src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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_ElecCSTab_VdwLJ_GeomW3P1_VF_avx_128_fma_double
  51  * Electrostatics interaction: CubicSplineTable
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCSTab_VdwLJ_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     nvdwtype         = fr->ntype;
 120     vdwparam         = fr->nbfp;
 121     vdwtype          = mdatoms->typeA;
 122
 123     vftab            = kernel_data->table_elec->data;
 124     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 125
 126     /* Setup water-specific parameters */
 127     inr              = nlist->iinr[0];
 128     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 129     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 130     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 131     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 132
 133     /* Avoid stupid compiler warnings */
 134     jnrA = jnrB = 0;
 135     j_coord_offsetA = 0;
 136     j_coord_offsetB = 0;
 137
 138     outeriter        = 0;
 139     inneriter        = 0;
 140
 141     /* Start outer loop over neighborlists */
 142     for(iidx=0; iidx<nri; iidx++)
 143     {
 144         /* Load shift vector for this list */
 145         i_shift_offset   = DIM*shiftidx[iidx];
 146
 147         /* Load limits for loop over neighbors */
 148         j_index_start    = jindex[iidx];
 149         j_index_end      = jindex[iidx+1];
 150
 151         /* Get outer coordinate index */
 152         inr              = iinr[iidx];
 153         i_coord_offset   = DIM*inr;
 154
 155         /* Load i particle coords and add shift vector */
 156         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 157                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 158
 159         fix0             = _mm_setzero_pd();
 160         fiy0             = _mm_setzero_pd();
 161         fiz0             = _mm_setzero_pd();
 162         fix1             = _mm_setzero_pd();
 163         fiy1             = _mm_setzero_pd();
 164         fiz1             = _mm_setzero_pd();
 165         fix2             = _mm_setzero_pd();
 166         fiy2             = _mm_setzero_pd();
 167         fiz2             = _mm_setzero_pd();
 168
 169         /* Reset potential sums */
 170         velecsum         = _mm_setzero_pd();
 171         vvdwsum          = _mm_setzero_pd();
 172
 173         /* Start inner kernel loop */
 174         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 175         {
 176
 177             /* Get j neighbor index, and coordinate index */
 178             jnrA             = jjnr[jidx];
 179             jnrB             = jjnr[jidx+1];
 180             j_coord_offsetA  = DIM*jnrA;
 181             j_coord_offsetB  = DIM*jnrB;
 182
 183             /* load j atom coordinates */
 184             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 185                                               &jx0,&jy0,&jz0);
 186
 187             /* Calculate displacement vector */
 188             dx00             = _mm_sub_pd(ix0,jx0);
 189             dy00             = _mm_sub_pd(iy0,jy0);
 190             dz00             = _mm_sub_pd(iz0,jz0);
 191             dx10             = _mm_sub_pd(ix1,jx0);
 192             dy10             = _mm_sub_pd(iy1,jy0);
 193             dz10             = _mm_sub_pd(iz1,jz0);
 194             dx20             = _mm_sub_pd(ix2,jx0);
 195             dy20             = _mm_sub_pd(iy2,jy0);
 196             dz20             = _mm_sub_pd(iz2,jz0);
 197
 198             /* Calculate squared distance and things based on it */
 199             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 200             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 201             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 202
 203             rinv00           = avx128fma_invsqrt_d(rsq00);
 204             rinv10           = avx128fma_invsqrt_d(rsq10);
 205             rinv20           = avx128fma_invsqrt_d(rsq20);
 206
 207             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 208
 209             /* Load parameters for j particles */
 210             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 211             vdwjidx0A        = 2*vdwtype[jnrA+0];
 212             vdwjidx0B        = 2*vdwtype[jnrB+0];
 213
 214             fjx0             = _mm_setzero_pd();
 215             fjy0             = _mm_setzero_pd();
 216             fjz0             = _mm_setzero_pd();
 217
 218             /**************************
 219              * CALCULATE INTERACTIONS *
 220              **************************/
 221
 222             r00              = _mm_mul_pd(rsq00,rinv00);
 223
 224             /* Compute parameters for interactions between i and j atoms */
 225             qq00             = _mm_mul_pd(iq0,jq0);
 226             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 227                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 228
 229             /* Calculate table index by multiplying r with table scale and truncate to integer */
 230             rt               = _mm_mul_pd(r00,vftabscale);
 231             vfitab           = _mm_cvttpd_epi32(rt);
 232 #ifdef __XOP__
 233             vfeps            = _mm_frcz_pd(rt);
 234 #else
 235             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 236 #endif
 237             twovfeps         = _mm_add_pd(vfeps,vfeps);
 238             vfitab           = _mm_slli_epi32(vfitab,2);
 239
 240             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 241             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 242             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 243             GMX_MM_TRANSPOSE2_PD(Y,F);
 244             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 245             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 246             GMX_MM_TRANSPOSE2_PD(G,H);
 247             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 248             VV               = _mm_macc_pd(vfeps,Fp,Y);
 249             velec            = _mm_mul_pd(qq00,VV);
 250             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 251             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 252
 253             /* LENNARD-JONES DISPERSION/REPULSION */
 254
 255             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 256             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 257             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 258             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 259             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 260
 261             /* Update potential sum for this i atom from the interaction with this j atom. */
 262             velecsum         = _mm_add_pd(velecsum,velec);
 263             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 264
 265             fscal            = _mm_add_pd(felec,fvdw);
 266
 267             /* Update vectorial force */
 268             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 269             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 270             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 271
 272             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 273             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 274             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 275
 276             /**************************
 277              * CALCULATE INTERACTIONS *
 278              **************************/
 279
 280             r10              = _mm_mul_pd(rsq10,rinv10);
 281
 282             /* Compute parameters for interactions between i and j atoms */
 283             qq10             = _mm_mul_pd(iq1,jq0);
 284
 285             /* Calculate table index by multiplying r with table scale and truncate to integer */
 286             rt               = _mm_mul_pd(r10,vftabscale);
 287             vfitab           = _mm_cvttpd_epi32(rt);
 288 #ifdef __XOP__
 289             vfeps            = _mm_frcz_pd(rt);
 290 #else
 291             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 292 #endif
 293             twovfeps         = _mm_add_pd(vfeps,vfeps);
 294             vfitab           = _mm_slli_epi32(vfitab,2);
 295
 296             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 297             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 298             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 299             GMX_MM_TRANSPOSE2_PD(Y,F);
 300             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 301             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 302             GMX_MM_TRANSPOSE2_PD(G,H);
 303             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 304             VV               = _mm_macc_pd(vfeps,Fp,Y);
 305             velec            = _mm_mul_pd(qq10,VV);
 306             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 307             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 308
 309             /* Update potential sum for this i atom from the interaction with this j atom. */
 310             velecsum         = _mm_add_pd(velecsum,velec);
 311
 312             fscal            = felec;
 313
 314             /* Update vectorial force */
 315             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 316             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 317             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 318
 319             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 320             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 321             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 322
 323             /**************************
 324              * CALCULATE INTERACTIONS *
 325              **************************/
 326
 327             r20              = _mm_mul_pd(rsq20,rinv20);
 328
 329             /* Compute parameters for interactions between i and j atoms */
 330             qq20             = _mm_mul_pd(iq2,jq0);
 331
 332             /* Calculate table index by multiplying r with table scale and truncate to integer */
 333             rt               = _mm_mul_pd(r20,vftabscale);
 334             vfitab           = _mm_cvttpd_epi32(rt);
 335 #ifdef __XOP__
 336             vfeps            = _mm_frcz_pd(rt);
 337 #else
 338             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 339 #endif
 340             twovfeps         = _mm_add_pd(vfeps,vfeps);
 341             vfitab           = _mm_slli_epi32(vfitab,2);
 342
 343             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 344             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 345             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 346             GMX_MM_TRANSPOSE2_PD(Y,F);
 347             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 348             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 349             GMX_MM_TRANSPOSE2_PD(G,H);
 350             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 351             VV               = _mm_macc_pd(vfeps,Fp,Y);
 352             velec            = _mm_mul_pd(qq20,VV);
 353             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 354             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 355
 356             /* Update potential sum for this i atom from the interaction with this j atom. */
 357             velecsum         = _mm_add_pd(velecsum,velec);
 358
 359             fscal            = felec;
 360
 361             /* Update vectorial force */
 362             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 363             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 364             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 365
 366             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 367             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 368             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 369
 370             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 371
 372             /* Inner loop uses 154 flops */
 373         }
 374
 375         if(jidx<j_index_end)
 376         {
 377
 378             jnrA             = jjnr[jidx];
 379             j_coord_offsetA  = DIM*jnrA;
 380
 381             /* load j atom coordinates */
 382             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 383                                               &jx0,&jy0,&jz0);
 384
 385             /* Calculate displacement vector */
 386             dx00             = _mm_sub_pd(ix0,jx0);
 387             dy00             = _mm_sub_pd(iy0,jy0);
 388             dz00             = _mm_sub_pd(iz0,jz0);
 389             dx10             = _mm_sub_pd(ix1,jx0);
 390             dy10             = _mm_sub_pd(iy1,jy0);
 391             dz10             = _mm_sub_pd(iz1,jz0);
 392             dx20             = _mm_sub_pd(ix2,jx0);
 393             dy20             = _mm_sub_pd(iy2,jy0);
 394             dz20             = _mm_sub_pd(iz2,jz0);
 395
 396             /* Calculate squared distance and things based on it */
 397             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 398             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 399             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 400
 401             rinv00           = avx128fma_invsqrt_d(rsq00);
 402             rinv10           = avx128fma_invsqrt_d(rsq10);
 403             rinv20           = avx128fma_invsqrt_d(rsq20);
 404
 405             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 406
 407             /* Load parameters for j particles */
 408             jq0              = _mm_load_sd(charge+jnrA+0);
 409             vdwjidx0A        = 2*vdwtype[jnrA+0];
 410
 411             fjx0             = _mm_setzero_pd();
 412             fjy0             = _mm_setzero_pd();
 413             fjz0             = _mm_setzero_pd();
 414
 415             /**************************
 416              * CALCULATE INTERACTIONS *
 417              **************************/
 418
 419             r00              = _mm_mul_pd(rsq00,rinv00);
 420
 421             /* Compute parameters for interactions between i and j atoms */
 422             qq00             = _mm_mul_pd(iq0,jq0);
 423             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 424
 425             /* Calculate table index by multiplying r with table scale and truncate to integer */
 426             rt               = _mm_mul_pd(r00,vftabscale);
 427             vfitab           = _mm_cvttpd_epi32(rt);
 428 #ifdef __XOP__
 429             vfeps            = _mm_frcz_pd(rt);
 430 #else
 431             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 432 #endif
 433             twovfeps         = _mm_add_pd(vfeps,vfeps);
 434             vfitab           = _mm_slli_epi32(vfitab,2);
 435
 436             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 437             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 438             F                = _mm_setzero_pd();
 439             GMX_MM_TRANSPOSE2_PD(Y,F);
 440             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 441             H                = _mm_setzero_pd();
 442             GMX_MM_TRANSPOSE2_PD(G,H);
 443             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 444             VV               = _mm_macc_pd(vfeps,Fp,Y);
 445             velec            = _mm_mul_pd(qq00,VV);
 446             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 447             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 448
 449             /* LENNARD-JONES DISPERSION/REPULSION */
 450
 451             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 452             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 453             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 454             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 455             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 456
 457             /* Update potential sum for this i atom from the interaction with this j atom. */
 458             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 459             velecsum         = _mm_add_pd(velecsum,velec);
 460             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 461             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 462
 463             fscal            = _mm_add_pd(felec,fvdw);
 464
 465             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 466
 467             /* Update vectorial force */
 468             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 469             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 470             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 471
 472             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 473             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 474             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 475
 476             /**************************
 477              * CALCULATE INTERACTIONS *
 478              **************************/
 479
 480             r10              = _mm_mul_pd(rsq10,rinv10);
 481
 482             /* Compute parameters for interactions between i and j atoms */
 483             qq10             = _mm_mul_pd(iq1,jq0);
 484
 485             /* Calculate table index by multiplying r with table scale and truncate to integer */
 486             rt               = _mm_mul_pd(r10,vftabscale);
 487             vfitab           = _mm_cvttpd_epi32(rt);
 488 #ifdef __XOP__
 489             vfeps            = _mm_frcz_pd(rt);
 490 #else
 491             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 492 #endif
 493             twovfeps         = _mm_add_pd(vfeps,vfeps);
 494             vfitab           = _mm_slli_epi32(vfitab,2);
 495
 496             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 497             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 498             F                = _mm_setzero_pd();
 499             GMX_MM_TRANSPOSE2_PD(Y,F);
 500             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 501             H                = _mm_setzero_pd();
 502             GMX_MM_TRANSPOSE2_PD(G,H);
 503             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 504             VV               = _mm_macc_pd(vfeps,Fp,Y);
 505             velec            = _mm_mul_pd(qq10,VV);
 506             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 507             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 508
 509             /* Update potential sum for this i atom from the interaction with this j atom. */
 510             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 511             velecsum         = _mm_add_pd(velecsum,velec);
 512
 513             fscal            = felec;
 514
 515             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 516
 517             /* Update vectorial force */
 518             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 519             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 520             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 521
 522             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 523             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 524             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 525
 526             /**************************
 527              * CALCULATE INTERACTIONS *
 528              **************************/
 529
 530             r20              = _mm_mul_pd(rsq20,rinv20);
 531
 532             /* Compute parameters for interactions between i and j atoms */
 533             qq20             = _mm_mul_pd(iq2,jq0);
 534
 535             /* Calculate table index by multiplying r with table scale and truncate to integer */
 536             rt               = _mm_mul_pd(r20,vftabscale);
 537             vfitab           = _mm_cvttpd_epi32(rt);
 538 #ifdef __XOP__
 539             vfeps            = _mm_frcz_pd(rt);
 540 #else
 541             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 542 #endif
 543             twovfeps         = _mm_add_pd(vfeps,vfeps);
 544             vfitab           = _mm_slli_epi32(vfitab,2);
 545
 546             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 547             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 548             F                = _mm_setzero_pd();
 549             GMX_MM_TRANSPOSE2_PD(Y,F);
 550             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 551             H                = _mm_setzero_pd();
 552             GMX_MM_TRANSPOSE2_PD(G,H);
 553             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 554             VV               = _mm_macc_pd(vfeps,Fp,Y);
 555             velec            = _mm_mul_pd(qq20,VV);
 556             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 557             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 558
 559             /* Update potential sum for this i atom from the interaction with this j atom. */
 560             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 561             velecsum         = _mm_add_pd(velecsum,velec);
 562
 563             fscal            = felec;
 564
 565             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 566
 567             /* Update vectorial force */
 568             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 569             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 570             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 571
 572             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 573             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 574             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 575
 576             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 577
 578             /* Inner loop uses 154 flops */
 579         }
 580
 581         /* End of innermost loop */
 582
 583         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 584                                               f+i_coord_offset,fshift+i_shift_offset);
 585
 586         ggid                        = gid[iidx];
 587         /* Update potential energies */
 588         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 589         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 590
 591         /* Increment number of inner iterations */
 592         inneriter                  += j_index_end - j_index_start;
 593
 594         /* Outer loop uses 20 flops */
 595     }
 596
 597     /* Increment number of outer iterations */
 598     outeriter        += nri;
 599
 600     /* Update outer/inner flops */
 601
 602     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*154);
 603 }
 604 /*
 605  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_avx_128_fma_double
 606  * Electrostatics interaction: CubicSplineTable
 607  * VdW interaction:            LennardJones
 608  * Geometry:                   Water3-Particle
 609  * Calculate force/pot:        Force
 610  */
 611 void
 612 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_avx_128_fma_double
 613                     (t_nblist                    * gmx_restrict       nlist,
 614                      rvec                        * gmx_restrict          xx,
 615                      rvec                        * gmx_restrict          ff,
 616                      struct t_forcerec           * gmx_restrict          fr,
 617                      t_mdatoms                   * gmx_restrict     mdatoms,
 618                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 619                      t_nrnb                      * gmx_restrict        nrnb)
 620 {
 621     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 622      * just 0 for non-waters.
 623      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 624      * jnr indices corresponding to data put in the four positions in the SIMD register.
 625      */
 626     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 627     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 628     int              jnrA,jnrB;
 629     int              j_coord_offsetA,j_coord_offsetB;
 630     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 631     real             rcutoff_scalar;
 632     real             *shiftvec,*fshift,*x,*f;
 633     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 634     int              vdwioffset0;
 635     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 636     int              vdwioffset1;
 637     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 638     int              vdwioffset2;
 639     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 640     int              vdwjidx0A,vdwjidx0B;
 641     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 642     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 643     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 644     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 645     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 646     real             *charge;
 647     int              nvdwtype;
 648     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 649     int              *vdwtype;
 650     real             *vdwparam;
 651     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 652     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 653     __m128i          vfitab;
 654     __m128i          ifour       = _mm_set1_epi32(4);
 655     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 656     real             *vftab;
 657     __m128d          dummy_mask,cutoff_mask;
 658     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 659     __m128d          one     = _mm_set1_pd(1.0);
 660     __m128d          two     = _mm_set1_pd(2.0);
 661     x                = xx[0];
 662     f                = ff[0];
 663
 664     nri              = nlist->nri;
 665     iinr             = nlist->iinr;
 666     jindex           = nlist->jindex;
 667     jjnr             = nlist->jjnr;
 668     shiftidx         = nlist->shift;
 669     gid              = nlist->gid;
 670     shiftvec         = fr->shift_vec[0];
 671     fshift           = fr->fshift[0];
 672     facel            = _mm_set1_pd(fr->ic->epsfac);
 673     charge           = mdatoms->chargeA;
 674     nvdwtype         = fr->ntype;
 675     vdwparam         = fr->nbfp;
 676     vdwtype          = mdatoms->typeA;
 677
 678     vftab            = kernel_data->table_elec->data;
 679     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 680
 681     /* Setup water-specific parameters */
 682     inr              = nlist->iinr[0];
 683     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 684     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 685     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 686     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 687
 688     /* Avoid stupid compiler warnings */
 689     jnrA = jnrB = 0;
 690     j_coord_offsetA = 0;
 691     j_coord_offsetB = 0;
 692
 693     outeriter        = 0;
 694     inneriter        = 0;
 695
 696     /* Start outer loop over neighborlists */
 697     for(iidx=0; iidx<nri; iidx++)
 698     {
 699         /* Load shift vector for this list */
 700         i_shift_offset   = DIM*shiftidx[iidx];
 701
 702         /* Load limits for loop over neighbors */
 703         j_index_start    = jindex[iidx];
 704         j_index_end      = jindex[iidx+1];
 705
 706         /* Get outer coordinate index */
 707         inr              = iinr[iidx];
 708         i_coord_offset   = DIM*inr;
 709
 710         /* Load i particle coords and add shift vector */
 711         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 712                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 713
 714         fix0             = _mm_setzero_pd();
 715         fiy0             = _mm_setzero_pd();
 716         fiz0             = _mm_setzero_pd();
 717         fix1             = _mm_setzero_pd();
 718         fiy1             = _mm_setzero_pd();
 719         fiz1             = _mm_setzero_pd();
 720         fix2             = _mm_setzero_pd();
 721         fiy2             = _mm_setzero_pd();
 722         fiz2             = _mm_setzero_pd();
 723
 724         /* Start inner kernel loop */
 725         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 726         {
 727
 728             /* Get j neighbor index, and coordinate index */
 729             jnrA             = jjnr[jidx];
 730             jnrB             = jjnr[jidx+1];
 731             j_coord_offsetA  = DIM*jnrA;
 732             j_coord_offsetB  = DIM*jnrB;
 733
 734             /* load j atom coordinates */
 735             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 736                                               &jx0,&jy0,&jz0);
 737
 738             /* Calculate displacement vector */
 739             dx00             = _mm_sub_pd(ix0,jx0);
 740             dy00             = _mm_sub_pd(iy0,jy0);
 741             dz00             = _mm_sub_pd(iz0,jz0);
 742             dx10             = _mm_sub_pd(ix1,jx0);
 743             dy10             = _mm_sub_pd(iy1,jy0);
 744             dz10             = _mm_sub_pd(iz1,jz0);
 745             dx20             = _mm_sub_pd(ix2,jx0);
 746             dy20             = _mm_sub_pd(iy2,jy0);
 747             dz20             = _mm_sub_pd(iz2,jz0);
 748
 749             /* Calculate squared distance and things based on it */
 750             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 751             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 752             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 753
 754             rinv00           = avx128fma_invsqrt_d(rsq00);
 755             rinv10           = avx128fma_invsqrt_d(rsq10);
 756             rinv20           = avx128fma_invsqrt_d(rsq20);
 757
 758             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 759
 760             /* Load parameters for j particles */
 761             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 762             vdwjidx0A        = 2*vdwtype[jnrA+0];
 763             vdwjidx0B        = 2*vdwtype[jnrB+0];
 764
 765             fjx0             = _mm_setzero_pd();
 766             fjy0             = _mm_setzero_pd();
 767             fjz0             = _mm_setzero_pd();
 768
 769             /**************************
 770              * CALCULATE INTERACTIONS *
 771              **************************/
 772
 773             r00              = _mm_mul_pd(rsq00,rinv00);
 774
 775             /* Compute parameters for interactions between i and j atoms */
 776             qq00             = _mm_mul_pd(iq0,jq0);
 777             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 778                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 779
 780             /* Calculate table index by multiplying r with table scale and truncate to integer */
 781             rt               = _mm_mul_pd(r00,vftabscale);
 782             vfitab           = _mm_cvttpd_epi32(rt);
 783 #ifdef __XOP__
 784             vfeps            = _mm_frcz_pd(rt);
 785 #else
 786             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 787 #endif
 788             twovfeps         = _mm_add_pd(vfeps,vfeps);
 789             vfitab           = _mm_slli_epi32(vfitab,2);
 790
 791             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 792             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 793             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 794             GMX_MM_TRANSPOSE2_PD(Y,F);
 795             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 796             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 797             GMX_MM_TRANSPOSE2_PD(G,H);
 798             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 799             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 800             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 801
 802             /* LENNARD-JONES DISPERSION/REPULSION */
 803
 804             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 805             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 806
 807             fscal            = _mm_add_pd(felec,fvdw);
 808
 809             /* Update vectorial force */
 810             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 811             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 812             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 813
 814             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 815             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 816             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 817
 818             /**************************
 819              * CALCULATE INTERACTIONS *
 820              **************************/
 821
 822             r10              = _mm_mul_pd(rsq10,rinv10);
 823
 824             /* Compute parameters for interactions between i and j atoms */
 825             qq10             = _mm_mul_pd(iq1,jq0);
 826
 827             /* Calculate table index by multiplying r with table scale and truncate to integer */
 828             rt               = _mm_mul_pd(r10,vftabscale);
 829             vfitab           = _mm_cvttpd_epi32(rt);
 830 #ifdef __XOP__
 831             vfeps            = _mm_frcz_pd(rt);
 832 #else
 833             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 834 #endif
 835             twovfeps         = _mm_add_pd(vfeps,vfeps);
 836             vfitab           = _mm_slli_epi32(vfitab,2);
 837
 838             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 839             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 840             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 841             GMX_MM_TRANSPOSE2_PD(Y,F);
 842             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 843             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 844             GMX_MM_TRANSPOSE2_PD(G,H);
 845             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 846             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 847             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 848
 849             fscal            = felec;
 850
 851             /* Update vectorial force */
 852             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 853             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 854             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 855
 856             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 857             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 858             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 859
 860             /**************************
 861              * CALCULATE INTERACTIONS *
 862              **************************/
 863
 864             r20              = _mm_mul_pd(rsq20,rinv20);
 865
 866             /* Compute parameters for interactions between i and j atoms */
 867             qq20             = _mm_mul_pd(iq2,jq0);
 868
 869             /* Calculate table index by multiplying r with table scale and truncate to integer */
 870             rt               = _mm_mul_pd(r20,vftabscale);
 871             vfitab           = _mm_cvttpd_epi32(rt);
 872 #ifdef __XOP__
 873             vfeps            = _mm_frcz_pd(rt);
 874 #else
 875             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 876 #endif
 877             twovfeps         = _mm_add_pd(vfeps,vfeps);
 878             vfitab           = _mm_slli_epi32(vfitab,2);
 879
 880             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 881             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 882             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 883             GMX_MM_TRANSPOSE2_PD(Y,F);
 884             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 885             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 886             GMX_MM_TRANSPOSE2_PD(G,H);
 887             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 888             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 889             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 890
 891             fscal            = felec;
 892
 893             /* Update vectorial force */
 894             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 895             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 896             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 897
 898             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 899             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 900             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 901
 902             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 903
 904             /* Inner loop uses 137 flops */
 905         }
 906
 907         if(jidx<j_index_end)
 908         {
 909
 910             jnrA             = jjnr[jidx];
 911             j_coord_offsetA  = DIM*jnrA;
 912
 913             /* load j atom coordinates */
 914             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 915                                               &jx0,&jy0,&jz0);
 916
 917             /* Calculate displacement vector */
 918             dx00             = _mm_sub_pd(ix0,jx0);
 919             dy00             = _mm_sub_pd(iy0,jy0);
 920             dz00             = _mm_sub_pd(iz0,jz0);
 921             dx10             = _mm_sub_pd(ix1,jx0);
 922             dy10             = _mm_sub_pd(iy1,jy0);
 923             dz10             = _mm_sub_pd(iz1,jz0);
 924             dx20             = _mm_sub_pd(ix2,jx0);
 925             dy20             = _mm_sub_pd(iy2,jy0);
 926             dz20             = _mm_sub_pd(iz2,jz0);
 927
 928             /* Calculate squared distance and things based on it */
 929             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 930             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 931             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 932
 933             rinv00           = avx128fma_invsqrt_d(rsq00);
 934             rinv10           = avx128fma_invsqrt_d(rsq10);
 935             rinv20           = avx128fma_invsqrt_d(rsq20);
 936
 937             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 938
 939             /* Load parameters for j particles */
 940             jq0              = _mm_load_sd(charge+jnrA+0);
 941             vdwjidx0A        = 2*vdwtype[jnrA+0];
 942
 943             fjx0             = _mm_setzero_pd();
 944             fjy0             = _mm_setzero_pd();
 945             fjz0             = _mm_setzero_pd();
 946
 947             /**************************
 948              * CALCULATE INTERACTIONS *
 949              **************************/
 950
 951             r00              = _mm_mul_pd(rsq00,rinv00);
 952
 953             /* Compute parameters for interactions between i and j atoms */
 954             qq00             = _mm_mul_pd(iq0,jq0);
 955             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 956
 957             /* Calculate table index by multiplying r with table scale and truncate to integer */
 958             rt               = _mm_mul_pd(r00,vftabscale);
 959             vfitab           = _mm_cvttpd_epi32(rt);
 960 #ifdef __XOP__
 961             vfeps            = _mm_frcz_pd(rt);
 962 #else
 963             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 964 #endif
 965             twovfeps         = _mm_add_pd(vfeps,vfeps);
 966             vfitab           = _mm_slli_epi32(vfitab,2);
 967
 968             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 969             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 970             F                = _mm_setzero_pd();
 971             GMX_MM_TRANSPOSE2_PD(Y,F);
 972             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 973             H                = _mm_setzero_pd();
 974             GMX_MM_TRANSPOSE2_PD(G,H);
 975             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 976             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 977             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 978
 979             /* LENNARD-JONES DISPERSION/REPULSION */
 980
 981             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 982             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 983
 984             fscal            = _mm_add_pd(felec,fvdw);
 985
 986             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 987
 988             /* Update vectorial force */
 989             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 990             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 991             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 992
 993             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 994             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 995             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 996
 997             /**************************
 998              * CALCULATE INTERACTIONS *
 999              **************************/
1000
1001             r10              = _mm_mul_pd(rsq10,rinv10);
1002
1003             /* Compute parameters for interactions between i and j atoms */
1004             qq10             = _mm_mul_pd(iq1,jq0);
1005
1006             /* Calculate table index by multiplying r with table scale and truncate to integer */
1007             rt               = _mm_mul_pd(r10,vftabscale);
1008             vfitab           = _mm_cvttpd_epi32(rt);
1009 #ifdef __XOP__
1010             vfeps            = _mm_frcz_pd(rt);
1011 #else
1012             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1013 #endif
1014             twovfeps         = _mm_add_pd(vfeps,vfeps);
1015             vfitab           = _mm_slli_epi32(vfitab,2);
1016
1017             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1018             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1019             F                = _mm_setzero_pd();
1020             GMX_MM_TRANSPOSE2_PD(Y,F);
1021             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1022             H                = _mm_setzero_pd();
1023             GMX_MM_TRANSPOSE2_PD(G,H);
1024             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1025             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1026             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1027
1028             fscal            = felec;
1029
1030             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1031
1032             /* Update vectorial force */
1033             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1034             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1035             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1036
1037             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1038             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1039             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1040
1041             /**************************
1042              * CALCULATE INTERACTIONS *
1043              **************************/
1044
1045             r20              = _mm_mul_pd(rsq20,rinv20);
1046
1047             /* Compute parameters for interactions between i and j atoms */
1048             qq20             = _mm_mul_pd(iq2,jq0);
1049
1050             /* Calculate table index by multiplying r with table scale and truncate to integer */
1051             rt               = _mm_mul_pd(r20,vftabscale);
1052             vfitab           = _mm_cvttpd_epi32(rt);
1053 #ifdef __XOP__
1054             vfeps            = _mm_frcz_pd(rt);
1055 #else
1056             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1057 #endif
1058             twovfeps         = _mm_add_pd(vfeps,vfeps);
1059             vfitab           = _mm_slli_epi32(vfitab,2);
1060
1061             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1062             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1063             F                = _mm_setzero_pd();
1064             GMX_MM_TRANSPOSE2_PD(Y,F);
1065             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1066             H                = _mm_setzero_pd();
1067             GMX_MM_TRANSPOSE2_PD(G,H);
1068             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1069             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1070             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1071
1072             fscal            = felec;
1073
1074             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1075
1076             /* Update vectorial force */
1077             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1078             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1079             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1080
1081             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1082             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1083             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1084
1085             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1086
1087             /* Inner loop uses 137 flops */
1088         }
1089
1090         /* End of innermost loop */
1091
1092         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1093                                               f+i_coord_offset,fshift+i_shift_offset);
1094
1095         /* Increment number of inner iterations */
1096         inneriter                  += j_index_end - j_index_start;
1097
1098         /* Outer loop uses 18 flops */
1099     }
1100
1101     /* Increment number of outer iterations */
1102     outeriter        += nri;
1103
1104     /* Update outer/inner flops */
1105
1106     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*137);
1107 }