             jq0              = _mm_load_sd(charge+jnrA+0);
 860             vdwjidx0A        = 2*vdwtype[jnrA+0];
 861
 862             fjx0             = _mm_setzero_pd();
 863             fjy0             = _mm_setzero_pd();
 864             fjz0             = _mm_setzero_pd();
 865
 866             /**************************
 867              * CALCULATE INTERACTIONS *
 868              **************************/
 869
 870             r00              = _mm_mul_pd(rsq00,rinv00);
 871
 872             /* Compute parameters for interactions between i and j atoms */
 873             qq00             = _mm_mul_pd(iq0,jq0);
 874             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 875
 876             /* Calculate table index by multiplying r with table scale and truncate to integer */
 877             rt               = _mm_mul_pd(r00,vftabscale);
 878             vfitab           = _mm_cvttpd_epi32(rt);
 879 #ifdef __XOP__
 880             vfeps            = _mm_frcz_pd(rt);
 881 #else
 882             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 883 #endif
 884             twovfeps         = _mm_add_pd(vfeps,vfeps);
 885             vfitab           = _mm_slli_epi32(vfitab,3);
 886
 887             /* REACTION-FIELD ELECTROSTATICS */
 888             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 889
 890             /* CUBIC SPLINE TABLE DISPERSION */
 891             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 892             F                = _mm_setzero_pd();
 893             GMX_MM_TRANSPOSE2_PD(Y,F);
 894             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 895             H                = _mm_setzero_pd();
 896             GMX_MM_TRANSPOSE2_PD(G,H);
 897             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 898             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 899             fvdw6            = _mm_mul_pd(c6_00,FF);
 900
 901             /* CUBIC SPLINE TABLE REPULSION */
 902             vfitab           = _mm_add_epi32(vfitab,ifour);
 903             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 904             F                = _mm_setzero_pd();
 905             GMX_MM_TRANSPOSE2_PD(Y,F);
 906             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 907             H                = _mm_setzero_pd();
 908             GMX_MM_TRANSPOSE2_PD(G,H);
 909             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 910             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 911             fvdw12           = _mm_mul_pd(c12_00,FF);
 912             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 913
 914             fscal            = _mm_add_pd(felec,fvdw);
 915
 916             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 917
 918             /* Update vectorial force */
 919             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 920             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 921             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 922
 923             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 924             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 925             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 926
 927             /**************************
 928              * CALCULATE INTERACTIONS *
 929              **************************/
 930
 931             /* Compute parameters for interactions between i and j atoms */
 932             qq10             = _mm_mul_pd(iq1,jq0);
 933
 934             /* REACTION-FIELD ELECTROSTATICS */
 935             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 936
 937             fscal            = felec;
 938
 939             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 940
 941             /* Update vectorial force */
 942             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 943             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 944             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 945
 946             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 947             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 948             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 949
 950             /**************************
 951              * CALCULATE INTERACTIONS *
 952              **************************/
 953
 954             /* Compute parameters for interactions between i and j atoms */
 955             qq20             = _mm_mul_pd(iq2,jq0);
 956
 957             /* REACTION-FIELD ELECTROSTATICS */
 958             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 959
 960             fscal            = felec;
 961
 962             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 963
 964             /* Update vectorial force */
 965             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 966             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 967             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 968
 969             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 970             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 971             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 972
 973             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 974
 975             /* Inner loop uses 120 flops */
 976         }
 977
 978         /* End of innermost loop */
 979
 980         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 981                                               f+i_coord_offset,fshift+i_shift_offset);
 982
 983         /* Increment number of inner iterations */
 984         inneriter                  += j_index_end - j_index_start;
 985
 986         /* Outer loop uses 18 flops */
 987     }
 988
 989     /* Increment number of outer iterations */
 990     outeriter        += nri;
 991
 992     /* Update outer/inner flops */
 993
 994     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*120);
 995 }