src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCSTab_VdwCSTab_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_VdwCSTab_GeomW3P1_VF_avx_128_fma_double
  51  * Electrostatics interaction: CubicSplineTable
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCSTab_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     nvdwtype         = fr->ntype;
 120     vdwparam         = fr->nbfp;
 121     vdwtype          = mdatoms->typeA;
 122
 123     vftab            = kernel_data->table_elec_vdw->data;
 124     vftabscale       = _mm_set1_pd(kernel_data->table_elec_vdw->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             /* Load parameters for j particles */
 208             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 209             vdwjidx0A        = 2*vdwtype[jnrA+0];
 210             vdwjidx0B        = 2*vdwtype[jnrB+0];
 211
 212             fjx0             = _mm_setzero_pd();
 213             fjy0             = _mm_setzero_pd();
 214             fjz0             = _mm_setzero_pd();
 215
 216             /**************************
 217              * CALCULATE INTERACTIONS *
 218              **************************/
 219
 220             r00              = _mm_mul_pd(rsq00,rinv00);
 221
 222             /* Compute parameters for interactions between i and j atoms */
 223             qq00             = _mm_mul_pd(iq0,jq0);
 224             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 225                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 226
 227             /* Calculate table index by multiplying r with table scale and truncate to integer */
 228             rt               = _mm_mul_pd(r00,vftabscale);
 229             vfitab           = _mm_cvttpd_epi32(rt);
 230 #ifdef __XOP__
 231             vfeps            = _mm_frcz_pd(rt);
 232 #else
 233             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 234 #endif
 235             twovfeps         = _mm_add_pd(vfeps,vfeps);
 236             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 237
 238             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 239             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 240             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 241             GMX_MM_TRANSPOSE2_PD(Y,F);
 242             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 243             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 244             GMX_MM_TRANSPOSE2_PD(G,H);
 245             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 246             VV               = _mm_macc_pd(vfeps,Fp,Y);
 247             velec            = _mm_mul_pd(qq00,VV);
 248             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 249             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 250
 251             /* CUBIC SPLINE TABLE DISPERSION */
 252             vfitab           = _mm_add_epi32(vfitab,ifour);
 253             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 254             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 255             GMX_MM_TRANSPOSE2_PD(Y,F);
 256             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 257             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 258             GMX_MM_TRANSPOSE2_PD(G,H);
 259             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 260             VV               = _mm_macc_pd(vfeps,Fp,Y);
 261             vvdw6            = _mm_mul_pd(c6_00,VV);
 262             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 263             fvdw6            = _mm_mul_pd(c6_00,FF);
 264
 265             /* CUBIC SPLINE TABLE REPULSION */
 266             vfitab           = _mm_add_epi32(vfitab,ifour);
 267             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 268             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 269             GMX_MM_TRANSPOSE2_PD(Y,F);
 270             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 271             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 272             GMX_MM_TRANSPOSE2_PD(G,H);
 273             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 274             VV               = _mm_macc_pd(vfeps,Fp,Y);
 275             vvdw12           = _mm_mul_pd(c12_00,VV);
 276             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 277             fvdw12           = _mm_mul_pd(c12_00,FF);
 278             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 279             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 280
 281             /* Update potential sum for this i atom from the interaction with this j atom. */
 282             velecsum         = _mm_add_pd(velecsum,velec);
 283             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 284
 285             fscal            = _mm_add_pd(felec,fvdw);
 286
 287             /* Update vectorial force */
 288             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 289             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 290             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 291
 292             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 293             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 294             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 295
 296             /**************************
 297              * CALCULATE INTERACTIONS *
 298              **************************/
 299
 300             r10              = _mm_mul_pd(rsq10,rinv10);
 301
 302             /* Compute parameters for interactions between i and j atoms */
 303             qq10             = _mm_mul_pd(iq1,jq0);
 304
 305             /* Calculate table index by multiplying r with table scale and truncate to integer */
 306             rt               = _mm_mul_pd(r10,vftabscale);
 307             vfitab           = _mm_cvttpd_epi32(rt);
 308 #ifdef __XOP__
 309             vfeps            = _mm_frcz_pd(rt);
 310 #else
 311             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 312 #endif
 313             twovfeps         = _mm_add_pd(vfeps,vfeps);
 314             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 315
 316             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 317             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 318             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 319             GMX_MM_TRANSPOSE2_PD(Y,F);
 320             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 321             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 322             GMX_MM_TRANSPOSE2_PD(G,H);
 323             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 324             VV               = _mm_macc_pd(vfeps,Fp,Y);
 325             velec            = _mm_mul_pd(qq10,VV);
 326             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 327             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 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             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 336             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 337             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 338
 339             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 340             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 341             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 342
 343             /**************************
 344              * CALCULATE INTERACTIONS *
 345              **************************/
 346
 347             r20              = _mm_mul_pd(rsq20,rinv20);
 348
 349             /* Compute parameters for interactions between i and j atoms */
 350             qq20             = _mm_mul_pd(iq2,jq0);
 351
 352             /* Calculate table index by multiplying r with table scale and truncate to integer */
 353             rt               = _mm_mul_pd(r20,vftabscale);
 354             vfitab           = _mm_cvttpd_epi32(rt);
 355 #ifdef __XOP__
 356             vfeps            = _mm_frcz_pd(rt);
 357 #else
 358             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 359 #endif
 360             twovfeps         = _mm_add_pd(vfeps,vfeps);
 361             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 362
 363             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 364             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 365             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 366             GMX_MM_TRANSPOSE2_PD(Y,F);
 367             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 368             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 369             GMX_MM_TRANSPOSE2_PD(G,H);
 370             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 371             VV               = _mm_macc_pd(vfeps,Fp,Y);
 372             velec            = _mm_mul_pd(qq20,VV);
 373             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 374             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 375
 376             /* Update potential sum for this i atom from the interaction with this j atom. */
 377             velecsum         = _mm_add_pd(velecsum,velec);
 378
 379             fscal            = felec;
 380
 381             /* Update vectorial force */
 382             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 383             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 384             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 385
 386             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 387             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 388             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 389
 390             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 391
 392             /* Inner loop uses 171 flops */
 393         }
 394
 395         if(jidx<j_index_end)
 396         {
 397
 398             jnrA             = jjnr[jidx];
 399             j_coord_offsetA  = DIM*jnrA;
 400
 401             /* load j atom coordinates */
 402             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 403                                               &jx0,&jy0,&jz0);
 404
 405             /* Calculate displacement vector */
 406             dx00             = _mm_sub_pd(ix0,jx0);
 407             dy00             = _mm_sub_pd(iy0,jy0);
 408             dz00             = _mm_sub_pd(iz0,jz0);
 409             dx10             = _mm_sub_pd(ix1,jx0);
 410             dy10             = _mm_sub_pd(iy1,jy0);
 411             dz10             = _mm_sub_pd(iz1,jz0);
 412             dx20             = _mm_sub_pd(ix2,jx0);
 413             dy20             = _mm_sub_pd(iy2,jy0);
 414             dz20             = _mm_sub_pd(iz2,jz0);
 415
 416             /* Calculate squared distance and things based on it */
 417             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 418             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 419             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 420
 421             rinv00           = avx128fma_invsqrt_d(rsq00);
 422             rinv10           = avx128fma_invsqrt_d(rsq10);
 423             rinv20           = avx128fma_invsqrt_d(rsq20);
 424
 425             /* Load parameters for j particles */
 426             jq0              = _mm_load_sd(charge+jnrA+0);
 427             vdwjidx0A        = 2*vdwtype[jnrA+0];
 428
 429             fjx0             = _mm_setzero_pd();
 430             fjy0             = _mm_setzero_pd();
 431             fjz0             = _mm_setzero_pd();
 432
 433             /**************************
 434              * CALCULATE INTERACTIONS *
 435              **************************/
 436
 437             r00              = _mm_mul_pd(rsq00,rinv00);
 438
 439             /* Compute parameters for interactions between i and j atoms */
 440             qq00             = _mm_mul_pd(iq0,jq0);
 441             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 442
 443             /* Calculate table index by multiplying r with table scale and truncate to integer */
 444             rt               = _mm_mul_pd(r00,vftabscale);
 445             vfitab           = _mm_cvttpd_epi32(rt);
 446 #ifdef __XOP__
 447             vfeps            = _mm_frcz_pd(rt);
 448 #else
 449             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 450 #endif
 451             twovfeps         = _mm_add_pd(vfeps,vfeps);
 452             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 453
 454             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 455             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 456             F                = _mm_setzero_pd();
 457             GMX_MM_TRANSPOSE2_PD(Y,F);
 458             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 459             H                = _mm_setzero_pd();
 460             GMX_MM_TRANSPOSE2_PD(G,H);
 461             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 462             VV               = _mm_macc_pd(vfeps,Fp,Y);
 463             velec            = _mm_mul_pd(qq00,VV);
 464             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 465             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 466
 467             /* CUBIC SPLINE TABLE DISPERSION */
 468             vfitab           = _mm_add_epi32(vfitab,ifour);
 469             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 470             F                = _mm_setzero_pd();
 471             GMX_MM_TRANSPOSE2_PD(Y,F);
 472             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 473             H                = _mm_setzero_pd();
 474             GMX_MM_TRANSPOSE2_PD(G,H);
 475             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 476             VV               = _mm_macc_pd(vfeps,Fp,Y);
 477             vvdw6            = _mm_mul_pd(c6_00,VV);
 478             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 479             fvdw6            = _mm_mul_pd(c6_00,FF);
 480
 481             /* CUBIC SPLINE TABLE REPULSION */
 482             vfitab           = _mm_add_epi32(vfitab,ifour);
 483             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 484             F                = _mm_setzero_pd();
 485             GMX_MM_TRANSPOSE2_PD(Y,F);
 486             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 487             H                = _mm_setzero_pd();
 488             GMX_MM_TRANSPOSE2_PD(G,H);
 489             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 490             VV               = _mm_macc_pd(vfeps,Fp,Y);
 491             vvdw12           = _mm_mul_pd(c12_00,VV);
 492             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 493             fvdw12           = _mm_mul_pd(c12_00,FF);
 494             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 495             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 496
 497             /* Update potential sum for this i atom from the interaction with this j atom. */
 498             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 499             velecsum         = _mm_add_pd(velecsum,velec);
 500             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 501             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 502
 503             fscal            = _mm_add_pd(felec,fvdw);
 504
 505             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 506
 507             /* Update vectorial force */
 508             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 509             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 510             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 511
 512             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 513             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 514             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 515
 516             /**************************
 517              * CALCULATE INTERACTIONS *
 518              **************************/
 519
 520             r10              = _mm_mul_pd(rsq10,rinv10);
 521
 522             /* Compute parameters for interactions between i and j atoms */
 523             qq10             = _mm_mul_pd(iq1,jq0);
 524
 525             /* Calculate table index by multiplying r with table scale and truncate to integer */
 526             rt               = _mm_mul_pd(r10,vftabscale);
 527             vfitab           = _mm_cvttpd_epi32(rt);
 528 #ifdef __XOP__
 529             vfeps            = _mm_frcz_pd(rt);
 530 #else
 531             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 532 #endif
 533             twovfeps         = _mm_add_pd(vfeps,vfeps);
 534             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 535
 536             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 537             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 538             F                = _mm_setzero_pd();
 539             GMX_MM_TRANSPOSE2_PD(Y,F);
 540             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 541             H                = _mm_setzero_pd();
 542             GMX_MM_TRANSPOSE2_PD(G,H);
 543             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 544             VV               = _mm_macc_pd(vfeps,Fp,Y);
 545             velec            = _mm_mul_pd(qq10,VV);
 546             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 547             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 548
 549             /* Update potential sum for this i atom from the interaction with this j atom. */
 550             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 551             velecsum         = _mm_add_pd(velecsum,velec);
 552
 553             fscal            = felec;
 554
 555             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 556
 557             /* Update vectorial force */
 558             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 559             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 560             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 561
 562             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 563             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 564             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 565
 566             /**************************
 567              * CALCULATE INTERACTIONS *
 568              **************************/
 569
 570             r20              = _mm_mul_pd(rsq20,rinv20);
 571
 572             /* Compute parameters for interactions between i and j atoms */
 573             qq20             = _mm_mul_pd(iq2,jq0);
 574
 575             /* Calculate table index by multiplying r with table scale and truncate to integer */
 576             rt               = _mm_mul_pd(r20,vftabscale);
 577             vfitab           = _mm_cvttpd_epi32(rt);
 578 #ifdef __XOP__
 579             vfeps            = _mm_frcz_pd(rt);
 580 #else
 581             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 582 #endif
 583             twovfeps         = _mm_add_pd(vfeps,vfeps);
 584             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 585
 586             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 587             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 588             F                = _mm_setzero_pd();
 589             GMX_MM_TRANSPOSE2_PD(Y,F);
 590             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 591             H                = _mm_setzero_pd();
 592             GMX_MM_TRANSPOSE2_PD(G,H);
 593             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 594             VV               = _mm_macc_pd(vfeps,Fp,Y);
 595             velec            = _mm_mul_pd(qq20,VV);
 596             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 597             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 598
 599             /* Update potential sum for this i atom from the interaction with this j atom. */
 600             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 601             velecsum         = _mm_add_pd(velecsum,velec);
 602
 603             fscal            = felec;
 604
 605             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 606
 607             /* Update vectorial force */
 608             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 609             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 610             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 611
 612             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 613             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 614             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 615
 616             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 617
 618             /* Inner loop uses 171 flops */
 619         }
 620
 621         /* End of innermost loop */
 622
 623         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 624                                               f+i_coord_offset,fshift+i_shift_offset);
 625
 626         ggid                        = gid[iidx];
 627         /* Update potential energies */
 628         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 629         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 630
 631         /* Increment number of inner iterations */
 632         inneriter                  += j_index_end - j_index_start;
 633
 634         /* Outer loop uses 20 flops */
 635     }
 636
 637     /* Increment number of outer iterations */
 638     outeriter        += nri;
 639
 640     /* Update outer/inner flops */
 641
 642     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*171);
 643 }
 644 /*
 645  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_128_fma_double
 646  * Electrostatics interaction: CubicSplineTable
 647  * VdW interaction:            CubicSplineTable
 648  * Geometry:                   Water3-Particle
 649  * Calculate force/pot:        Force
 650  */
 651 void
 652 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_128_fma_double
 653                     (t_nblist                    * gmx_restrict       nlist,
 654                      rvec                        * gmx_restrict          xx,
 655                      rvec                        * gmx_restrict          ff,
 656                      struct t_forcerec           * gmx_restrict          fr,
 657                      t_mdatoms                   * gmx_restrict     mdatoms,
 658                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 659                      t_nrnb                      * gmx_restrict        nrnb)
 660 {
 661     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 662      * just 0 for non-waters.
 663      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 664      * jnr indices corresponding to data put in the four positions in the SIMD register.
 665      */
 666     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 667     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 668     int              jnrA,jnrB;
 669     int              j_coord_offsetA,j_coord_offsetB;
 670     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 671     real             rcutoff_scalar;
 672     real             *shiftvec,*fshift,*x,*f;
 673     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 674     int              vdwioffset0;
 675     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 676     int              vdwioffset1;
 677     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 678     int              vdwioffset2;
 679     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 680     int              vdwjidx0A,vdwjidx0B;
 681     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 682     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 683     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 684     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 685     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 686     real             *charge;
 687     int              nvdwtype;
 688     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 689     int              *vdwtype;
 690     real             *vdwparam;
 691     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 692     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 693     __m128i          vfitab;
 694     __m128i          ifour       = _mm_set1_epi32(4);
 695     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 696     real             *vftab;
 697     __m128d          dummy_mask,cutoff_mask;
 698     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 699     __m128d          one     = _mm_set1_pd(1.0);
 700     __m128d          two     = _mm_set1_pd(2.0);
 701     x                = xx[0];
 702     f                = ff[0];
 703
 704     nri              = nlist->nri;
 705     iinr             = nlist->iinr;
 706     jindex           = nlist->jindex;
 707     jjnr             = nlist->jjnr;
 708     shiftidx         = nlist->shift;
 709     gid              = nlist->gid;
 710     shiftvec         = fr->shift_vec[0];
 711     fshift           = fr->fshift[0];
 712     facel            = _mm_set1_pd(fr->ic->epsfac);
 713     charge           = mdatoms->chargeA;
 714     nvdwtype         = fr->ntype;
 715     vdwparam         = fr->nbfp;
 716     vdwtype          = mdatoms->typeA;
 717
 718     vftab            = kernel_data->table_elec_vdw->data;
 719     vftabscale       = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
 720
 721     /* Setup water-specific parameters */
 722     inr              = nlist->iinr[0];
 723     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 724     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 725     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 726     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 727
 728     /* Avoid stupid compiler warnings */
 729     jnrA = jnrB = 0;
 730     j_coord_offsetA = 0;
 731     j_coord_offsetB = 0;
 732
 733     outeriter        = 0;
 734     inneriter        = 0;
 735
 736     /* Start outer loop over neighborlists */
 737     for(iidx=0; iidx<nri; iidx++)
 738     {
 739         /* Load shift vector for this list */
 740         i_shift_offset   = DIM*shiftidx[iidx];
 741
 742         /* Load limits for loop over neighbors */
 743         j_index_start    = jindex[iidx];
 744         j_index_end      = jindex[iidx+1];
 745
 746         /* Get outer coordinate index */
 747         inr              = iinr[iidx];
 748         i_coord_offset   = DIM*inr;
 749
 750         /* Load i particle coords and add shift vector */
 751         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 752                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 753
 754         fix0             = _mm_setzero_pd();
 755         fiy0             = _mm_setzero_pd();
 756         fiz0             = _mm_setzero_pd();
 757         fix1             = _mm_setzero_pd();
 758         fiy1             = _mm_setzero_pd();
 759         fiz1             = _mm_setzero_pd();
 760         fix2             = _mm_setzero_pd();
 761         fiy2             = _mm_setzero_pd();
 762         fiz2             = _mm_setzero_pd();
 763
 764         /* Start inner kernel loop */
 765         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 766         {
 767
 768             /* Get j neighbor index, and coordinate index */
 769             jnrA             = jjnr[jidx];
 770             jnrB             = jjnr[jidx+1];
 771             j_coord_offsetA  = DIM*jnrA;
 772             j_coord_offsetB  = DIM*jnrB;
 773
 774             /* load j atom coordinates */
 775             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 776                                               &jx0,&jy0,&jz0);
 777
 778             /* Calculate displacement vector */
 779             dx00             = _mm_sub_pd(ix0,jx0);
 780             dy00             = _mm_sub_pd(iy0,jy0);
 781             dz00             = _mm_sub_pd(iz0,jz0);
 782             dx10             = _mm_sub_pd(ix1,jx0);
 783             dy10             = _mm_sub_pd(iy1,jy0);
 784             dz10             = _mm_sub_pd(iz1,jz0);
 785             dx20             = _mm_sub_pd(ix2,jx0);
 786             dy20             = _mm_sub_pd(iy2,jy0);
 787             dz20             = _mm_sub_pd(iz2,jz0);
 788
 789             /* Calculate squared distance and things based on it */
 790             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 791             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 792             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 793
 794             rinv00           = avx128fma_invsqrt_d(rsq00);
 795             rinv10           = avx128fma_invsqrt_d(rsq10);
 796             rinv20           = avx128fma_invsqrt_d(rsq20);
 797
 798             /* Load parameters for j particles */
 799             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 800             vdwjidx0A        = 2*vdwtype[jnrA+0];
 801             vdwjidx0B        = 2*vdwtype[jnrB+0];
 802
 803             fjx0             = _mm_setzero_pd();
 804             fjy0             = _mm_setzero_pd();
 805             fjz0             = _mm_setzero_pd();
 806
 807             /**************************
 808              * CALCULATE INTERACTIONS *
 809              **************************/
 810
 811             r00              = _mm_mul_pd(rsq00,rinv00);
 812
 813             /* Compute parameters for interactions between i and j atoms */
 814             qq00             = _mm_mul_pd(iq0,jq0);
 815             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 816                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 817
 818             /* Calculate table index by multiplying r with table scale and truncate to integer */
 819             rt               = _mm_mul_pd(r00,vftabscale);
 820             vfitab           = _mm_cvttpd_epi32(rt);
 821 #ifdef __XOP__
 822             vfeps            = _mm_frcz_pd(rt);
 823 #else
 824             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 825 #endif
 826             twovfeps         = _mm_add_pd(vfeps,vfeps);
 827             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 828
 829             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 830             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 831             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 832             GMX_MM_TRANSPOSE2_PD(Y,F);
 833             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 834             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 835             GMX_MM_TRANSPOSE2_PD(G,H);
 836             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 837             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 838             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 839
 840             /* CUBIC SPLINE TABLE DISPERSION */
 841             vfitab           = _mm_add_epi32(vfitab,ifour);
 842             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 843             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 844             GMX_MM_TRANSPOSE2_PD(Y,F);
 845             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 846             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 847             GMX_MM_TRANSPOSE2_PD(G,H);
 848             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 849             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 850             fvdw6            = _mm_mul_pd(c6_00,FF);
 851
 852             /* CUBIC SPLINE TABLE REPULSION */
 853             vfitab           = _mm_add_epi32(vfitab,ifour);
 854             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 855             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 856             GMX_MM_TRANSPOSE2_PD(Y,F);
 857             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 858             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 859             GMX_MM_TRANSPOSE2_PD(G,H);
 860             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 861             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 862             fvdw12           = _mm_mul_pd(c12_00,FF);
 863             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 864
 865             fscal            = _mm_add_pd(felec,fvdw);
 866
 867             /* Update vectorial force */
 868             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 869             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 870             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 871
 872             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 873             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 874             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 875
 876             /**************************
 877              * CALCULATE INTERACTIONS *
 878              **************************/
 879
 880             r10              = _mm_mul_pd(rsq10,rinv10);
 881
 882             /* Compute parameters for interactions between i and j atoms */
 883             qq10             = _mm_mul_pd(iq1,jq0);
 884
 885             /* Calculate table index by multiplying r with table scale and truncate to integer */
 886             rt               = _mm_mul_pd(r10,vftabscale);
 887             vfitab           = _mm_cvttpd_epi32(rt);
 888 #ifdef __XOP__
 889             vfeps            = _mm_frcz_pd(rt);
 890 #else
 891             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 892 #endif
 893             twovfeps         = _mm_add_pd(vfeps,vfeps);
 894             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 895
 896             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 897             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 898             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 899             GMX_MM_TRANSPOSE2_PD(Y,F);
 900             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 901             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 902             GMX_MM_TRANSPOSE2_PD(G,H);
 903             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 904             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 905             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 906
 907             fscal            = felec;
 908
 909             /* Update vectorial force */
 910             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 911             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 912             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 913
 914             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 915             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 916             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 917
 918             /**************************
 919              * CALCULATE INTERACTIONS *
 920              **************************/
 921
 922             r20              = _mm_mul_pd(rsq20,rinv20);
 923
 924             /* Compute parameters for interactions between i and j atoms */
 925             qq20             = _mm_mul_pd(iq2,jq0);
 926
 927             /* Calculate table index by multiplying r with table scale and truncate to integer */
 928             rt               = _mm_mul_pd(r20,vftabscale);
 929             vfitab           = _mm_cvttpd_epi32(rt);
 930 #ifdef __XOP__
 931             vfeps            = _mm_frcz_pd(rt);
 932 #else
 933             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 934 #endif
 935             twovfeps         = _mm_add_pd(vfeps,vfeps);
 936             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 937
 938             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 939             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 940             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 941             GMX_MM_TRANSPOSE2_PD(Y,F);
 942             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 943             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 944             GMX_MM_TRANSPOSE2_PD(G,H);
 945             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 946             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 947             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 948
 949             fscal            = felec;
 950
 951             /* Update vectorial force */
 952             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 953             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 954             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 955
 956             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 957             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 958             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 959
 960             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 961
 962             /* Inner loop uses 151 flops */
 963         }
 964
 965         if(jidx<j_index_end)
 966         {
 967
 968             jnrA             = jjnr[jidx];
 969             j_coord_offsetA  = DIM*jnrA;
 970
 971             /* load j atom coordinates */
 972             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 973                                               &jx0,&jy0,&jz0);
 974
 975             /* Calculate displacement vector */
 976             dx00             = _mm_sub_pd(ix0,jx0);
 977             dy00             = _mm_sub_pd(iy0,jy0);
 978             dz00             = _mm_sub_pd(iz0,jz0);
 979             dx10             = _mm_sub_pd(ix1,jx0);
 980             dy10             = _mm_sub_pd(iy1,jy0);
 981             dz10             = _mm_sub_pd(iz1,jz0);
 982             dx20             = _mm_sub_pd(ix2,jx0);
 983             dy20             = _mm_sub_pd(iy2,jy0);
 984             dz20             = _mm_sub_pd(iz2,jz0);
 985
 986             /* Calculate squared distance and things based on it */
 987             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 988             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 989             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 990
 991             rinv00           = avx128fma_invsqrt_d(rsq00);
 992             rinv10           = avx128fma_invsqrt_d(rsq10);
 993             rinv20           = avx128fma_invsqrt_d(rsq20);
 994
 995             /* Load parameters for j particles */
 996             jq0              = _mm_load_sd(charge+jnrA+0);
 997             vdwjidx0A        = 2*vdwtype[jnrA+0];
 998
 999             fjx0             = _mm_setzero_pd();
1000             fjy0             = _mm_setzero_pd();
1001             fjz0             = _mm_setzero_pd();
1002
1003             /**************************
1004              * CALCULATE INTERACTIONS *
1005              **************************/
1006
1007             r00              = _mm_mul_pd(rsq00,rinv00);
1008
1009             /* Compute parameters for interactions between i and j atoms */
1010             qq00             = _mm_mul_pd(iq0,jq0);
1011             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1012
1013             /* Calculate table index by multiplying r with table scale and truncate to integer */
1014             rt               = _mm_mul_pd(r00,vftabscale);
1015             vfitab           = _mm_cvttpd_epi32(rt);
1016 #ifdef __XOP__
1017             vfeps            = _mm_frcz_pd(rt);
1018 #else
1019             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1020 #endif
1021             twovfeps         = _mm_add_pd(vfeps,vfeps);
1022             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1023
1024             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1025             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1026             F                = _mm_setzero_pd();
1027             GMX_MM_TRANSPOSE2_PD(Y,F);
1028             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1029             H                = _mm_setzero_pd();
1030             GMX_MM_TRANSPOSE2_PD(G,H);
1031             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1032             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1033             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
1034
1035             /* CUBIC SPLINE TABLE DISPERSION */
1036             vfitab           = _mm_add_epi32(vfitab,ifour);
1037             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1038             F                = _mm_setzero_pd();
1039             GMX_MM_TRANSPOSE2_PD(Y,F);
1040             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1041             H                = _mm_setzero_pd();
1042             GMX_MM_TRANSPOSE2_PD(G,H);
1043             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1044             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1045             fvdw6            = _mm_mul_pd(c6_00,FF);
1046
1047             /* CUBIC SPLINE TABLE REPULSION */
1048             vfitab           = _mm_add_epi32(vfitab,ifour);
1049             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1050             F                = _mm_setzero_pd();
1051             GMX_MM_TRANSPOSE2_PD(Y,F);
1052             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1053             H                = _mm_setzero_pd();
1054             GMX_MM_TRANSPOSE2_PD(G,H);
1055             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1056             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1057             fvdw12           = _mm_mul_pd(c12_00,FF);
1058             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1059
1060             fscal            = _mm_add_pd(felec,fvdw);
1061
1062             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1063
1064             /* Update vectorial force */
1065             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1066             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1067             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1068
1069             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1070             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1071             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1072
1073             /**************************
1074              * CALCULATE INTERACTIONS *
1075              **************************/
1076
1077             r10              = _mm_mul_pd(rsq10,rinv10);
1078
1079             /* Compute parameters for interactions between i and j atoms */
1080             qq10             = _mm_mul_pd(iq1,jq0);
1081
1082             /* Calculate table index by multiplying r with table scale and truncate to integer */
1083             rt               = _mm_mul_pd(r10,vftabscale);
1084             vfitab           = _mm_cvttpd_epi32(rt);
1085 #ifdef __XOP__
1086             vfeps            = _mm_frcz_pd(rt);
1087 #else
1088             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1089 #endif
1090             twovfeps         = _mm_add_pd(vfeps,vfeps);
1091             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1092
1093             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1094             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1095             F                = _mm_setzero_pd();
1096             GMX_MM_TRANSPOSE2_PD(Y,F);
1097             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1098             H                = _mm_setzero_pd();
1099             GMX_MM_TRANSPOSE2_PD(G,H);
1100             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1101             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1102             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1103
1104             fscal            = felec;
1105
1106             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1107
1108             /* Update vectorial force */
1109             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1110             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1111             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1112
1113             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1114             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1115             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1116
1117             /**************************
1118              * CALCULATE INTERACTIONS *
1119              **************************/
1120
1121             r20              = _mm_mul_pd(rsq20,rinv20);
1122
1123             /* Compute parameters for interactions between i and j atoms */
1124             qq20             = _mm_mul_pd(iq2,jq0);
1125
1126             /* Calculate table index by multiplying r with table scale and truncate to integer */
1127             rt               = _mm_mul_pd(r20,vftabscale);
1128             vfitab           = _mm_cvttpd_epi32(rt);
1129 #ifdef __XOP__
1130             vfeps            = _mm_frcz_pd(rt);
1131 #else
1132             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1133 #endif
1134             twovfeps         = _mm_add_pd(vfeps,vfeps);
1135             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1136
1137             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1138             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1139             F                = _mm_setzero_pd();
1140             GMX_MM_TRANSPOSE2_PD(Y,F);
1141             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1142             H                = _mm_setzero_pd();
1143             GMX_MM_TRANSPOSE2_PD(G,H);
1144             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1145             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1146             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1147
1148             fscal            = felec;
1149
1150             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1151
1152             /* Update vectorial force */
1153             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1154             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1155             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1156
1157             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1158             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1159             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1160
1161             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1162
1163             /* Inner loop uses 151 flops */
1164         }
1165
1166         /* End of innermost loop */
1167
1168         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1169                                               f+i_coord_offset,fshift+i_shift_offset);
1170
1171         /* Increment number of inner iterations */
1172         inneriter                  += j_index_end - j_index_start;
1173
1174         /* Outer loop uses 18 flops */
1175     }
1176
1177     /* Increment number of outer iterations */
1178     outeriter        += nri;
1179
1180     /* Update outer/inner flops */
1181
1182     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*151);
1183 }