src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_double.h"
  48 #include "kernelutil_x86_avx_256_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_avx_256_double
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_avx_256_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  84     real *           vdwioffsetptr0;
  85     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  86     real *           vdwioffsetptr1;
  87     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  88     real *           vdwioffsetptr2;
  89     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  91     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  96     real             *charge;
  97     int              nvdwtype;
  98     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  99     int              *vdwtype;
 100     real             *vdwparam;
 101     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 102     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 103     __m128i          vfitab;
 104     __m128i          ifour       = _mm_set1_epi32(4);
 105     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 106     real             *vftab;
 107     __m256d          dummy_mask,cutoff_mask;
 108     __m128           tmpmask0,tmpmask1;
 109     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 110     __m256d          one     = _mm256_set1_pd(1.0);
 111     __m256d          two     = _mm256_set1_pd(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm256_set1_pd(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     vftab            = kernel_data->table_vdw->data;
 130     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 131
 132     /* Setup water-specific parameters */
 133     inr              = nlist->iinr[0];
 134     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 135     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 136     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 137     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 138
 139     /* Avoid stupid compiler warnings */
 140     jnrA = jnrB = jnrC = jnrD = 0;
 141     j_coord_offsetA = 0;
 142     j_coord_offsetB = 0;
 143     j_coord_offsetC = 0;
 144     j_coord_offsetD = 0;
 145
 146     outeriter        = 0;
 147     inneriter        = 0;
 148
 149     for(iidx=0;iidx<4*DIM;iidx++)
 150     {
 151         scratch[iidx] = 0.0;
 152     }
 153
 154     /* Start outer loop over neighborlists */
 155     for(iidx=0; iidx<nri; iidx++)
 156     {
 157         /* Load shift vector for this list */
 158         i_shift_offset   = DIM*shiftidx[iidx];
 159
 160         /* Load limits for loop over neighbors */
 161         j_index_start    = jindex[iidx];
 162         j_index_end      = jindex[iidx+1];
 163
 164         /* Get outer coordinate index */
 165         inr              = iinr[iidx];
 166         i_coord_offset   = DIM*inr;
 167
 168         /* Load i particle coords and add shift vector */
 169         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 170                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 171
 172         fix0             = _mm256_setzero_pd();
 173         fiy0             = _mm256_setzero_pd();
 174         fiz0             = _mm256_setzero_pd();
 175         fix1             = _mm256_setzero_pd();
 176         fiy1             = _mm256_setzero_pd();
 177         fiz1             = _mm256_setzero_pd();
 178         fix2             = _mm256_setzero_pd();
 179         fiy2             = _mm256_setzero_pd();
 180         fiz2             = _mm256_setzero_pd();
 181
 182         /* Reset potential sums */
 183         velecsum         = _mm256_setzero_pd();
 184         vvdwsum          = _mm256_setzero_pd();
 185
 186         /* Start inner kernel loop */
 187         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 188         {
 189
 190             /* Get j neighbor index, and coordinate index */
 191             jnrA             = jjnr[jidx];
 192             jnrB             = jjnr[jidx+1];
 193             jnrC             = jjnr[jidx+2];
 194             jnrD             = jjnr[jidx+3];
 195             j_coord_offsetA  = DIM*jnrA;
 196             j_coord_offsetB  = DIM*jnrB;
 197             j_coord_offsetC  = DIM*jnrC;
 198             j_coord_offsetD  = DIM*jnrD;
 199
 200             /* load j atom coordinates */
 201             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 202                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 203                                                  &jx0,&jy0,&jz0);
 204
 205             /* Calculate displacement vector */
 206             dx00             = _mm256_sub_pd(ix0,jx0);
 207             dy00             = _mm256_sub_pd(iy0,jy0);
 208             dz00             = _mm256_sub_pd(iz0,jz0);
 209             dx10             = _mm256_sub_pd(ix1,jx0);
 210             dy10             = _mm256_sub_pd(iy1,jy0);
 211             dz10             = _mm256_sub_pd(iz1,jz0);
 212             dx20             = _mm256_sub_pd(ix2,jx0);
 213             dy20             = _mm256_sub_pd(iy2,jy0);
 214             dz20             = _mm256_sub_pd(iz2,jz0);
 215
 216             /* Calculate squared distance and things based on it */
 217             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 218             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 219             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 220
 221             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 222             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 223             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 224
 225             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 226             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 227             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 228
 229             /* Load parameters for j particles */
 230             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 231                                                                  charge+jnrC+0,charge+jnrD+0);
 232             vdwjidx0A        = 2*vdwtype[jnrA+0];
 233             vdwjidx0B        = 2*vdwtype[jnrB+0];
 234             vdwjidx0C        = 2*vdwtype[jnrC+0];
 235             vdwjidx0D        = 2*vdwtype[jnrD+0];
 236
 237             fjx0             = _mm256_setzero_pd();
 238             fjy0             = _mm256_setzero_pd();
 239             fjz0             = _mm256_setzero_pd();
 240
 241             /**************************
 242              * CALCULATE INTERACTIONS *
 243              **************************/
 244
 245             r00              = _mm256_mul_pd(rsq00,rinv00);
 246
 247             /* Compute parameters for interactions between i and j atoms */
 248             qq00             = _mm256_mul_pd(iq0,jq0);
 249             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 250                                             vdwioffsetptr0+vdwjidx0B,
 251                                             vdwioffsetptr0+vdwjidx0C,
 252                                             vdwioffsetptr0+vdwjidx0D,
 253                                             &c6_00,&c12_00);
 254
 255             /* Calculate table index by multiplying r with table scale and truncate to integer */
 256             rt               = _mm256_mul_pd(r00,vftabscale);
 257             vfitab           = _mm256_cvttpd_epi32(rt);
 258             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 259             vfitab           = _mm_slli_epi32(vfitab,3);
 260
 261             /* COULOMB ELECTROSTATICS */
 262             velec            = _mm256_mul_pd(qq00,rinv00);
 263             felec            = _mm256_mul_pd(velec,rinvsq00);
 264
 265             /* CUBIC SPLINE TABLE DISPERSION */
 266             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 267             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 268             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 269             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 270             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 271             Heps             = _mm256_mul_pd(vfeps,H);
 272             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 273             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 274             vvdw6            = _mm256_mul_pd(c6_00,VV);
 275             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 276             fvdw6            = _mm256_mul_pd(c6_00,FF);
 277
 278             /* CUBIC SPLINE TABLE REPULSION */
 279             vfitab           = _mm_add_epi32(vfitab,ifour);
 280             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 281             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 282             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 283             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 284             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 285             Heps             = _mm256_mul_pd(vfeps,H);
 286             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 287             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 288             vvdw12           = _mm256_mul_pd(c12_00,VV);
 289             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 290             fvdw12           = _mm256_mul_pd(c12_00,FF);
 291             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 292             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 293
 294             /* Update potential sum for this i atom from the interaction with this j atom. */
 295             velecsum         = _mm256_add_pd(velecsum,velec);
 296             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 297
 298             fscal            = _mm256_add_pd(felec,fvdw);
 299
 300             /* Calculate temporary vectorial force */
 301             tx               = _mm256_mul_pd(fscal,dx00);
 302             ty               = _mm256_mul_pd(fscal,dy00);
 303             tz               = _mm256_mul_pd(fscal,dz00);
 304
 305             /* Update vectorial force */
 306             fix0             = _mm256_add_pd(fix0,tx);
 307             fiy0             = _mm256_add_pd(fiy0,ty);
 308             fiz0             = _mm256_add_pd(fiz0,tz);
 309
 310             fjx0             = _mm256_add_pd(fjx0,tx);
 311             fjy0             = _mm256_add_pd(fjy0,ty);
 312             fjz0             = _mm256_add_pd(fjz0,tz);
 313
 314             /**************************
 315              * CALCULATE INTERACTIONS *
 316              **************************/
 317
 318             /* Compute parameters for interactions between i and j atoms */
 319             qq10             = _mm256_mul_pd(iq1,jq0);
 320
 321             /* COULOMB ELECTROSTATICS */
 322             velec            = _mm256_mul_pd(qq10,rinv10);
 323             felec            = _mm256_mul_pd(velec,rinvsq10);
 324
 325             /* Update potential sum for this i atom from the interaction with this j atom. */
 326             velecsum         = _mm256_add_pd(velecsum,velec);
 327
 328             fscal            = felec;
 329
 330             /* Calculate temporary vectorial force */
 331             tx               = _mm256_mul_pd(fscal,dx10);
 332             ty               = _mm256_mul_pd(fscal,dy10);
 333             tz               = _mm256_mul_pd(fscal,dz10);
 334
 335             /* Update vectorial force */
 336             fix1             = _mm256_add_pd(fix1,tx);
 337             fiy1             = _mm256_add_pd(fiy1,ty);
 338             fiz1             = _mm256_add_pd(fiz1,tz);
 339
 340             fjx0             = _mm256_add_pd(fjx0,tx);
 341             fjy0             = _mm256_add_pd(fjy0,ty);
 342             fjz0             = _mm256_add_pd(fjz0,tz);
 343
 344             /**************************
 345              * CALCULATE INTERACTIONS *
 346              **************************/
 347
 348             /* Compute parameters for interactions between i and j atoms */
 349             qq20             = _mm256_mul_pd(iq2,jq0);
 350
 351             /* COULOMB ELECTROSTATICS */
 352             velec            = _mm256_mul_pd(qq20,rinv20);
 353             felec            = _mm256_mul_pd(velec,rinvsq20);
 354
 355             /* Update potential sum for this i atom from the interaction with this j atom. */
 356             velecsum         = _mm256_add_pd(velecsum,velec);
 357
 358             fscal            = felec;
 359
 360             /* Calculate temporary vectorial force */
 361             tx               = _mm256_mul_pd(fscal,dx20);
 362             ty               = _mm256_mul_pd(fscal,dy20);
 363             tz               = _mm256_mul_pd(fscal,dz20);
 364
 365             /* Update vectorial force */
 366             fix2             = _mm256_add_pd(fix2,tx);
 367             fiy2             = _mm256_add_pd(fiy2,ty);
 368             fiz2             = _mm256_add_pd(fiz2,tz);
 369
 370             fjx0             = _mm256_add_pd(fjx0,tx);
 371             fjy0             = _mm256_add_pd(fjy0,ty);
 372             fjz0             = _mm256_add_pd(fjz0,tz);
 373
 374             fjptrA             = f+j_coord_offsetA;
 375             fjptrB             = f+j_coord_offsetB;
 376             fjptrC             = f+j_coord_offsetC;
 377             fjptrD             = f+j_coord_offsetD;
 378
 379             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 380
 381             /* Inner loop uses 119 flops */
 382         }
 383
 384         if(jidx<j_index_end)
 385         {
 386
 387             /* Get j neighbor index, and coordinate index */
 388             jnrlistA         = jjnr[jidx];
 389             jnrlistB         = jjnr[jidx+1];
 390             jnrlistC         = jjnr[jidx+2];
 391             jnrlistD         = jjnr[jidx+3];
 392             /* Sign of each element will be negative for non-real atoms.
 393              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 394              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 395              */
 396             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 397
 398             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 399             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 400             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 401
 402             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 403             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 404             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 405             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 406             j_coord_offsetA  = DIM*jnrA;
 407             j_coord_offsetB  = DIM*jnrB;
 408             j_coord_offsetC  = DIM*jnrC;
 409             j_coord_offsetD  = DIM*jnrD;
 410
 411             /* load j atom coordinates */
 412             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 413                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 414                                                  &jx0,&jy0,&jz0);
 415
 416             /* Calculate displacement vector */
 417             dx00             = _mm256_sub_pd(ix0,jx0);
 418             dy00             = _mm256_sub_pd(iy0,jy0);
 419             dz00             = _mm256_sub_pd(iz0,jz0);
 420             dx10             = _mm256_sub_pd(ix1,jx0);
 421             dy10             = _mm256_sub_pd(iy1,jy0);
 422             dz10             = _mm256_sub_pd(iz1,jz0);
 423             dx20             = _mm256_sub_pd(ix2,jx0);
 424             dy20             = _mm256_sub_pd(iy2,jy0);
 425             dz20             = _mm256_sub_pd(iz2,jz0);
 426
 427             /* Calculate squared distance and things based on it */
 428             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 429             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 430             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 431
 432             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 433             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 434             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 435
 436             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 437             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 438             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 439
 440             /* Load parameters for j particles */
 441             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 442                                                                  charge+jnrC+0,charge+jnrD+0);
 443             vdwjidx0A        = 2*vdwtype[jnrA+0];
 444             vdwjidx0B        = 2*vdwtype[jnrB+0];
 445             vdwjidx0C        = 2*vdwtype[jnrC+0];
 446             vdwjidx0D        = 2*vdwtype[jnrD+0];
 447
 448             fjx0             = _mm256_setzero_pd();
 449             fjy0             = _mm256_setzero_pd();
 450             fjz0             = _mm256_setzero_pd();
 451
 452             /**************************
 453              * CALCULATE INTERACTIONS *
 454              **************************/
 455
 456             r00              = _mm256_mul_pd(rsq00,rinv00);
 457             r00              = _mm256_andnot_pd(dummy_mask,r00);
 458
 459             /* Compute parameters for interactions between i and j atoms */
 460             qq00             = _mm256_mul_pd(iq0,jq0);
 461             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 462                                             vdwioffsetptr0+vdwjidx0B,
 463                                             vdwioffsetptr0+vdwjidx0C,
 464                                             vdwioffsetptr0+vdwjidx0D,
 465                                             &c6_00,&c12_00);
 466
 467             /* Calculate table index by multiplying r with table scale and truncate to integer */
 468             rt               = _mm256_mul_pd(r00,vftabscale);
 469             vfitab           = _mm256_cvttpd_epi32(rt);
 470             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 471             vfitab           = _mm_slli_epi32(vfitab,3);
 472
 473             /* COULOMB ELECTROSTATICS */
 474             velec            = _mm256_mul_pd(qq00,rinv00);
 475             felec            = _mm256_mul_pd(velec,rinvsq00);
 476
 477             /* CUBIC SPLINE TABLE DISPERSION */
 478             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 479             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 480             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 481             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 482             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 483             Heps             = _mm256_mul_pd(vfeps,H);
 484             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 485             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 486             vvdw6            = _mm256_mul_pd(c6_00,VV);
 487             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 488             fvdw6            = _mm256_mul_pd(c6_00,FF);
 489
 490             /* CUBIC SPLINE TABLE REPULSION */
 491             vfitab           = _mm_add_epi32(vfitab,ifour);
 492             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 493             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 494             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 495             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 496             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 497             Heps             = _mm256_mul_pd(vfeps,H);
 498             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 499             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 500             vvdw12           = _mm256_mul_pd(c12_00,VV);
 501             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 502             fvdw12           = _mm256_mul_pd(c12_00,FF);
 503             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 504             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 505
 506             /* Update potential sum for this i atom from the interaction with this j atom. */
 507             velec            = _mm256_andnot_pd(dummy_mask,velec);
 508             velecsum         = _mm256_add_pd(velecsum,velec);
 509             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 510             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 511
 512             fscal            = _mm256_add_pd(felec,fvdw);
 513
 514             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 515
 516             /* Calculate temporary vectorial force */
 517             tx               = _mm256_mul_pd(fscal,dx00);
 518             ty               = _mm256_mul_pd(fscal,dy00);
 519             tz               = _mm256_mul_pd(fscal,dz00);
 520
 521             /* Update vectorial force */
 522             fix0             = _mm256_add_pd(fix0,tx);
 523             fiy0             = _mm256_add_pd(fiy0,ty);
 524             fiz0             = _mm256_add_pd(fiz0,tz);
 525
 526             fjx0             = _mm256_add_pd(fjx0,tx);
 527             fjy0             = _mm256_add_pd(fjy0,ty);
 528             fjz0             = _mm256_add_pd(fjz0,tz);
 529
 530             /**************************
 531              * CALCULATE INTERACTIONS *
 532              **************************/
 533
 534             /* Compute parameters for interactions between i and j atoms */
 535             qq10             = _mm256_mul_pd(iq1,jq0);
 536
 537             /* COULOMB ELECTROSTATICS */
 538             velec            = _mm256_mul_pd(qq10,rinv10);
 539             felec            = _mm256_mul_pd(velec,rinvsq10);
 540
 541             /* Update potential sum for this i atom from the interaction with this j atom. */
 542             velec            = _mm256_andnot_pd(dummy_mask,velec);
 543             velecsum         = _mm256_add_pd(velecsum,velec);
 544
 545             fscal            = felec;
 546
 547             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 548
 549             /* Calculate temporary vectorial force */
 550             tx               = _mm256_mul_pd(fscal,dx10);
 551             ty               = _mm256_mul_pd(fscal,dy10);
 552             tz               = _mm256_mul_pd(fscal,dz10);
 553
 554             /* Update vectorial force */
 555             fix1             = _mm256_add_pd(fix1,tx);
 556             fiy1             = _mm256_add_pd(fiy1,ty);
 557             fiz1             = _mm256_add_pd(fiz1,tz);
 558
 559             fjx0             = _mm256_add_pd(fjx0,tx);
 560             fjy0             = _mm256_add_pd(fjy0,ty);
 561             fjz0             = _mm256_add_pd(fjz0,tz);
 562
 563             /**************************
 564              * CALCULATE INTERACTIONS *
 565              **************************/
 566
 567             /* Compute parameters for interactions between i and j atoms */
 568             qq20             = _mm256_mul_pd(iq2,jq0);
 569
 570             /* COULOMB ELECTROSTATICS */
 571             velec            = _mm256_mul_pd(qq20,rinv20);
 572             felec            = _mm256_mul_pd(velec,rinvsq20);
 573
 574             /* Update potential sum for this i atom from the interaction with this j atom. */
 575             velec            = _mm256_andnot_pd(dummy_mask,velec);
 576             velecsum         = _mm256_add_pd(velecsum,velec);
 577
 578             fscal            = felec;
 579
 580             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 581
 582             /* Calculate temporary vectorial force */
 583             tx               = _mm256_mul_pd(fscal,dx20);
 584             ty               = _mm256_mul_pd(fscal,dy20);
 585             tz               = _mm256_mul_pd(fscal,dz20);
 586
 587             /* Update vectorial force */
 588             fix2             = _mm256_add_pd(fix2,tx);
 589             fiy2             = _mm256_add_pd(fiy2,ty);
 590             fiz2             = _mm256_add_pd(fiz2,tz);
 591
 592             fjx0             = _mm256_add_pd(fjx0,tx);
 593             fjy0             = _mm256_add_pd(fjy0,ty);
 594             fjz0             = _mm256_add_pd(fjz0,tz);
 595
 596             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 597             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 598             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 599             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 600
 601             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 602
 603             /* Inner loop uses 120 flops */
 604         }
 605
 606         /* End of innermost loop */
 607
 608         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 609                                                  f+i_coord_offset,fshift+i_shift_offset);
 610
 611         ggid                        = gid[iidx];
 612         /* Update potential energies */
 613         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 614         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 615
 616         /* Increment number of inner iterations */
 617         inneriter                  += j_index_end - j_index_start;
 618
 619         /* Outer loop uses 20 flops */
 620     }
 621
 622     /* Increment number of outer iterations */
 623     outeriter        += nri;
 624
 625     /* Update outer/inner flops */
 626
 627     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*120);
 628 }
 629 /*
 630  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_avx_256_double
 631  * Electrostatics interaction: Coulomb
 632  * VdW interaction:            CubicSplineTable
 633  * Geometry:                   Water3-Particle
 634  * Calculate force/pot:        Force
 635  */
 636 void
 637 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_avx_256_double
 638                     (t_nblist                    * gmx_restrict       nlist,
 639                      rvec                        * gmx_restrict          xx,
 640                      rvec                        * gmx_restrict          ff,
 641                      t_forcerec                  * gmx_restrict          fr,
 642                      t_mdatoms                   * gmx_restrict     mdatoms,
 643                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 644                      t_nrnb                      * gmx_restrict        nrnb)
 645 {
 646     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 647      * just 0 for non-waters.
 648      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 649      * jnr indices corresponding to data put in the four positions in the SIMD register.
 650      */
 651     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 652     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 653     int              jnrA,jnrB,jnrC,jnrD;
 654     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 655     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 656     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 657     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 658     real             rcutoff_scalar;
 659     real             *shiftvec,*fshift,*x,*f;
 660     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 661     real             scratch[4*DIM];
 662     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 663     real *           vdwioffsetptr0;
 664     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 665     real *           vdwioffsetptr1;
 666     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 667     real *           vdwioffsetptr2;
 668     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 669     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 670     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 671     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 672     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 673     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 674     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 675     real             *charge;
 676     int              nvdwtype;
 677     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 678     int              *vdwtype;
 679     real             *vdwparam;
 680     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 681     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 682     __m128i          vfitab;
 683     __m128i          ifour       = _mm_set1_epi32(4);
 684     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 685     real             *vftab;
 686     __m256d          dummy_mask,cutoff_mask;
 687     __m128           tmpmask0,tmpmask1;
 688     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 689     __m256d          one     = _mm256_set1_pd(1.0);
 690     __m256d          two     = _mm256_set1_pd(2.0);
 691     x                = xx[0];
 692     f                = ff[0];
 693
 694     nri              = nlist->nri;
 695     iinr             = nlist->iinr;
 696     jindex           = nlist->jindex;
 697     jjnr             = nlist->jjnr;
 698     shiftidx         = nlist->shift;
 699     gid              = nlist->gid;
 700     shiftvec         = fr->shift_vec[0];
 701     fshift           = fr->fshift[0];
 702     facel            = _mm256_set1_pd(fr->epsfac);
 703     charge           = mdatoms->chargeA;
 704     nvdwtype         = fr->ntype;
 705     vdwparam         = fr->nbfp;
 706     vdwtype          = mdatoms->typeA;
 707
 708     vftab            = kernel_data->table_vdw->data;
 709     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 710
 711     /* Setup water-specific parameters */
 712     inr              = nlist->iinr[0];
 713     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 714     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 715     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 716     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 717
 718     /* Avoid stupid compiler warnings */
 719     jnrA = jnrB = jnrC = jnrD = 0;
 720     j_coord_offsetA = 0;
 721     j_coord_offsetB = 0;
 722     j_coord_offsetC = 0;
 723     j_coord_offsetD = 0;
 724
 725     outeriter        = 0;
 726     inneriter        = 0;
 727
 728     for(iidx=0;iidx<4*DIM;iidx++)
 729     {
 730         scratch[iidx] = 0.0;
 731     }
 732
 733     /* Start outer loop over neighborlists */
 734     for(iidx=0; iidx<nri; iidx++)
 735     {
 736         /* Load shift vector for this list */
 737         i_shift_offset   = DIM*shiftidx[iidx];
 738
 739         /* Load limits for loop over neighbors */
 740         j_index_start    = jindex[iidx];
 741         j_index_end      = jindex[iidx+1];
 742
 743         /* Get outer coordinate index */
 744         inr              = iinr[iidx];
 745         i_coord_offset   = DIM*inr;
 746
 747         /* Load i particle coords and add shift vector */
 748         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 749                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 750
 751         fix0             = _mm256_setzero_pd();
 752         fiy0             = _mm256_setzero_pd();
 753         fiz0             = _mm256_setzero_pd();
 754         fix1             = _mm256_setzero_pd();
 755         fiy1             = _mm256_setzero_pd();
 756         fiz1             = _mm256_setzero_pd();
 757         fix2             = _mm256_setzero_pd();
 758         fiy2             = _mm256_setzero_pd();
 759         fiz2             = _mm256_setzero_pd();
 760
 761         /* Start inner kernel loop */
 762         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 763         {
 764
 765             /* Get j neighbor index, and coordinate index */
 766             jnrA             = jjnr[jidx];
 767             jnrB             = jjnr[jidx+1];
 768             jnrC             = jjnr[jidx+2];
 769             jnrD             = jjnr[jidx+3];
 770             j_coord_offsetA  = DIM*jnrA;
 771             j_coord_offsetB  = DIM*jnrB;
 772             j_coord_offsetC  = DIM*jnrC;
 773             j_coord_offsetD  = DIM*jnrD;
 774
 775             /* load j atom coordinates */
 776             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 777                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 778                                                  &jx0,&jy0,&jz0);
 779
 780             /* Calculate displacement vector */
 781             dx00             = _mm256_sub_pd(ix0,jx0);
 782             dy00             = _mm256_sub_pd(iy0,jy0);
 783             dz00             = _mm256_sub_pd(iz0,jz0);
 784             dx10             = _mm256_sub_pd(ix1,jx0);
 785             dy10             = _mm256_sub_pd(iy1,jy0);
 786             dz10             = _mm256_sub_pd(iz1,jz0);
 787             dx20             = _mm256_sub_pd(ix2,jx0);
 788             dy20             = _mm256_sub_pd(iy2,jy0);
 789             dz20             = _mm256_sub_pd(iz2,jz0);
 790
 791             /* Calculate squared distance and things based on it */
 792             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 793             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 794             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 795
 796             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 797             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 798             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 799
 800             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 801             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 802             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 803
 804             /* Load parameters for j particles */
 805             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 806                                                                  charge+jnrC+0,charge+jnrD+0);
 807             vdwjidx0A        = 2*vdwtype[jnrA+0];
 808             vdwjidx0B        = 2*vdwtype[jnrB+0];
 809             vdwjidx0C        = 2*vdwtype[jnrC+0];
 810             vdwjidx0D        = 2*vdwtype[jnrD+0];
 811
 812             fjx0             = _mm256_setzero_pd();
 813             fjy0             = _mm256_setzero_pd();
 814             fjz0             = _mm256_setzero_pd();
 815
 816             /**************************
 817              * CALCULATE INTERACTIONS *
 818              **************************/
 819
 820             r00              = _mm256_mul_pd(rsq00,rinv00);
 821
 822             /* Compute parameters for interactions between i and j atoms */
 823             qq00             = _mm256_mul_pd(iq0,jq0);
 824             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 825                                             vdwioffsetptr0+vdwjidx0B,
 826                                             vdwioffsetptr0+vdwjidx0C,
 827                                             vdwioffsetptr0+vdwjidx0D,
 828                                             &c6_00,&c12_00);
 829
 830             /* Calculate table index by multiplying r with table scale and truncate to integer */
 831             rt               = _mm256_mul_pd(r00,vftabscale);
 832             vfitab           = _mm256_cvttpd_epi32(rt);
 833             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 834             vfitab           = _mm_slli_epi32(vfitab,3);
 835
 836             /* COULOMB ELECTROSTATICS */
 837             velec            = _mm256_mul_pd(qq00,rinv00);
 838             felec            = _mm256_mul_pd(velec,rinvsq00);
 839
 840             /* CUBIC SPLINE TABLE DISPERSION */
 841             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 842             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 843             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 844             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 845             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 846             Heps             = _mm256_mul_pd(vfeps,H);
 847             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 848             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 849             fvdw6            = _mm256_mul_pd(c6_00,FF);
 850
 851             /* CUBIC SPLINE TABLE REPULSION */
 852             vfitab           = _mm_add_epi32(vfitab,ifour);
 853             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 854             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 855             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 856             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 857             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 858             Heps             = _mm256_mul_pd(vfeps,H);
 859             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 860             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 861             fvdw12           = _mm256_mul_pd(c12_00,FF);
 862             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 863
 864             fscal            = _mm256_add_pd(felec,fvdw);
 865
 866             /* Calculate temporary vectorial force */
 867             tx               = _mm256_mul_pd(fscal,dx00);
 868             ty               = _mm256_mul_pd(fscal,dy00);
 869             tz               = _mm256_mul_pd(fscal,dz00);
 870
 871             /* Update vectorial force */
 872             fix0             = _mm256_add_pd(fix0,tx);
 873             fiy0             = _mm256_add_pd(fiy0,ty);
 874             fiz0             = _mm256_add_pd(fiz0,tz);
 875
 876             fjx0             = _mm256_add_pd(fjx0,tx);
 877             fjy0             = _mm256_add_pd(fjy0,ty);
 878             fjz0             = _mm256_add_pd(fjz0,tz);
 879
 880             /**************************
 881              * CALCULATE INTERACTIONS *
 882              **************************/
 883
 884             /* Compute parameters for interactions between i and j atoms */
 885             qq10             = _mm256_mul_pd(iq1,jq0);
 886
 887             /* COULOMB ELECTROSTATICS */
 888             velec            = _mm256_mul_pd(qq10,rinv10);
 889             felec            = _mm256_mul_pd(velec,rinvsq10);
 890
 891             fscal            = felec;
 892
 893             /* Calculate temporary vectorial force */
 894             tx               = _mm256_mul_pd(fscal,dx10);
 895             ty               = _mm256_mul_pd(fscal,dy10);
 896             tz               = _mm256_mul_pd(fscal,dz10);
 897
 898             /* Update vectorial force */
 899             fix1             = _mm256_add_pd(fix1,tx);
 900             fiy1             = _mm256_add_pd(fiy1,ty);
 901             fiz1             = _mm256_add_pd(fiz1,tz);
 902
 903             fjx0             = _mm256_add_pd(fjx0,tx);
 904             fjy0             = _mm256_add_pd(fjy0,ty);
 905             fjz0             = _mm256_add_pd(fjz0,tz);
 906
 907             /**************************
 908              * CALCULATE INTERACTIONS *
 909              **************************/
 910
 911             /* Compute parameters for interactions between i and j atoms */
 912             qq20             = _mm256_mul_pd(iq2,jq0);
 913
 914             /* COULOMB ELECTROSTATICS */
 915             velec            = _mm256_mul_pd(qq20,rinv20);
 916             felec            = _mm256_mul_pd(velec,rinvsq20);
 917
 918             fscal            = felec;
 919
 920             /* Calculate temporary vectorial force */
 921             tx               = _mm256_mul_pd(fscal,dx20);
 922             ty               = _mm256_mul_pd(fscal,dy20);
 923             tz               = _mm256_mul_pd(fscal,dz20);
 924
 925             /* Update vectorial force */
 926             fix2             = _mm256_add_pd(fix2,tx);
 927             fiy2             = _mm256_add_pd(fiy2,ty);
 928             fiz2             = _mm256_add_pd(fiz2,tz);
 929
 930             fjx0             = _mm256_add_pd(fjx0,tx);
 931             fjy0             = _mm256_add_pd(fjy0,ty);
 932             fjz0             = _mm256_add_pd(fjz0,tz);
 933
 934             fjptrA             = f+j_coord_offsetA;
 935             fjptrB             = f+j_coord_offsetB;
 936             fjptrC             = f+j_coord_offsetC;
 937             fjptrD             = f+j_coord_offsetD;
 938
 939             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 940
 941             /* Inner loop uses 108 flops */
 942         }
 943
 944         if(jidx<j_index_end)
 945         {
 946
 947             /* Get j neighbor index, and coordinate index */
 948             jnrlistA         = jjnr[jidx];
 949             jnrlistB         = jjnr[jidx+1];
 950             jnrlistC         = jjnr[jidx+2];
 951             jnrlistD         = jjnr[jidx+3];
 952             /* Sign of each element will be negative for non-real atoms.
 953              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 954              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 955              */
 956             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 957
 958             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 959             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 960             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 961
 962             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 963             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 964             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 965             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 966             j_coord_offsetA  = DIM*jnrA;
 967             j_coord_offsetB  = DIM*jnrB;
 968             j_coord_offsetC  = DIM*jnrC;
 969             j_coord_offsetD  = DIM*jnrD;
 970
 971             /* load j atom coordinates */
 972             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 973                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 974                                                  &jx0,&jy0,&jz0);
 975
 976             /* Calculate displacement vector */
 977             dx00             = _mm256_sub_pd(ix0,jx0);
 978             dy00             = _mm256_sub_pd(iy0,jy0);
 979             dz00             = _mm256_sub_pd(iz0,jz0);
 980             dx10             = _mm256_sub_pd(ix1,jx0);
 981             dy10             = _mm256_sub_pd(iy1,jy0);
 982             dz10             = _mm256_sub_pd(iz1,jz0);
 983             dx20             = _mm256_sub_pd(ix2,jx0);
 984             dy20             = _mm256_sub_pd(iy2,jy0);
 985             dz20             = _mm256_sub_pd(iz2,jz0);
 986
 987             /* Calculate squared distance and things based on it */
 988             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 989             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 990             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 991
 992             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 993             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 994             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 995
 996             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 997             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 998             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 999
1000             /* Load parameters for j particles */
1001             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1002                                                                  charge+jnrC+0,charge+jnrD+0);
1003             vdwjidx0A        = 2*vdwtype[jnrA+0];
1004             vdwjidx0B        = 2*vdwtype[jnrB+0];
1005             vdwjidx0C        = 2*vdwtype[jnrC+0];
1006             vdwjidx0D        = 2*vdwtype[jnrD+0];
1007
1008             fjx0             = _mm256_setzero_pd();
1009             fjy0             = _mm256_setzero_pd();
1010             fjz0             = _mm256_setzero_pd();
1011
1012             /**************************
1013              * CALCULATE INTERACTIONS *
1014              **************************/
1015
1016             r00              = _mm256_mul_pd(rsq00,rinv00);
1017             r00              = _mm256_andnot_pd(dummy_mask,r00);
1018
1019             /* Compute parameters for interactions between i and j atoms */
1020             qq00             = _mm256_mul_pd(iq0,jq0);
1021             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1022                                             vdwioffsetptr0+vdwjidx0B,
1023                                             vdwioffsetptr0+vdwjidx0C,
1024                                             vdwioffsetptr0+vdwjidx0D,
1025                                             &c6_00,&c12_00);
1026
1027             /* Calculate table index by multiplying r with table scale and truncate to integer */
1028             rt               = _mm256_mul_pd(r00,vftabscale);
1029             vfitab           = _mm256_cvttpd_epi32(rt);
1030             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1031             vfitab           = _mm_slli_epi32(vfitab,3);
1032
1033             /* COULOMB ELECTROSTATICS */
1034             velec            = _mm256_mul_pd(qq00,rinv00);
1035             felec            = _mm256_mul_pd(velec,rinvsq00);
1036
1037             /* CUBIC SPLINE TABLE DISPERSION */
1038             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1039             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1040             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1041             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1042             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1043             Heps             = _mm256_mul_pd(vfeps,H);
1044             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1045             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1046             fvdw6            = _mm256_mul_pd(c6_00,FF);
1047
1048             /* CUBIC SPLINE TABLE REPULSION */
1049             vfitab           = _mm_add_epi32(vfitab,ifour);
1050             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1051             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1052             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1053             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1054             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1055             Heps             = _mm256_mul_pd(vfeps,H);
1056             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1057             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1058             fvdw12           = _mm256_mul_pd(c12_00,FF);
1059             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1060
1061             fscal            = _mm256_add_pd(felec,fvdw);
1062
1063             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1064
1065             /* Calculate temporary vectorial force */
1066             tx               = _mm256_mul_pd(fscal,dx00);
1067             ty               = _mm256_mul_pd(fscal,dy00);
1068             tz               = _mm256_mul_pd(fscal,dz00);
1069
1070             /* Update vectorial force */
1071             fix0             = _mm256_add_pd(fix0,tx);
1072             fiy0             = _mm256_add_pd(fiy0,ty);
1073             fiz0             = _mm256_add_pd(fiz0,tz);
1074
1075             fjx0             = _mm256_add_pd(fjx0,tx);
1076             fjy0             = _mm256_add_pd(fjy0,ty);
1077             fjz0             = _mm256_add_pd(fjz0,tz);
1078
1079             /**************************
1080              * CALCULATE INTERACTIONS *
1081              **************************/
1082
1083             /* Compute parameters for interactions between i and j atoms */
1084             qq10             = _mm256_mul_pd(iq1,jq0);
1085
1086             /* COULOMB ELECTROSTATICS */
1087             velec            = _mm256_mul_pd(qq10,rinv10);
1088             felec            = _mm256_mul_pd(velec,rinvsq10);
1089
1090             fscal            = felec;
1091
1092             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1093
1094             /* Calculate temporary vectorial force */
1095             tx               = _mm256_mul_pd(fscal,dx10);
1096             ty               = _mm256_mul_pd(fscal,dy10);
1097             tz               = _mm256_mul_pd(fscal,dz10);
1098
1099             /* Update vectorial force */
1100             fix1             = _mm256_add_pd(fix1,tx);
1101             fiy1             = _mm256_add_pd(fiy1,ty);
1102             fiz1             = _mm256_add_pd(fiz1,tz);
1103
1104             fjx0             = _mm256_add_pd(fjx0,tx);
1105             fjy0             = _mm256_add_pd(fjy0,ty);
1106             fjz0             = _mm256_add_pd(fjz0,tz);
1107
1108             /**************************
1109              * CALCULATE INTERACTIONS *
1110              **************************/
1111
1112             /* Compute parameters for interactions between i and j atoms */
1113             qq20             = _mm256_mul_pd(iq2,jq0);
1114
1115             /* COULOMB ELECTROSTATICS */
1116             velec            = _mm256_mul_pd(qq20,rinv20);
1117             felec            = _mm256_mul_pd(velec,rinvsq20);
1118
1119             fscal            = felec;
1120
1121             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1122
1123             /* Calculate temporary vectorial force */
1124             tx               = _mm256_mul_pd(fscal,dx20);
1125             ty               = _mm256_mul_pd(fscal,dy20);
1126             tz               = _mm256_mul_pd(fscal,dz20);
1127
1128             /* Update vectorial force */
1129             fix2             = _mm256_add_pd(fix2,tx);
1130             fiy2             = _mm256_add_pd(fiy2,ty);
1131             fiz2             = _mm256_add_pd(fiz2,tz);
1132
1133             fjx0             = _mm256_add_pd(fjx0,tx);
1134             fjy0             = _mm256_add_pd(fjy0,ty);
1135             fjz0             = _mm256_add_pd(fjz0,tz);
1136
1137             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1138             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1139             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1140             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1141
1142             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1143
1144             /* Inner loop uses 109 flops */
1145         }
1146
1147         /* End of innermost loop */
1148
1149         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1150                                                  f+i_coord_offset,fshift+i_shift_offset);
1151
1152         /* Increment number of inner iterations */
1153         inneriter                  += j_index_end - j_index_start;
1154
1155         /* Outer loop uses 18 flops */
1156     }
1157
1158     /* Increment number of outer iterations */
1159     outeriter        += nri;
1160
1161     /* Update outer/inner flops */
1162
1163     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*109);
1164 }